輝達 (NVDA) 2021 Q1 法說會逐字稿

Good afternoon. My name is Josh, and I will be your conference operator today. At this time, I would like to welcome everyone to NVIDIA's Financial Results Conference call. (Operator Instructions) Thank you.

下午好。我的名字是喬希，今天我將成為您的會議接線員。在這個時候，我想歡迎大家參加 NVIDIA 的財務業績電話會議。 （操作員說明）謝謝。

Simona Jankowski, you may begin your conference.

Simona Jankowski，你可以開始你的會議了。

Thank you. Good afternoon, everyone, and welcome to NVIDIA's conference call for the first quarter of fiscal 2021. With me on the call today from NVIDIA are Jensen Huang, President and Chief Executive Officer; and Colette Kress, Executive Vice President and Chief Financial Officer. I'd like to remind you that our call is being webcast live on NVIDIA's Investor Relations website. The webcast will be available for replay until the conference call to discuss our financial results for the second quarter of fiscal 2021. The content of today's call is NVIDIA's property. It can't be reproduced or transcribed without our prior written consent.

謝謝你。大家下午好，歡迎參加 NVIDIA 2021 財年第一季度電話會議。今天與我一起參加電話會議的是 NVIDIA 總裁兼首席執行官 Jensen Huang；和 Colette Kress，執行副總裁兼首席財務官。我想提醒您，我們的電話會議正在 NVIDIA 的投資者關係網站上進行網絡直播。網絡直播將在電話會議召開之前進行重播，以討論我們 2021 財年第二季度的財務業績。今天電話會議的內容是 NVIDIA 的財產。未經我們事先書面同意，不得複製或轉錄。

During this call, we may make forward-looking statements based on current expectations. These are subject to a number of significant risks and uncertainties, and our actual results may vary materially. For a discussion of factors that could affect our future financial results and business, please refer to the disclosure in today's earnings release, our most recent Form 10-K and 10-Q and the reports that we may file on Form 8-K with the Securities and Exchange Commission. All our statements are made as of today, May 21, 2020, based on information currently available to us. Except as required by law, we assume no obligation to update any such statements.

在本次電話會議期間，我們可能會根據當前預期做出前瞻性陳述。這些受到許多重大風險和不確定性的影響，我們的實際結果可能會有很大差異。有關可能影響我們未來財務業績和業務的因素的討論，請參閱今天的收益發布中的披露、我們最近的 10-K 和 10-Q 表格以及我們可能在 8-K 表格中與證券交易委員會。我們所有的聲明都是基於我們目前可獲得的信息，截至今天，2020 年 5 月 21 日。除法律要求外，我們不承擔更新任何此類聲明的義務。

During this call, we will discuss non-GAAP financial measures. You can find a reconciliation of these non-GAAP financial measures to GAAP financial measures in our CFO commentary, which is posted on our website.

在本次電話會議中，我們將討論非 GAAP 財務指標。您可以在我們的 CFO 評論中找到這些非 GAAP 財務指標與 GAAP 財務指標的對賬，該評論發佈在我們的網站上。

With that, let me turn the call over to Jensen.

有了這個，讓我把電話轉給 Jensen。

Thanks, Simona. Before Colette describes our quarterly results, I'd like to thank those who are on the front lines of this crisis, first responders, health care workers, service providers, who inspires every day with their bravery and selflessness. I also want to acknowledge the incredible efforts of our colleagues here at NVIDIA. Despite many challenges, they have barely broken stride during one of the busiest periods in our history.

謝謝，西蒙娜。在科萊特描述我們的季度業績之前，我要感謝那些站在這場危機前線的人們、急救人員、衛生保健工作者、服務提供者，他們每天都以勇敢和無私的精神激勵著他們。我還要感謝我們在 NVIDIA 的同事所做的令人難以置信的努力。儘管面臨許多挑戰，但在我們歷史上最繁忙的時期之一，他們幾乎沒有大踏步前進。

Our efforts related to the virus are focused in 3 areas. First, we're taking care of our families and communities. We've pooled in raises by 6 months to put more money in our employees' hands, and NVIDIA and our people have donated thus far more than $10 million to those in need.

我們與病毒相關的工作集中在三個領域。首先，我們正在照顧我們的家庭和社區。我們已經在 6 個月內籌集到更多資金，以便為我們的員工提供更多資金，而 NVIDIA 和我們的員工迄今已向有需要的人捐贈了超過 1000 萬美元。

Second, we're using NVIDIA's unique capabilities to fight the virus. A great deal of science being done on COVID-19 uses NVIDIA technology for acceleration when every second counts. Some of the many examples, including sequencing the virus, analyzing drug candidates, imaging the virus at molecular resolution with cryo-electron microscopy and identifying elevated body temperature with AI cameras.

其次，我們正在利用 NVIDIA 的獨特能力來對抗病毒。對 COVID-19 進行的大量科學研究在分秒必爭的情況下使用 NVIDIA 技術進行加速。許多例子中的一些，包括對病毒進行測序、分析候選藥物、使用冷凍電子顯微鏡以分子分辨率對病毒進行成像以及使用 AI 相機識別體溫升高。

And third, because COVID-19 won't be the last killer virus, we need to be ready for the next outbreak. NVIDIA technology is essential for the scientific community to develop an end-to-end computational defense system, a system that can detect early, accelerate the development of a vaccine, contain the spread of disease and continuously test and monitor.

第三，由於 COVID-19 不會成為最後的殺手病毒，我們需要為下一次爆發做好準備。 NVIDIA 技術對於科學界開發端到端計算防禦系統至關重要，該系統可以及早發現、加速疫苗開發、遏制疾病傳播並持續測試和監控。

We are racing to deploy the NVIDIA Clara computational health care platforms, Clara Parabricks can accelerate genomics analysis from days to minutes. Clara Imaging will continue to partner with leading research institutes to develop state-of-the-art AI models to detect infections, and Clara Guardian will connect AI to cameras and microphones and hospitals to help overloaded staff watch over patients.

我們正在競相部署 NVIDIA Clara 計算醫療保健平台，Clara Parabricks 可以將基因組學分析從幾天縮短到幾分鐘。 Clara Imaging 將繼續與領先的研究機構合作，開發最先進的 AI 模型來檢測感染，Clara Guardian 將把 AI 連接到攝像頭、麥克風和醫院，以幫助超負荷工作人員看護病人。

We completed the acquisition of Mellanox on April 27. Mellanox is now NVIDIA's networking brand and business unit and will be reported as part of our data center market platform, and Israel is now one of NVIDIA's major technology centers.

我們於 4 月 27 日完成了對 Mellanox 的收購。Mellanox 現在是 NVIDIA 的網絡品牌和業務部門，將被報告為我們數據中心市場平台的一部分，而以色列現在是 NVIDIA 的主要技術中心之一。

The new NVIDIA has a much larger footprint in data center computing, end-to-end and full-stack expertise in data center architectures and tremendous scale to accelerate innovation. NVIDIA Mellanox are a perfect combination and position us for the major forces shaping the IT industry today, data center scale computing and AI.

新的 NVIDIA 在數據中心計算、端到端和全棧數據中心架構專業知識方面擁有更大的足跡，並具有加速創新的巨大規模。 NVIDIA Mellanox 是一個完美的組合，讓我們成為塑造當今 IT 行業、數據中心規模計算和人工智能的主要力量。

From micro service cloud applications to machine learning and AI, accelerated computing and high-performance networking are critical to modern data centers. Previously, a CPU compute node was the unit of computing. Going forward, the new unit of computing is an entire data center. The basic computing elements are now storage servers, CPU servers and GPU servers, and are composed and orchestrated by hyperscale applications that are serving millions of users simultaneously. Connecting these computing elements together is the high-performance Mellanox networking. This is the era of data center scale computing. And together, NVIDIA Mellanox can architect end to end. Mellanox is an extraordinary company, and I'm thrilled that we're now one force to invent the future together.

從微服務雲應用到機器學習和人工智能，加速計算和高性能網絡對於現代數據中心至關重要。以前，CPU 計算節點是計算單元。展望未來，新的計算單元是一個完整的數據中心。基本計算元素現在是存儲服務器、CPU 服務器和 GPU 服務器，由同時為數百萬用戶提供服務的超大規模應用程序組成和編排。將這些計算元素連接在一起的是高性能的 Mellanox 網絡。這是數據中心規模計算的時代。 NVIDIA Mellanox 可以一起進行端到端的架構設計。 Mellanox 是一家非凡的公司，我很高興我們現在成為共同創造未來的力量。

Now let me turn the call over to Colette.

現在讓我把電話轉給 Colette。

Thanks, Jensen. Against the backdrop of the extraordinary events unfolding around the globe, we had a very strong quarter. Q1 revenue was $3.08 billion, up 39% year-on-year, down 1% sequentially and slightly ahead of our outlook, reflecting upside in our data center and gaming platforms.

謝謝，詹森。在全球範圍內發生的非凡事件的背景下，我們有一個非常強勁的季度。第一季度收入為 30.8 億美元，同比增長 39%，環比下降 1%，略高於我們的預期，反映了我們的數據中心和遊戲平台的上行空間。

Starting with gaming. Revenue of $1.34 billion was up 27% year-on-year and down 10% sequentially. We are pleased with these results, which exceeded expectations in the quarter, marked by the unprecedented challenge of the COVID-19.

從遊戲開始。收入為 13.4 億美元，同比增長 27%，環比下降 10%。我們對這些結果感到滿意，這些結果超出了本季度的預期，以 COVID-19 的前所未有的挑戰為標誌。

Let me give you some color. Early in Q1, as the epidemic unfolded, demand in China was impacted with iCafes closing for an extended period. As the virus spread globally, much of the world started working and learning from home, and gameplay surged. Globally, we have seen 50% rise in gaming hours played on our GeForce platform, driven both by more people playing and more gameplay per user.

讓我給你一些顏色。第一季度初，隨著疫情的蔓延，中國的需求受到影響，iCafes 長時間關閉。隨著病毒在全球蔓延，世界上大部分人開始在家工作和學習，遊戲玩法激增。在全球範圍內，我們看到在我們的 GeForce 平台上玩的遊戲時長增加了 50%，這得益於更多的遊戲人數和更多的每位用戶的遊戲玩法。

With many retail outlets closed, demand for our products has shifted quite efficiently to e-tail channels globally. Gaming laptops revenue accelerated to its fastest year-on-year growth in 6 quarters. We are working with our OEMs, channel partners to meet the growing needs of the professionals and students engaged in working, learning and playing at home. In early April, our global OEM partners announced a record new 100 NVIDIA GeForce-powered laptops with availability starting in Q1 and the most to ship in Q2. These laptops are the first to use our high-end GeForce RTX 2080 SUPER and 2070 SUPER GPUs, which have been available for desktop since last summer.

隨著許多零售店的關閉，對我們產品的需求已經非常有效地轉移到全球的電子零售渠道。遊戲筆記本電腦收入加速至 6 個季度以來最快的同比增長。我們正在與我們的原始設備製造商、渠道合作夥伴合作，以滿足從事在家工作、學習和娛樂的專業人士和學生日益增長的需求。 4 月初，我們的全球 OEM 合作夥伴宣布推出創紀錄的 100 台採用 NVIDIA GeForce 的筆記本電腦，第一季度開始供貨，第二季度出貨最多。這些筆記本電腦是第一款使用我們的高端 GeForce RTX 2080 SUPER 和 2070 SUPER GPU 的筆記本電腦，這些 GPU 從去年夏天就已用於台式機。

In addition, OEMs are bringing to market laptops based on the RTX 2060 GPU at just $999, a price point that enables a larger audience to take advantage of the power and features of RTX, including its unique ray tracing and AI capabilities. These launches are well-timed as mobile and remote computing needs accelerate.

此外，OEM 將基於 RTX 2060 GPU 的筆記本電腦以 999 美元的價格推向市場，這一價位使更多的受眾能夠利用 RTX 的強大功能和功能，包括其獨特的光線追踪和 AI 功能。隨著移動和遠程計算需求的加速，這些發布恰逢其時。

The global rise in gaming also lifted sales of NVIDIA Nintendo Switch and our console business, driving strong growth both sequentially and year-over-year. We collaborated with Microsoft and Mojang to bring RTX ray tracing to Minecraft, the world's most popular game with over 100 million gamers monthly and over 100 billion total views on YouTube. Minecraft with RTX looks astounding with realistic shadows and reflections. Light that reflects, refracts and scatters through surfaces as naturalistic effects like fog. Reviews for it are off the charts. Ars Technica called it a jaw-dropping stunner, and PC World said it was glorious to behold.

全球遊戲市場的增長也提升了 NVIDIA Nintendo Switch 和我們的遊戲機業務的銷售額，推動了環比和同比的強勁增長。我們與 Microsoft 和 Mojang 合作，將 RTX 光線追踪引入 Minecraft，這是世界上最受歡迎的遊戲，每月有超過 1 億遊戲玩家，在 YouTube 上的總瀏覽量超過 1000 億。帶有 RTX 的 Minecraft 看起來令人驚嘆，具有逼真的陰影和反射。光線通過表面反射、折射和散射，形成像霧一樣的自然效果。對它的評論不在圖表之列。 Ars Technica 稱其為令人瞠目結舌的絕技，PC World 稱其光彩奪目。

Our RTX technology stands apart, not only with our 2-year lead in ray tracing but with its use of AI to speed up and enhance games using the Tensor Core silicon on our RTX class GPUs. We introduced the next version of our AI algorithm called Deep Learning Super Sampling. In real time, DLSS 2.0 can fill the missing bits from every frame, doubling performance. It represents a major step function from the original, and it can be trained on nongaming-specific images, making it universal and easy to implement.

我們的 RTX 技術與眾不同，不僅因為我們在光線追踪方面擁有 2 年的領先優勢，而且還通過我們的 RTX 級 GPU 上的 Tensor Core 芯片使用 AI 來加速和增強遊戲。我們介紹了稱為深度學習超級採樣的 AI 算法的下一個版本。 DLSS 2.0 可以實時填補每一幀的缺失位，使性能翻倍。它代表了原始的一個主要步驟函數，它可以在非遊戲特定的圖像上進行訓練，使其通用且易於實現。

The value and momentum of our RTX GPUs continue to grow. We have a significant upgrade opportunity over the next year with the rise and tide of RTX-enabled games, including major blockbusters like Minecraft and Cyberpunk. Let me also touch on our game streaming service, GFN, which exited beta this quarter. It gives gamers access to more than 650 games with another 1,500 in line to get onboarded. These include Epic Games, Fortnite, which is the most played game on GFN; and other popular titles such as CONTROL, Destiny 2 and League of Lighting in the fall. Since launching in February, GFN has added 2 million users around the world, with both sign-ups and hours of gameplays boosted by stay-at-home measures. GFN expands our market reach to the billions of gamers with underpowered devices. It is the most publisher-friendly, developer-friendly game streaming service with the greatest number of games and the only one that supports ray tracing.

我們 RTX GPU 的價值和發展勢頭持續增長。隨著支持 RTX 的遊戲（包括 Minecraft 和 Cyberpunk 等主要大片）的興起和浪潮，我們在明年有一個重大的升級機會。讓我也談談我們的遊戲流媒體服務 GFN，它在本季度退出了測試版。它使遊戲玩家可以訪問超過 650 款遊戲，另外還有 1,500 款在線遊戲可供使用。其中包括 Epic Games、Fortnite，這是 GFN 上玩得最多的遊戲；以及秋季的其他熱門遊戲，例如 CONTROL、Destiny 2 和 League of Lighting。自 2 月推出以來，GFN 已在全球新增 200 萬用戶，通過居家措施提高了註冊人數和遊戲時長。 GFN 將我們的市場範圍擴大到數十億設備性能不足的遊戲玩家。它是對發行商最友好、對開發者最友好的遊戲流媒體服務，擁有最多的遊戲，也是唯一支持光線追踪的服務。

Moving to Pro Visualization. Revenue was $307 million, up 15% year-on-year and down 7% sequentially. Year-on-year revenue growth accelerated in Q1 driven by laptop workstations and Turing adoption.

轉向專業可視化。收入為 3.07 億美元，同比增長 15%，環比下降 7%。受筆記本電腦工作站和圖靈採用的推動，第一季度收入同比增長加速。

We are seeing continued momentum in our ecosystem for RTX ray tracing. We now have RTX support for all major rendering visualization and design software packages, including Autodesk Maya, Dassault's CATIA, Pixar's RenderMan, Chaos Group's V-Ray and many others. Autodesk has announced that the latest release of VRED, its automotive 3D visualization software, supports NVIDIA RTX GPUs. This enables designers to take advantage of RTX to produce more like-life designs in a fraction of the time versus CPU-based systems. Over 45 leading creative and design applications now take advantage of RTX, driving a sustained upgrade opportunity for Quadro-powered systems while also expanding their reach.

我們看到 RTX 光線追踪生態系統的持續發展勢頭。我們現在為所有主要的渲染可視化和設計軟件包提供 RTX 支持，包括 Autodesk Maya、Dassault 的 CATIA、Pixar 的 RenderMan、Chaos Group 的 V-Ray 等等。 Autodesk 宣布其汽車 3D 可視化軟件 VRED 的最新版本支持 NVIDIA RTX GPU。與基於 CPU 的系統相比，這使設計人員能夠利用 RTX 在更短的時間內製作出更逼真的設計。現在，超過 45 種領先的創意和設計應用程序利用了 RTX，為 Quadro 供電的系統帶來了持續升級的機會，同時也擴大了它們的覆蓋範圍。

We see strong demand in verticals, including health care, media and entertainment and higher education, among others. Higher health care demand was fueled in part by COVID-19 related research at Siemens, Oxford and Caption Health. Caption Health received FDA clearance for an update to its AI-guided ultrasound, making it easier to perform diagnostics-quality cardiac ultrasounds. And in media and entertainment, demand increased as companies like Disney deployed remote workforce initiatives.

我們看到垂直領域的需求強勁，包括醫療保健、媒體和娛樂以及高等教育等。西門子、牛津和 Caption Health 的 COVID-19 相關研究部分推動了更高的醫療保健需求。 Caption Health 獲得了 FDA 的批准，以更新其 AI 引導的超聲，使其更容易執行診斷質量的心臟超聲。在媒體和娛樂領域，隨著迪士尼等公司部署遠程勞動力計劃，需求增加。

Turning to automotive and robotic autonomous machines. Automotive revenue was $155 million, down 7% year-on-year and down 5% sequentially. The automotive industry is seeing a significant impact from the pandemic, and we expect that to affect our revenue in the second quarter as well, likely declining about 40% from Q1.

轉向汽車和機器人自主機器。汽車收入為 1.55 億美元，同比下降 7%，環比下降 5%。汽車行業正受到大流行的重大影響，我們預計這也會影響我們第二季度的收入，可能比第一季度下降約 40%。

Despite the near-term challenges, our important work continues. We believe that every machine that moves someday will have autonomous capabilities. During the quarter, Xpeng introduced the P7, an all-electric sports sedan with innovative Level 3 automated driving features, powered by the NVIDIA DRIVE AGX Xavier AI compute platform. Our open, programmable, software-defined platform enables Xpeng to run its proprietary software while also delivering over-the-air updates for new driving features and capabilities. Production deliveries of the P7 with NVIDIA DRIVE begin next month.

儘管近期面臨挑戰，但我們的重要工作仍在繼續。我們相信，有朝一日每台移動的機器都將具有自主能力。在本季度，小鵬推出了 P7，這是一款具有創新 3 級自動駕駛功能的全電動運動轎車，由 NVIDIA DRIVE AGX Xavier AI 計算平台提供支持。我們開放、可編程、軟件定義的平台使小鵬能夠運行其專有軟件，同時還為新的駕駛特性和功能提供無線更新。搭載 NVIDIA DRIVE 的 P7 將於下個月開始生產交付。

Our Ampere architecture will power our next-generation NVIDIA DRIVE platform called Orin, delivering more than 6x the performance of Xavier Solutions and 4x better power efficiency. With Ampere scalability, the DRIVE platform will extend from driverless robotaxis all the way down to in windshield driver assistant systems sipping just a few watts of power. Customers appreciate the top-to-bottom platform all based on a single architecture, letting them build one software-defined platform for every vehicle in their fleet.

我們的 Ampere 架構將為我們稱為 Orin 的下一代 NVIDIA DRIVE 平台提供動力，提供超過 Xavier Solutions 6 倍的性能和 4 倍的能效。憑藉 Ampere 可擴展性，DRIVE 平台將從無人駕駛機器人出租車一直延伸到僅消耗幾瓦功率的擋風玻璃駕駛員輔助系統。客戶欣賞所有基於單一架構的自上而下的平台，讓他們為車隊中的每輛車構建一個軟件定義的平台。

Lastly, in the area of robotics, we announced that BMW Group has selected the new NVIDIA as a robotics platforms to automate their factories, utilizing logistic robots built on advanced AI computing and visualization technologies.

最後，在機器人領域，我們宣布寶馬集團選擇了新的 NVIDIA 作為機器人平台，利用基於先進人工智能計算和可視化技術的物流機器人實現工廠自動化。

Turning to data center. Quarterly revenue was a record $1.14 billion, up 80% year-on-year and up 18% sequentially, crossing the $1 billion mark for the first time. Announced last week, the A100 is the first Ampere architecture GPU. Although just announced, A100 is in full production, contributed meaningful to Q1 revenue and demand is strong. Overall, data center demand was solid throughout the quarter. It was also broad-based across hyperscale and vertical industry customers as well as across workloads, including training, inference and high-performance computing. We continue to have solid visibility into Q2.

轉向數據中心。季度收入達到創紀錄的 11.4 億美元，同比增長 80%，環比增長 18%，首次突破 10 億美元大關。上周宣布，A100 是第一款安培架構 GPU。雖然剛剛宣布，但 A100 已全面投產，對第一季度的收入做出了有意義的貢獻，需求強勁。總體而言，整個季度的數據中心需求穩定。它還廣泛應用於超大規模和垂直行業客戶以及工作負載，包括培訓、推理和高性能計算。我們繼續對第二季度有堅實的可見性。

The A100 offers the largest leap in performance to date over our 8 generations of GPUs, boosting performance by up to 20x over its predecessor. It is exceptionally versatile, serving as a universal accelerator for the most important high-performance workloads, including AI training and inference as well as data analytics, scientific computing and cloud graphics.

與我們的 8 代 GPU 相比，A100 提供了迄今為止最大的性能飛躍，性能比其前身提升了 20 倍。它非常通用，可作為最重要的高性能工作負載的通用加速器，包括 AI 訓練和推理以及數據分析、科學計算和雲圖形。

Beyond its leap performance and versatility, the A100 introduces new elastic computing technologies that make it possible to bring rightsized computing power to every job. A multi-instance GPU capability allows each A100 to be partitioned into as many as 7 smaller GPU instances. Conversely, multiple A100 interconnected by our third-generation NVLink can operate as one giant GPU for ever larger training tasks. This makes the A100 ideal for both training and for inference. The A100 will be deployed by the world's leading cloud service providers and system builders, including Alibaba cloud, Amazon Web Services, Baidu Cloud, Dell Technologies, Google Cloud platform, HPE and Microsoft Azure, among others. It is also getting adopted by several supercomputing centers, including the National Energy Research Scientific Computing Center and the Jülich Supercomputing Centre in Germany and Argonne National Laboratory.

除了飛躍的性能和多功能性之外，A100 還引入了新的彈性計算技術，可以為每項工作帶來合適的計算能力。多實例 GPU 功能允許將每個 A100 劃分為多達 7 個較小的 GPU 實例。相反，通過我們的第三代 NVLink 互連的多個 A100 可以作為一個巨大的 GPU 運行，以完成更大的訓練任務。這使得 A100 非常適合訓練和推理。 A100 將由全球領先的雲服務提供商和系統構建商部署，包括阿里雲、亞馬遜網絡服務、百度雲、戴爾科技、谷歌云平台、HPE 和微軟 Azure 等。它還被多個超級計算中心採用，包括國家能源研究科學計算中心和德國的於利希超級計算中心以及阿貢國家實驗室。

We launched and shipped the DGX A100, our third-generation DGX and the most advanced AI system in the world. The DGX A100 is configurable from 1 to 56 independent GPUs to deliver elastic software-defined data center infrastructure for the most demanding workloads from AI training and inference to data analytics.

我們推出並交付了 DGX A100，這是我們的第三代 DGX 和世界上最先進的 AI 系統。 DGX A100 可配置 1 到 56 個獨立 GPU，為從 AI 訓練和推理到數據分析的最苛刻工作負載提供彈性的軟件定義數據中心基礎設施。

We announced 2 products for edge AI: the EGX A100 for larger commercial off-the-shelf servers; and the EGX Jetson Xavier NX for micro-edge servers. Supported by full AI optimized cloud, native and secure software, the EGX platform is built for AI computing at the edge. With the EGX, hospitals, retail stores, farms and factories can securely carry out real-time processing of the massive amounts of data streaming from trillions of edge sensors. NVIDIA EGX makes it possible to securely, deploy and manage and update fleets of servers remotely. EGX is also ideal for the massive computational challenge of 5G networks, which we are working on with our partners like Ericsson and Mavenir.

我們宣布了 2 款用於邊緣 AI 的產品：用於大型商用現成服務器的 EGX A100；以及用於微型邊緣服務器的 EGX Jetson Xavier NX。 EGX 平台由全面的 AI 優化雲、原生和安全軟件支持，專為邊緣的 AI 計算而構建。借助 EGX，醫院、零售店、農場和工廠可以安全地實時處理來自數万億個邊緣傳感器的大量數據流。 NVIDIA EGX 使遠程安全、部署、管理和更新服務器群成為可能。 EGX 也是應對 5G 網絡的大規模計算挑戰的理想選擇，我們正在與愛立信和 Mavenir 等合作夥伴合作。

Additionally, we announced CUDA 11 and other important software harnessing the A100's performance and universatility (sic) [universality] to accelerate 3 of the most complex and fast-growing workloads: recommendation systems, conversational AI and data science.

此外，我們還發布了 CUDA 11 和其他重要軟件，它們利用 A100 的性能和通用性 (sic) [通用性] 來加速 3 個最複雜和快速增長的工作負載：推薦系統、對話式 AI 和數據科學。

First, NVIDIA Merlin is a deep recommendator (sic) [recommender] application framework that enables developers to quickly build state-of-the-art recommendation systems, leveraging our pretrained models. With billions of users and trillions of items on the Internet, deep recommendators are the critical engine powering virtually every internet service.

首先，NVIDIA Merlin 是一個深度推薦器 (sic) [推薦器] 應用程序框架，使開發人員能夠利用我們的預訓練模型快速構建最先進的推薦系統。互聯網上有數十億用戶和數万億個項目，深度推薦器是支持幾乎所有互聯網服務的關鍵引擎。

Second, NVIDIA Jarvis is a GPU-accelerated application framework that makes it easy for developers to create, deploy and run end-to-end real-time conversational AI applications that understand terminology unique to each company and its customers using both vision and speech. Demand for these applications are surging. Amid the shift to working from home, telemedicine and remote learning.

其次，NVIDIA Jarvis 是一個 GPU 加速的應用程序框架，它使開發人員可以輕鬆地創建、部署和運行端到端實時對話式 AI 應用程序，這些應用程序可以使用視覺和語音來理解每個公司及其客戶所特有的術語。對這些應用的需求正在激增。在向在家工作、遠程醫療和遠程學習的轉變中。

And third, in the field of data science and data analytics, we announced that we are bringing end-to-end GPU acceleration to Apache Spark, an analytics engine for big data processing that uses more than 500,000 data scientists worldwide. Native GPU acceleration for the entire Spark pipeline, from extracting, transforming and loading the data to training to inference, delivers the performance and the scale needed to finally connect the potential of big data with the power of AI. Adobe has achieved a 7x performance improvement and a 90% cost savings in an initial test using GPU-accelerated data analytics with Spark.

第三，在數據科學和數據分析領域，我們宣布我們將為 Apache Spark 帶來端到端 GPU 加速，Apache Spark 是一種用於大數據處理的分析引擎，在全球擁有超過 500,000 名數據科學家。整個 Spark 管道的原生 GPU 加速，從提取、轉換和加載數據到訓練到推理，提供最終將大數據的潛力與 AI 的力量聯繫起來所需的性能和規模。 Adobe 在使用 Spark 的 GPU 加速數據分析的初始測試中實現了 7 倍的性能提升和 90% 的成本節約。

Our accelerated computing platform continues to gain momentum, underscored by the tremendous success of GTC Digital, our annual GPU technology conference, which shifted this spring to an online format. More than 55,000 online developers and AI research registered for the online event, which includes hundreds of hours of free content from AI practitioners and industry experts who leverage NVIDIA's platforms.

我們的加速計算平台繼續獲得動力，我們的年度 GPU 技術會議 GTC Digital 的巨大成功突顯了這一點，該會議於今年春天轉變為在線形式。超過 55,000 名在線開發人員和 AI 研究人員註冊了在線活動，其中包括來自利用 NVIDIA 平台的 AI 從業者和行業專家的數百小時免費內容。

Our ecosystem is now 1.8 million developers strong. Times like these truly test a computing platform's metal in the utility it brings to scientist racing for solutions. Researchers around the world are deploying our GPU computing platform in the fight against COVID-19. Scientists are combining AI simulation to detect changes in pneumonia cases, sequence, the virus and seek effective biomolecular compounds for a vaccine or treatment.

我們的生態系統現在擁有 180 萬強大的開發人員。像這樣的時代真正測試了計算平台的金屬在它為科學家競相解決方案帶來的效用中的作用。世界各地的研究人員正在部署我們的 GPU 計算平台來對抗 COVID-19。科學家們正在結合人工智能模擬來檢測肺炎病例、序列、病毒的變化，並尋找用於疫苗或治療的有效生物分子化合物。

The first breakthrough came from researchers at the University of Texas at Austin and National Institute of Health, who used the GPU-accelerated application to create the first 3D atomic scale map of virus using NVIDIA GPUs. This was followed by researchers at Oak Ridge National Laboratory who screened 8,000 compounds to identify 77 promising drug targets using the world's fastest supercomputer, Summit, which is powered by more than 27,000 NVIDIA GPUs.

第一個突破來自德克薩斯大學奧斯汀分校和美國國立衛生研究院的研究人員，他們使用 GPU 加速的應用程序使用 NVIDIA GPU 創建了第一個 3D 原子尺度病毒圖。緊隨其後的是橡樹嶺國家實驗室的研究人員，他們使用世界上最快的超級計算機 Summit 篩選了 8,000 種化合物，以確定 77 個有前途的藥物靶點，該超級計算機由超過 27,000 個 NVIDIA GPU 提供支持。

The V100 GPUs at Oak Ridge are in high demand as they can analyze 17 million compound protein combinations in a day. They'll help understand the virus spread pattern, the University of California, San Diego, researchers ported their microbiomic analysis software to GPUs in the San Diego supercomputing cluster of 500x analysis speed up from what some people are more susceptible to the virus.

Oak Ridge 的 V100 GPU 需求量很大，因為它們一天可以分析 1700 萬種複合蛋白質組合。他們將幫助了解病毒的傳播模式，加利福尼亞大學聖地亞哥分校的研究人員將他們的微生物組分析軟件移植到聖地亞哥超級計算集群中的 GPU 上，分析速度比某些人更容易感染病毒的速度提高了 500 倍。

Okay. Moving to the rest of the P&L. Q1 GAAP gross margins was 65.1% and non-GAAP was 65.8%, up sequentially and year-on-year, primarily driven by GeForce GPU product mix and higher data center sales. Q1 GAAP operating expenses were $1.03 billion, and non-GAAP operating expenses were $821 million, up 10% and 9% year-on-year, respectively. Q1 GAAP EPS was $1.47, up 130% from a year earlier, and non-GAAP EPS was $1.80, up 105% from a year ago. Q1 cash flow from operations was $909 million.

好的。轉到損益表的其餘部分。第一季度 GAAP 毛利率為 65.1%，非 GAAP 毛利率為 65.8%，環比和同比增長，主要受 GeForce GPU 產品組合和更高的數據中心銷售額的推動。 Q1 GAAP 運營費用為 10.3 億美元，非 GAAP 運營費用為 8.21 億美元，同比分別增長 10% 和 9%。第一季度 GAAP 每股收益為 1.47 美元，同比增長 130%，非 GAAP 每股收益為 1.80 美元，同比增長 105%。第一季度運營現金流為 9.09 億美元。

Before I turn to the outlook, let me make a few comments on our Mellanox acquisition. Beyond the strong strategic and cultural fit that Jensen has discussed, Mellanox has exceptionally strong financial profile. The company reported revenue of $429 million in its March quarter, accelerating to 40% year-on-year growth, with GAAP and non-GAAP gross margins in the mid- to high 60% range. We expect the acquisition to be immediately accretive to non-GAAP gross margins, non-GAAP earnings per share and free cash flow. We aim to retain the full Mellanox team and accelerate investments in our combined road map as we jointly innovate on our shared vision for the future of accelerated computing.

在談到前景之前，讓我對收購 Mellanox 發表一些評論。除了 Jensen 所討論的強大的戰略和文化契合度之外，Mellanox 還擁有異常強大的財務狀況。該公司報告其 3 月季度的收入為 4.29 億美元，同比增長 40%，GAAP 和非 GAAP 毛利率在 60% 的中高範圍內。我們預計此次收購將立即增加非 GAAP 毛利率、非 GAAP 每股收益和自由現金流。我們的目標是保留完整的 Mellanox 團隊並加快對我們合併路線圖的投資，因為我們共同創新我們對未來加速計算的共同願景。

With that, let me turn to the outlook of the second quarter of fiscal 2021, which includes a full quarter contribution from Mellanox. We have assumed in our outlook the potential ongoing impact from COVID-19. We expect our automotive platform sales to be down 40% on a sequential basis and Pro Viz to decline sequentially. In gaming, while we will likely see ongoing impact from the partial operations or closures of iCafes and retail stores, we expect that to be largely offset by a shift to e-tail channels. Overall, the precise magnitude of the impact is difficult to predict, given uncertainties around the reopening of the economy. Overall, we expect second quarter revenue to be $3.65 billion, plus or minus 2%. The contribution of Mellanox revenue is likely to be in the low teens percentage range of our total Q2 revenue. We are providing this breakout to help with comparability between Q1 and Q2. But going forward, it will become an integrated part of our data center market platform.

有了這個，讓我轉向 2021 財年第二季度的展望，其中包括 Mellanox 的整個季度貢獻。我們在展望中假設了 COVID-19 的潛在持續影響。我們預計我們的汽車平台銷售額將環比下降 40%，Pro Viz 將環比下降。在遊戲方面，雖然我們可能會看到部分運營或 iCafes 和零售店關閉的持續影響，但我們預計這將在很大程度上被轉向電子零售渠道所抵消。總體而言，鑑於經濟重新開放存在不確定性，影響的確切程度難以預測。總體而言，我們預計第二季度收入為 36.5 億美元，上下浮動 2%。 Mellanox 收入的貢獻可能在我們第二季度總收入的低百分比範圍內。我們提供此突破以幫助提高第一季度和第二季度之間的可比性。但展望未來，它將成為我們數據中心市場平台的一個組成部分。

GAAP and non-GAAP gross margins are expected to be 58.6% and 66%, respectively, plus or minus 50 basis points. The sequential decline in GAAP gross margins primarily reflects an increase in acquisition-related costs, most of which are nonrecurring.

GAAP 和非 GAAP 毛利率預計分別為 58.6% 和 66%，上下浮動 50 個基點。 GAAP 毛利率的連續下降主要反映了與收購相關的成本增加，其中大部分是非經常性的。

GAAP and non-GAAP operating expenses are expected to be approximately $1.52 billion, and $1.04 billion, respectively. The sequential change in GAAP operating expenses reflects an increase in stock-based compensation and acquisition-related costs. GAAP and non-GAAP operating expenses for the full year are expected to be approximately $5.7 billion and $4.1 billion, respectively. For the full year, stock-based compensation and acquisition-related costs also influence.

GAAP 和非 GAAP 運營費用預計分別約為 15.2 億美元和 10.4 億美元。 GAAP 運營費用的連續變化反映了基於股票的薪酬和收購相關成本的增加。全年 GAAP 和非 GAAP 運營費用預計分別約為 57 億美元和 41 億美元。全年，股票薪酬和收購相關成本也有影響。

GAAP and non-GAAP OI&E are both expected to be an increase of approximately $50 million and $45 million, respectively. GAAP and non-GAAP tax rates are both expected to be 9%, plus or minus 1%, excluding discrete items. Capital expenditures are expected to be approximately $225 million to $250 million. Further financial details are included in the CFO commentary and other information available on our IR website.

GAAP 和非 GAAP OI&E 預計將分別增加約 5000 萬美元和 4500 萬美元。 GAAP 和非 GAAP 稅率預計均為 9%，正負 1%，不包括離散項目。資本支出預計約為 2.25 億美元至 2.5 億美元。更多財務細節包含在 CFO 評論和我們投資者關係網站上提供的其他信息中。

New this quarter, we have also posted an investor presentation summarizing our results and key highlights. In closing, let me highlight upcoming events for the financial community. Next Thursday, May 28, we will webcast a presentation and Q&A with Jensen on our recent product announcement moderated by Evercore. We will also be at Cowen's TMT Conference on May 27; Morgan Stanley's Cloud Secular Winners Conference on June 1; BoFa's Technology Conference on June 2; Needham's Fourth Automotive Technology Conference on June 3 and Nasdaq Investor Conference on June 16.

本季度新增功能，我們還發布了一份投資者報告，總結了我們的業績和主要亮點。最後，讓我強調一下金融界即將發生的事件。下週四，5 月 28 日，我們將與 Jensen 就我們最近由 Evercore 主持的產品公告進行網絡直播演示和問答。我們還將在 5 月 27 日參加 Cowen 的 TMT 會議；摩根士丹利 6 月 1 日的雲長期贏家大會； BoFa 6 月 2 日技術大會； 6 月 3 日李約瑟第四次汽車技術會議和 6 月 16 日納斯達克投資者會議。

Operator, we will now open for questions. Can you please poll for questions, please.

接線員，我們現在開始提問。請您投票提問。

(Operator Instructions) Aaron Rakers with Wells Fargo.

（操作員說明）富國銀行的 Aaron Rakers。

Congratulations on a solid quarter. Colette, I'm curious of your commentary around visibility in the data center side, that that's comments over the last couple of quarters, how would you characterize your visibility today relative to maybe what it was last quarter? And how do we think about the visibility in the context of trends maybe into the back half of the calendar year.

祝賀一個堅實的季度。 Colette，我很好奇你對數據中心方面可見性的評論，這是過去幾個季度的評論，你如何描述你今天相對於上一季度的可見性？以及我們如何考慮趨勢背景下的可見性可能會進入日曆年的後半段。

Thanks, Will, for the question. You are correct. We have indicated a couple of quarters ago that we were starting to see improved visibility after we came out of the digestion period in the prior overall fiscal year. As we move into Q2, we still have visibility and solid visibility into our Q2 results for overall data centers. So at this time, I'd say they are relatively about the same of what we had seen going into the Q1 period. And we think that is a true indication of their excitement about our platform and most particularly our excitement regarding A100, and that's launched and its additional products.

謝謝，威爾，這個問題。你是對的。幾個季度前，我們已經表示，在我們走出上一財年的消化期後，我們開始看到能見度有所提高。隨著我們進入第二季度，我們仍然對整個數據中心的第二季度業績有可見性和可靠的可見性。所以在這個時候，我想說它們與我們在第一季度看到的情況大致相同。我們認為這真實地表明了他們對我們平台的興奮，尤其是我們對 A100 的興奮，以及它的推出及其附加產品。

Now regarding the second half of the year, as you know, we have seen broad-based growth in both the hyperscale and the vertical industries, both of them in terms of at record levels. In our Q1 results. And we see in terms of inferencing continuing to grow as well, as well as we're also expanding in terms of edge AI. Our strong demand of the A100 products, including the Delta Board, but also in terms of our DGXs, was just starting an initial ramp. However, we do guide only 1 quarter at a time. So it's still a little bit too early for us to give a true certainty in terms of the macro situation that's in front of us. But again, we feel very good about the demand for A100.

如您所知，現在就下半年而言，我們已經看到超大規模和垂直行業的廣泛增長，這兩個行業都達到了創紀錄的水平。在我們的第一季度結果中。我們看到在推理方面也在繼續增長，而且我們也在邊緣 AI 方面進行擴展。我們對 A100 產品（包括 Delta Board）以及我們的 DGX 的強勁需求才剛剛開始起步。但是，我們一次只指導 1 個季度。因此，我們現在就擺在我們面前的宏觀形勢給出真正的確定性還為時過早。但同樣，我們對 A100 的需求感到非常滿意。

Your next question comes from Stacy Rasgon with Bernstein Research.

您的下一個問題來自 Bernstein Research 的 Stacy Rasgon。

I first wanted to follow-up on your gaming commentary. You sort of mentioned a couple of offsets. COVID potentially still a headwind, e-tail or tailwind and maybe offsetting each other. Were you trying to suggest that those did offset completely and gaming was kind of flattish into Q2? Because I know it has a typical seasonal pattern switches typically up. I guess what were you trying to say with those kind of factors? And what are the kinds of things we should be thinking about when it comes to seasonality, Colette, into Q2 around that business segment?

我首先想跟進您的遊戲評論。你有點提到了幾個偏移量。 COVID 可能仍然是逆風、電子尾風或順風，並且可能相互抵消。您是否試圖暗示這些確實完全抵消了，並且遊戲在第二季度有點平淡？因為我知道它有一個典型的季節性模式，通常會向上切換。我猜你想用這些因素說什麼？當談到季節性問題時，我們應該考慮哪些事情，科萊特，進入第二季度圍繞該業務部門？

So let me start, and I'll see if Jensen also wants to add on to it. I think you're talking about our sequential between Q1 and Q2. Some of the...

所以讓我開始吧，我會看看 Jensen 是否也想加入它。我認為您是在談論我們在第一季度和第二季度之間的順序。某些...

Yes. That's right.

是的。這是正確的。

Right. Some of the pieces that we had seen related to COVID-19 in Q1 may carry over into Q2. COVID-19, in fact, had an impact in terms of our retail channels as well as our iCafes. However, as we discussed, efficiently, moved to overall e-tail. We have normally been seasonally down in desktop between Q1 and Q2, and that will likely happen. But we do see the strength in terms of laptops and overall consoles as we move for Q1 to Q2. So in summary, we do expect grow sequentially between Q1 and Q2 for our overall gaming business. And I'll turn it over to Jensen to see if he has additional commentary.

正確的。我們在第一季度看到的一些與 COVID-19 相關的部分可能會延續到第二季度。事實上，COVID-19 對我們的零售渠道和 iCafes 產生了影響。然而，正如我們所討論的，有效地轉移到了整體電子零售。我們通常在第一季度和第二季度之間的台式機季節性下降，這很可能會發生。但隨著我們從第一季度到第二季度，我們確實看到了筆記本電腦和整體遊戲機方面的實力。因此，總而言之，我們確實預計我們的整體遊戲業務在第一季度和第二季度之間會連續增長。我會把它交給 Jensen 看他是否有其他評論。

No, that was great. That was fantastic.

不，那太好了。那太棒了。

Yes. I guess just to follow up on that, though, if it's growing. I mean like in prior years, we've seen it grow like very strong double digits. Obviously, the mix of the business was different back then. But do you think that the kind of -- I mean are we thinking kind of it's up somewhat? You don't -- is there any chance that it could be up like on -- for what we've seen in terms of like typical levels in the past? Like can you give us any sense of magnitude, that would be really helpful?

是的。不過，如果它在增長，我想只是為了跟進。我的意思是，就像往年一樣，我們已經看到它以非常強勁的兩位數增長。顯然，當時的業務組合有所不同。但是你認為那種 - 我的意思是我們認為它有點上升嗎？你不 - 它有沒有可能像過去一樣 - 我們在過去看到的典型水平？就像你能給我們任何規模感，那真的很有幫助嗎？

Yes. I think when we think about that sequential growth, we'll probably be in the low -- moving up to probably the mid-single digits in terms of -- that's what our guidance right now, and we'll just have to see how the quarter goes.

是的。我認為當我們考慮這種連續增長時，我們可能會處於低位 - 可能會上升到中個位數 - 這就是我們現在的指導，我們只需要看看如何季度過去了。

Yes. That's very helpful.

是的。這很有幫助。

Stacy, the thing that I would add is this. I would say, I think the guidance is exactly what Colette mentioned. But if you look at the big picture, there's a few dynamics that are working really well in our favor. First, of course, is that RTX and ray tracing is just the home run. Minecraft was phenomenal. We have 33 games in the pipe that has already been announced or shipping. Just about every game developers signed on to RTX and ray tracing, and I think it's a foregone conclusion that this is the next generation. This is the way computer graphics is going to be in the future. And so I think RTX is a home run.

Stacy，我要補充的是這個。我想說，我認為指導正是科萊特提到的。但如果你從大局來看，有一些動態對我們來說非常有效。首先，當然，RTX 和光線追踪只是本壘打。我的世界是非凡的。我們有 33 款遊戲已經發布或發售。幾乎每個遊戲開發者都簽署了 RTX 和光線追踪，我認為這是下一代遊戲已成定局。這就是未來計算機圖形學的發展方向。所以我認為 RTX 是本壘打。

The second, the notebooks that we create is just doing great. We got 100 notebooks in gaming. We have 75 notebooks designed for either mobile workstations or what we call NVIDIA studio for designers and creators. And the timing was just perfect. With everybody needing to stay at home, the ability to have a mobile gaming platform and a mobile workstation, it was just perfect timing. And then, of course, you guys know quite well that our Nintendo Switch is doing fantastic. There are 3 -- the top 3 games in the world. The top games in the world today are Fortnite, Minecraft and Animal Crossing. All 3 games are NVIDIA platforms. And so I think we have all the dynamics working in our favor. And then we just got to see how it turns out.

第二，我們創建的筆記本做得很好。我們在遊戲中有 100 台筆記本電腦。我們有 75 款筆記本電腦，專為移動工作站或我們稱為 NVIDIA 工作室的設計師和創作者而設計。時機恰到好處。每個人都需要待在家裡，擁有移動遊戲平台和移動工作站的能力，這正是完美的時機。然後，當然，你們很清楚我們的 Nintendo Switch 做得很棒。有3個——世界排名前3的遊戲。當今世界上最熱門的遊戲是 Fortnite、Minecraft 和 Animal Crossing。所有 3 款遊戲都是 NVIDIA 平台。所以我認為我們的所有動態都對我們有利。然後我們只需要看看結果如何。

Your next question comes from Joe Moore with Morgan Stanley.

您的下一個問題來自摩根士丹利的喬摩爾。

I wanted to ask about the rollout of Ampere how quickly does that roll in the various segments between hyperscale as well as on the DGX side as well as on the HPC side. And is it a smooth transition? Is there -- I remember when you launched Volta, there was a little bit of a transitional pause. Just can you tell us how you see that ramping up with the different customer segments?

我想問一下 Ampere 的推出在超大規模、DGX 端和 HPC 端之間的各個細分市場中的推出速度有多快。它是平穩過渡嗎？有沒有——我記得當你推出 Volta 時，有一點過渡性的停頓。你能告訴我們你如何看待不同客戶群的增長嗎？

Yes. Thanks a lot, Joe. So first of all, taking a step back. Accelerated computing is now common sense in data centers. It wasn't the case when we first launched Volta. If you went back to Volta, Volta was the first generation that the deep learning training in a really serious way, and it was really focused on training. It was focused on training and high-performance computing. We didn't come until later with the inference version called T4.

是的。非常感謝，喬。所以首先，退後一步。加速計算現在是數據中心的常識。當我們第一次推出 Volta 時，情況並非如此。如果你回到 Volta，Volta 是第一代以非常認真的方式進行深度學習訓練的人，而且它真的專注於訓練。它專注於培訓和高性能計算。直到後來，我們才推出了名為 T4 的推理版本。

But over the course of the last 5 years, we've been accelerating workloads that are now diversifying in data centers. If you take a look at most of the hyperscalers, machine learning is now pervasive. Deep learning is now pervasive. The notion of accelerated deep learning and machine learning using our GPUs is now common sense. It didn't used to be. People still saw it as something esoteric. But today, data centers all over the world expect a very significant part of their data center being accelerated with GPUs. The number of workloads that we've accelerated since in the last 5 years have expanded tremendously, whether it's imaging or video or conversational AI or deep recommender systems that probably unquestionably, at this point, the most important machine learning model in the world. And so the number of applications we now accelerate is quite diverse. And so that's really -- that's contributed greatly to the ramp of Ampere. When we came -- when we started to introduce Ampere to the data center, it was very commonsensical to them that they would adopt it. They have a large amount of workload that's already accelerated by NVIDIA GPUs.

但在過去的 5 年中，我們一直在加速工作負載，這些工作負載現在正在數據中心中實現多樣化。如果您看一下大多數超大規模計算機，機器學習現在已經無處不在。深度學習現在無處不在。使用我們的 GPU 加速深度學習和機器學習的概念現在已成為常識。以前不是這樣的。人們仍然認為它是深奧的。但是今天，世界各地的數據中心都希望其數據中心的很大一部分能夠通過 GPU 進行加速。自過去 5 年以來，我們加速的工作負載數量已經大大增加，無論是成像、視頻還是對話式 AI 或深度推薦系統，這可能毫無疑問是目前世界上最重要的機器學習模型。因此，我們現在加速的應用程序數量非常多樣化。所以這真的 - 這對安培的斜坡有很大貢獻。當我們來的時候——當我們開始將 Ampere 引入數據中心時，他們會採用它是非常常識的。他們有大量工作負載已經被 NVIDIA GPU 加速。

And as you know, our GPUs are architecturally compatible from generation to generation. We're forward compatible or backwards compatible. Everything that runs on T4 runs on A100, everything that runs on V100 runs on A100. And so I think the transition is going to be really, really smooth. On the other hand, because V100 and T4 -- which, by the way, V100 and T4 had a great quarter. It was sequentially up. And then on top of that, we grew with the A100 shipment. A100 -- or excuse me, V100 and T4 are now quite broadly adopted in hyperscalers for their AI services, in cloud computing, in a vertical industries, which is almost roughly half of our overall HPC business. All the way out to the edge, which had a great quarter. Much smaller part, of course -- supercomputing is important, but it's a very small part of the high-performance computing. But that's also -- we also shipped A100 to supercomputing centers. And so I think the general sense of it -- the summary of it is that the number of workloads for accelerated computing has continued to grow, the adoption of machine learning and AI and all the cloud and hyperscalers has grown. The common sense of using acceleration is now a foregone conclusion. And so I think we're ramping into a very receptive market with a really fantastic -- with a really fantastic product.

如您所知，我們的 GPU 在架構上一代代兼容。我們向前兼容或向後兼容。在 T4 上運行的一切都在 A100 上運行，在 V100 上運行的一切都在 A100 上運行。所以我認為過渡會非常非常順利。另一方面，因為 V100 和 T4 ——順便說一下，V100 和 T4 有一個很好的季度。它是依次向上的。最重要的是，我們隨著 A100 的出貨而增長。 A100——或者對不起，V100 和 T4 現在在超大規模企業的 AI 服務、雲計算和垂直行業中被廣泛採用，這幾乎占我們整體 HPC 業務的一半。一直到邊緣，那裡有一個很棒的季度。當然，要小得多——超級計算很重要，但它只是高性能計算的一小部分。但這也是——我們還將 A100 運送到超級計算中心。所以我認為它的一般意義 - 它的總結是加速計算的工作負載數量持續增長，機器學習和人工智能的採用以及所有云計算和超大規模計算都在增長。使用加速度的常識現在已成定局。所以我認為我們正在以非常棒的產品進入一個非常容易接受的市場。

Your next question comes from Vivek Arya with Bank of America.

您的下一個問題來自美國銀行的 Vivek Arya。

Congratulations on the strong growth and execution. Just a quick clarification. Colette, 66% kind of the new baseline for gross margin? And then the question, Jensen, for you, is give us a sense for how much inference as a workload and payer as a product are expected to contribute? I'm just curious where you are in terms of growing in the inference and edge AI market? And where are we kind of in the journey of Ampere penetration?

祝賀強勁的增長和執行力。只是一個快速的澄清。 Colette，66% 是毛利率的新基準？然後問題，Jensen，對你來說，是讓我們了解作為工作量的推理和作為產品的付款人預計會做出多少貢獻？我只是好奇您在推理和邊緣 AI 市場的增長情況如何？我們在安培滲透的旅程中處於什麼位置？

So let me start on the first question regarding the gross margin and our gross margin as we look into Q2. We are guiding Q2 non-GAAP gross margins at 66%. This is -- would be another record gross margin quarter just as we finished an overall record level, even as we are continuing right now to ramp our overall Ampere architecture within that. The Q2 also incorporates Mellanox. Mellanox has a very similar overall margins to our overall data center margins as well. But we see this new baseline as a great transition and likely to see some changes as we go forward. However, it's still a little early to see where these gross margins will go. But we're very pleased with the overall guidance right now at 66% for Q2.

因此，讓我從關於第二季度毛利率和我們的毛利率的第一個問題開始。我們將第二季度非 GAAP 毛利率指導為 66%。這將是另一個創紀錄的毛利率季度，就像我們完成了整體創紀錄水平一樣，即使我們現在正在繼續在其中提升我們的整體安培架構。 Q2 還採用了 Mellanox。 Mellanox 的整體利潤率也與我們的整體數據中心利潤率非常相似。但是我們認為這個新的基線是一個很好的過渡，並且在我們前進的過程中可能會看到一些變化。然而，現在看到這些毛利率將走向何方還為時過早。但我們對目前第二季度 66% 的總體指導感到非常滿意。

Accelerated computing is just at the beginning of its journey. If you look at -- I would characterize it as several segments. First is hyperscaler AI microservices, which is all the services that we enjoy today that has AI. Whenever you shop on the web, it recommends a product. When you're watching a movie, it recommends a movie or it recommends a song. All of those -- or recommends news or recommends a friend or recommends a website, the first 10 websites that they recommend. All of these recommenders that are powering the Internet are all based on machine learning today. It's the reason why they're collecting so much data. The more data they can collect, the more they could predict your preference, and that predicting your preference is the core to a personalized Internet. It used to be largely based on CPU approaches. But going forward, it's all based on deep learning approaches. The results are much more superior and a few percentage change in preference prediction accuracy could result in tens of billions of dollars of economics.

加速計算才剛剛開始。如果您看一下- 我會將其描述為幾個部分。首先是超大規模人工智能微服務，這是我們今天享受的所有具有人工智能的服務。每當您在網上購物時，它都會推荐一種產品。當您在看電影時，它會推荐一部電影或一首歌曲。所有這些 - 或推薦新聞或推薦朋友或推薦網站，他們推薦的前 10 個網站。所有這些為互聯網提供動力的推薦系統都基於當今的機器學習。這就是他們收集如此多數據的原因。他們收集的數據越多，就越能預測您的偏好，而預測您的偏好是個性化互聯網的核心。它過去主要基於 CPU 方法。但展望未來，這一切都基於深度學習方法。結果要好得多，偏好預測準確性的幾個百分比變化可能會帶來數百億美元的經濟收益。

And so this is very, very big deal. And the shift towards deep learning in hyperscale micro services or AI micro services is still ramping. Second is cloud. And as you know, cloud is a $100 billion market segment of it today, growing about 40% into $1 trillion opportunity. This -- cloud computing is the single largest IT industry transformation that we have ever seen. The 2 powers that is really -- the force -- the 2 forces that is really driving our data center business is AI and cloud computing. We're perfectly, perfectly positioned to benefit from these 2 powerful forces. So the second is cloud computing. And that journey is -- has a long ways to go. Then the third is industrial edge. In the future, today -- it's not the case today. But the combination of IoT, 5G, industrial 5G and artificial intelligence, it's going to turn every single industry into a tech industry. And whether it's logistics or warehousing or manufacturing or farming, construction, industrial, every single industry will become a tech industry. And there'll be trillions of sensors, and they'll be connected to little micro data centers.

所以這是非常非常重要的事情。超大規模微服務或人工智能微服務向深度學習的轉變仍在加速。二是雲。如您所知，雲是今天價值 1000 億美元的細分市場，增長了約 40%，成為 1 萬億美元的機會。這——雲計算是我們所見過的最大的 IT 行業轉型。真正推動我們數據中心業務的兩種力量是人工智能和雲計算。我們完全可以從這兩種強大的力量中受益。所以第二個是雲計算。而那段旅程——還有很長的路要走。第三是產業優勢。未來，今天——今天不是這樣。但物聯網、5G、工業 5G 和人工智能的結合，將把每一個行業都變成科技行業。無論是物流、倉儲、製造還是農業、建築、工業，每一個行業都將成為科技行業。而且會有數万億個傳感器，它們將連接到小型微型數據中心。

And those data centers will be in the millions. They'll be distributed all over the edge. And that journey has just barely started. We announced 3 very important partners in 3 domains. And they're the lead partners that we felt that people would know, but we have several hundred partners that are working with us on edge AI. We announced Walmart for smart retail. We announced the U.S. Postal Service, the world's largest mail sorting service and logistics service. And then we announced this last quarter, BMW, who is working with us to transform their factory into a robotics, automated factory of the future. And so these 3 applications are great examples of the next phase of artificial intelligence and where Ampere is going to ramp into. And that is just really at its early stages. And so I think it's fair to say that we're really well positioned in the 2 fundamental forces of IT today, data center scale computing and artificial intelligence. And the segments that it's going to make a real impact are all gigantic markets. Hyperscale AI, cloud and edge AI.

這些數據中心將達到數百萬。它們將分佈在整個邊緣。而那段旅程才剛剛開始。我們宣布了 3 個領域的 3 個非常重要的合作夥伴。他們是我們認為人們會知道的主要合作夥伴，但我們有數百個合作夥伴正在與我們合作開發邊緣人工智能。我們宣布了沃爾瑪的智能零售。我們宣布了美國郵政服務，這是世界上最大的郵件分揀服務和物流服務。然後我們在上個季度宣布了寶馬，他們正在與我們合作，將他們的工廠轉變為未來的機器人自動化工廠。因此，這 3 個應用程序是人工智能下一階段以及 Ampere 將進入的領域的絕佳示例。這還只是處於早期階段。因此，我認為可以公平地說，我們在當今 IT 的兩大基本力量——數據中心規模計算和人工智能方面處於有利地位。它將產生真正影響的細分市場都是巨大的市場。超大規模人工智能、雲和邊緣人工智能。

Your next question comes from C.J. Muse with Evercore.

您的下一個問題來自 C.J. Muse with Evercore。

I guess if I could ask 2. Colette, can you help us with what you think the growth rate for Mellanox could look like in calendar '20? And then Jensen, a bigger picture question for you and really not specific to health care, more broad-based. But how do you think about the long-lasting impact of COVID on worldwide demand for AI?

我想如果我能問一下 2. Colette，你能幫我們看看你認為 Mellanox 在 20 年日曆中的增長率嗎？然後是 Jensen，對你來說是一個更大的問題，實際上並不特定於醫療保健，而且基礎更廣泛。但是，您如何看待 COVID 對全球人工智能需求的長期影響？

C.J., can you help me? You cut out in the middle of your sentence to me. Can you repeat the first part of it for me?

C.J.，你能幫幫我嗎？你在你對我的句子中間中斷了。你能幫我重複一下它的第一部分嗎？

No, sorry about that. I'm curious if you could provide a little handholding on what we should think about for growth for Mellanox in calendar '20?

不，對此感到抱歉。我很好奇您能否提供一些關於我們應該為 Mellanox 在 20 年日曆中的增長考慮什麼？

At this time, it's a little early for us. And as you know, we generally just go 1 quarter out, and we're excited to bring the Mellanox team on board so we can start beginning the future of building products together. For the overall margin, their overall performance over the last couple of quarters, they had a great last year. They had a great March quarter as well. And we're just going to have to stay tuned to see equally with them what the second half of the year looks for them. Okay?

這個時候，對我們來說有點早了。如您所知，我們通常只完成 1 個季度的工作，我們很高興能將 Mellanox 團隊加入進來，這樣我們就可以開始共同構建產品的未來。對於整體利潤率，他們在過去幾個季度的整體表現，他們去年表現出色。他們也有一個偉大的三月季度。我們將不得不繼續關注，與他們一樣看到下半年對他們的期待。好的？

Yes, C.J., thanks for the question. This pandemic is really quite tragic, and it's reshaping industries and markets. And I think it's going to be structural. I think it's going to remain. And I think your question is really good because now it's a good time to think about where to double down. There's a few areas that I believe are going to be structurally changed. And I think that once I say it, it'll be very sensible. The first is that the world's enterprise digital transformation and moving to the cloud, that is going to accelerate. Every single company can't afford to rely just on on-prem IT. They have to be much more resilient. And having a hybrid cloud computing infrastructure is going to provide them the resilience they need. And so that's one. And when the world moves and accelerates into this $1 trillion IT infrastructure transformation, which is now $100 billion into that journey, it's growing 40% a year, I wouldn't be surprised to see that accelerate. And so cloud computing AI is going to accelerate because of that. The second is the importance of creating a computational defense system. The defense systems of most nations today are based on radar. And yet in the future, our defense systems are going to detect things that are unseeable. It's going to be infectious disease.

是的，C.J.，謝謝你的提問。這場流行病真的很悲慘，它正在重塑行業和市場。我認為這將是結構性的。我認為它會繼續存在。我認為你的問題非常好，因為現在是考慮在哪裡加倍努力的好時機。我相信有幾個領域會在結構上發生變化。而且我認為一旦我說出來，它就會非常明智。首先是全球企業數字化轉型和向雲遷移，這將加速。每家公司都不能僅僅依靠本地 IT。他們必須更有彈性。擁有混合雲計算基礎架構將為他們提供所需的彈性。這就是其中之一。當世界移動並加速進入這個價值 1 萬億美元的 IT 基礎設施轉型（現在是 1000 億美元），它以每年 40% 的速度增長，看到這種加速我不會感到驚訝。因此，雲計算 AI 將因此加速。第二個是創建計算防禦系統的重要性。今天大多數國家的防禦系統都是基於雷達的。然而在未來，我們的防禦系統將檢測到看不見的東西。這將是傳染病。

And I think every nation and government and scientific lab is now gearing up to think about what does it take to create a national defense system for each country that is based on computational methods? And NVIDIA is an accelerated computing company. We take something that otherwise would take a year in the case of Oak Ridge, and they filter 1 billion compounds in a day. And that's what you need to do. You need to find a way to have an accelerated computational defense system that allows you to find insight, detect early warning ASAP. And then, of course, the computational system has to go through the entire range from mitigation to containment to living within the monitoring. And so scientific labs are going to be gearing up. National labs are going to be gearing up. The third part is AI and robotics. We're going to have to have the ability to be able to do our work remotely.

我認為每個國家、政府和科學實驗室現在都在加緊思考如何為每個國家創建基於計算方法的國防系統？英偉達是一家加速計算公司。在橡樹嶺的情況下，我們採用了原本需要一年時間的東西，他們每天過濾 10 億種化合物。這就是你需要做的。您需要找到一種方法來擁有一個加速的計算防禦系統，讓您能夠發現洞察力，盡快檢測到預警。然後，當然，計算系統必須經歷從緩解到遏製到生活在監控中的整個範圍。因此，科學實驗室將加緊準備。國家實驗室將加緊準備。第三部分是人工智能和機器人。我們將必須有能力遠程完成我們的工作。

NVIDIA has a lot of robots that are helping us in our labs. And without those robots helping us in our labs, we'll have a hard time getting our work done. And so we need to have remote autonomous capability for -- to handle all of these -- either dangerous circumstances to disinfect environments, to fumigate environments autonomously, to clean environments, to be able to interact with people where as little as possible in the event of an outbreak. All kinds of robotics applications are being dreamed up right now to help society forward in the case of another outbreak. And then lastly, I think more and more people are going to work permanently from home. There's a strong movement of companies that are going to support a larger percentage of people working from home. And when people work from home, it's going to clearly increase the single best home entertainment, which is video games. I think video games is going to represent a much larger segment of the overall entertainment budget of society.

NVIDIA 有很多機器人在我們的實驗室中為我們提供幫助。如果沒有這些機器人在我們的實驗室中幫助我們，我們將很難完成工作。因此，我們需要擁有遠程自主能力——處理所有這些——無論是對環境進行消毒、對環境進行自主熏蒸、對環境進行清潔，還是能夠在事件中盡可能少地與人互動的爆發。現在正在構想各種機器人應用程序，以幫助社會在再次爆發的情況下向前發展。最後，我認為越來越多的人將永久在家工作。有一股強大的公司運動將支持更大比例的在家工作的人。當人們在家工作時，它顯然會增加最好的家庭娛樂，那就是視頻遊戲。我認為視頻遊戲將在社會整體娛樂預算中佔據更大的份額。

And so these are some of the trends, I would say. I would say cloud computing, AI. I would say national labs, a computational defense system, robotics and working from home are structural changes that are going to be here to stay. And these dynamics are really good for us.

所以這些是一些趨勢，我會說。我會說云計算，人工智能。我想說國家實驗室、計算防禦系統、機器人技術和在家工作是結構性變化，這些變化將持續存在。這些動態對我們來說真的很好。

Your next question comes from Toshiya Hari with Goldman Sachs.

您的下一個問題來自高盛的 Toshiya Hari。

I had one for Colette and then one for Jensen as well, if I may. Colette, I wanted to come back to the gross margin question. You're guiding July essentially flat sequentially, despite what I'm guessing is better mixed with non-ops coming in and automotive guided down 40% sequentially. I guess the question is, what are some of the offsets that are pulling down gross margins in the current quarter? And sort of related to that, how should we be thinking about the cadence and OpEx going forward, given the 6-month pull in that you guys talked about on the compensation side? And then one quick one for Jensen. I was hoping you could comment on the current trade landscape between the U.S. and China. I feel like you guys shouldn't be impacted in a material way directly nor indirectly. But at the same time, given the critical role you play in scientific computing, I can sort of see a scenario where some people may claim that you guys contribute to efforts outside of the U.S. So if you can kind of speak on that -- speak to that, that will be helpful.

如果可以的話，我給 Colette 買了一個，然後也給 Jensen 買了一個。 Colette，我想回到毛利率問題。儘管我猜想與非運營商的進入和汽車行業的引導性相比，它們會更好地連續下降 40%，但您指導的 7 月份基本持平。我想問題是，在本季度拉低毛利率的一些抵消因素是什麼？與此相關，考慮到你們在薪酬方面談到的 6 個月拉動，我們應該如何考慮前進的節奏和運營支出？然後給 Jensen 一個快速的。我希望你能評論當前美國和中國之間的貿易格局。我覺得你們不應該直接或間接地受到物質上的影響。但同時，考慮到你們在科學計算中扮演的關鍵角色，我可以看到一些人可能聲稱你們為美國以外的努力做出貢獻的情況所以如果你們能就此發表看法——說出來對此，這將是有幫助的。

Thanks, Toshiya, for your question. So regarding our gross margins in the second quarter, our second quarter guide at 66% is up sequentially from even a record level in terms of what we had in terms of Q1. This next record that we hope to achieve with our overall guidance is even with including our overall Ampere architecture. So typically, when we transition to a new architectures, margins can somewhat be a little bit lower on the onset but tend to kind of move up and trend up over time. Additionally, as you articulated, our automotive is lower. But also, we're going to see growth in some of our platforms in gaming such as consoles, which may offset those 2. But overall, there's nothing structural to really highlight other than our mix in business and the ramp of Ampere and its transition.

謝謝你，Toshiya，你的問題。因此，關於我們第二季度的毛利率，我們第二季度的指導為 66%，甚至比第一季度的創紀錄水平還要高。我們希望通過我們的整體指導實現的下一個記錄甚至包括我們的整體安培架構。因此，通常情況下，當我們過渡到新架構時，利潤率在開始時可能會稍低一些，但隨著時間的推移往往會上升並呈上升趨勢。此外，正如您所說，我們的汽車價格較低。而且，我們將看到我們的一些遊戲平台（例如游戲機）的增長，這可能會抵消這兩個平台。但總的來說，除了我們的業務組合和安培的增長及其過渡之外，沒有什麼結構性值得真正突出.

Let's see, the trade tension. We've been living in this environment for some time, Toshiya. And as you know, the trade tension has been in the background for coming up on a year, probably gotten longer. And China's high-performance computing systems are largely based on Chinese electronics anyhow. And so that's -- I think our condition won't materially change going forward.

讓我們看看，貿易緊張局勢。我們已經在這種環境中生活了一段時間，Toshiya。如您所知，貿易緊張局勢在過去一年中一直存在，可能會持續更長時間。無論如何，中國的高性能計算系統很大程度上是基於中國的電子產品。所以這就是 - 我認為我們的情況不會發生重大變化。

So Toshiya, let me respond to your second question that you had for me, which was regarding to our OpEx and our decision to pull forward our overall local into Q2. This is something that we've normally done later in the year. We felt it was prudent during the current COVID-19. Although our employees are quite safe. We just wanted to make sure that their family members also were safe and had the opportunity to have cash upfront. It is about a couple of months, about 4 months earlier than normal, and it is incorporated in our guidance for Q2.

所以 Toshiya，讓我回答你問我的第二個問題，這是關於我們的運營支出以及我們決定將我們的整體本地化推進到第二季度的決定。這是我們通常在今年晚些時候做的事情。在當前的 COVID-19 期間，我們認為這是謹慎的做法。雖然我們的員工很安全。我們只是想確保他們的家人也安全，並有機會預付現金。這比正常情況提前了大約幾個月，大約提前了 4 個月，並且已包含在我們對第二季度的指導中。

Your next question comes from Mark Lipacis with Jefferies.

您的下一個問題來自 Jefferies 的 Mark Lipacis。

A question coming back to the A100. I'm trying to understand how this kind of fits into the evolution of your solution set over time and the evolution of the demand for the applications. Is -- and I guess if I think about it going back, you had a solution, which is largely training based. And then you kind of introduced solutions that were targeted more inferencing. And now you have a solution, it sounds to my understanding that it solves both inferencing and training efficiently. And so I guess I'm wondering is 3 years, 5 years, 10 years down the line, is this part of the kind of general purpose computing or acceleration framework that you had talked about in the past, Jensen, where Ampere is kind of like an Ampere-class product? Or is this -- would you still -- should we still expect to see inferencing-specific solutions in the market and then training-specific solutions and then an Ampere solution for a different class application? If you could provide a framework for thinking about Ampere in those context, I think that would be helpful.

回到 A100 的問題。我試圖了解這種方式如何適應您的解決方案集隨時間的演變以及應用程序需求的演變。是——我想如果我回想一下，你有一個解決方案，主要是基於培訓的。然後您介紹了針對更多推理的解決方案。現在你有了一個解決方案，在我看來，它可以有效地解決推理和訓練問題。所以我想我想知道 3 年、5 年、10 年後，這是否屬於你過去談到的通用計算或加速框架的一部分，Jensen，安培就是像安培級產品？或者這 - 你仍然 - 我們是否仍然期望在市場上看到特定於推理的解決方案，然後是特定於培訓的解決方案，然後是針對不同類別應用的 Ampere 解決方案？如果您可以提供一個框架來在這些情況下思考安培，我認為這會有所幫助。

Yes. Thanks for the call, Mark. Good Question. I think the -- if you take a step back, currently in our data centers, the current setup in data centers, starting from probably all the way back, 6, 7 years ago, but really accelerating in the last 5 years and then really accelerating in the last couple of years, we learned our way into it. There are 3 classes of workloads, and they kind of came into acceleration over time. The first class of workload that we discovered was -- the major workload was deep learning training. Deep learning training. And the ideal setup for that today prior to Ampere or yesterday prior to Ampere is the V100 SXM with NVLink, 8 GPUs on one board, and that architecture is called scale up. It's like a supercomputer architecture. It's like a -- it's like a weather simulation architecture. You're trying to build the largest possible computing node you can for one operating system called scale up. And the second thing that we learned along the way was then cloud computing started to grow because researchers around the world needed to get access to an accelerated platform for developing their machine learning algorithms. And because they have a different degree of budget, and they want to get into it, a little bit more lightly and have the ability to scale up to larger nodes, the perfect model for that was actually a V100 PCI Express, not SXM, but PCI Express that allows you to offer 1 GPU all the way up to many GPUs.

是的。謝謝你的電話，馬克。好問題。我認為 - 如果你退後一步，目前在我們的數據中心，數據中心的當前設置，可能從 6、7 年前開始，但在過去 5 年真正加速，然後真的在過去的幾年裡加速發展，我們學會了進入它。有 3 類工作負載，隨著時間的推移它們會加速。我們發現的第一類工作負載是——主要的工作負載是深度學習訓練。深度學習訓練。今天在 Ampere 之前或昨天在 Ampere 之前的理想設置是帶有 NVLink 的 V100 SXM，一塊板上有 8 個 GPU，這種架構稱為縱向擴展。它就像一個超級計算機架構。這就像一個 - 它就像一個天氣模擬架構。您正在嘗試為一種稱為縱向擴展的操作系統構建盡可能大的計算節點。我們在此過程中學到的第二件事是雲計算開始增長，因為世界各地的研究人員需要訪問一個加速平台來開發他們的機器學習算法。而且因為他們有不同程度的預算，並且他們想要更輕鬆地進入它，並且能夠擴展到更大的節點，因此完美的模型實際上是 V100 PCI Express，而不是 SXM，而是PCI Express 允許您提供 1 個 GPU 到多個 GPU。

And so that versatility, V100 PCI Express, not as scalable in performance as the V100 SXMs, but it was much more flexible for rentals. Cloud renting was really quite ideal. And then we started to get into inference, and we're on our seventh generation of TensorRT, TensorRT 7.0. Along the way, we've been able to accelerate more and more. And today, we largely accelerate every deep learning inference computational graph that's out there. And the ideal GPU for that was something that has the reduced precision, which is called (inaudible), reduced precision not with electronics that is focused more for inference -- and because inference is a scale-out application, where you have millions of queries, and each one of the queries are quite small versus scale up, where you have 1 training job and that 1 training job is running for a day.

因此，V100 PCI Express 的多功能性在性能上不如 V100 SXM 那樣可擴展，但在租賃方面更加靈活。雲租真的很理想。然後我們開始進行推理，我們正在使用我們的第七代 TensorRT，TensorRT 7.0。在此過程中，我們能夠越來越加速。而今天，我們在很大程度上加速了每一個深度學習推理計算圖。理想的 GPU 是精度降低的東西，這被稱為（聽不清），精度降低而不是電子設備更專注於推理——因為推理是一個橫向擴展的應用程序，你有數百萬個查詢，並且每個查詢都非常小，而不是按比例放大，您有 1 個培訓作業並且該 1 個培訓作業正在運行一天。

It could be running for days and sometimes even weeks. And so scale-up application is for 1 user that uses it for a long period of time on a very large machine. Scale out, it's for millions of users, each one of them have a very small query and that query could last hundreds of milliseconds and where ideally, you like to get it done in hundreds of milliseconds. And so notice, I've said 3 different architecture in a data center today. Most data centers today has a storage server, has CPU servers, and it has scale-up acceleration service with Voltas has scaled out servers with GeForce and then it has scale cloud computing, flexible servers based on V100. And so the ability to predict workload is so hard, and therefore, the utilization of these systems will be spiky. And so we created an architecture that allows for 3 things.

它可能會運行數天，有時甚至數週。因此，擴展應用程序適用於在非常大的機器上長時間使用它的 1 個用戶。向外擴展，它適用於數百萬用戶，每個用戶都有一個非常小的查詢，該查詢可能持續數百毫秒，理想情況下，您希望在數百毫秒內完成。請注意，我今天已經在數據中心中提到了 3 種不同的架構。今天的大多數數據中心都有存儲服務器，有 CPU 服務器，它有 Voltas 的縱向擴展加速服務，有 GeForce 橫向擴展的服務器，然後它有基於 V100 的可擴展雲計算、靈活的服務器。因此，預測工作負載的能力是如此之難，因此，這些系統的利用率將非常高。所以我們創建了一個允許三件事的架構。

So things -- the 3 characteristics of Ampere are: number one, it is the greatest generational leap in history. I mean I don't remember a generation where we increased throughput for training and inference by 20x. And it's just a gigantic -- for training and for inference, it is a gigantic leap forward. The second, it's the first architecture that is unified. We could use this to the computational -- the computation engine of Ampere accelerates the moment the data comes into the data center. From data processing, it's called [ETO]; the engine, which many of you probably know, it's the single most important computational engine in the world today for big data. It used to be Hadoop, but now it's Spark. Spark is used all over the world, 16,000 customers. We finally have the ability to accelerate that.

所以事情——安培的三個特點是：第一，它是歷史上最偉大的代際飛躍。我的意思是我不記得我們將訓練和推理的吞吐量提高了 20 倍的一代人。這只是一個巨大的進步——對於訓練和推理來說，這是一個巨大的飛躍。第二，它是第一個統一的架構。我們可以將其用於計算——Ampere 的計算引擎在數據進入數據中心的那一刻加速。從數據處理來說，叫做[ETO]；這個引擎，你們很多人可能都知道，它是當今世界上最重要的大數據計算引擎。它曾經是 Hadoop，但現在是 Spark。 Spark 被全世界使用，有 16,000 名客戶。我們終於有能力加速這一進程。

And then it's -- Ampere is also good for training, deep learning, machine learning, extra boost as well as deep learning, all the way out to inference. And so we now have a unified acceleration platform for the entire workload. And then the third thing is it's the first GPU ever, the first acceleration platform ever that's elastic. You could reconfigure it. You could configure it for either scale up or you can configure it for scale out. When you configure it for scale up, you're gaining a whole bunch of GPUs together using NVLink, and it creates this 1 gigantic GPU. When you want to scale it out, that same computation node becomes 56 small GPUs. Each one of those 56 partitions, each 1 is more powerful than Volta.

然後是——Ampere 也適用於訓練、深度學習、機器學習、額外提升以及深度學習，一直到推理。因此，我們現在擁有一個針對整個工作負載的統一加速平台。然後第三件事是它是有史以來第一個 GPU，第一個彈性加速平台。你可以重新配置它。您可以將其配置為縱向擴展，也可以將其配置為橫向擴展。當您將其配置為擴大規模時，您將使用 NVLink 一起獲得一大堆 GPU，它會創建這 1 個巨大的 GPU。當你想擴展它時，同一個計算節點變成了 56 個小型 GPU。這 56 個分區中的每一個，每一個都比 Volta 更強大。

I mean it's really quite extraordinary. And so Ampere is a breakthrough on all of these fronts for performance for the fact that it unifies the workload, and you can now have 1 acceleration cluster; and then number three, it's elastic. You could use in the cloud, you could use for inference, you could use it for training. And so the versatility of Ampere is the thing that I'm most excited about. And now you could have 1 acceleration cluster that serves all of your needs. That's very helpful.

我的意思是這真的很不尋常。因此，Ampere 是所有這些方面的性能突破，因為它統一了工作負載，您現在可以擁有 1 個加速集群；然後是第三個，它是有彈性的。你可以在雲中使用，可以用於推理，也可以用於訓練。因此，Ampere 的多功能性是我最興奮的事情。現在您可以擁有 1 個加速集群來滿足您的所有需求。這很有幫助。

Your next question comes from Timothy Arcuri with UBS.

您的下一個問題來自瑞銀的 Timothy Arcuri。

Actually I had 2, I guess, Jensen, first for you. Just on the data center business, things have been very strong recently. Obviously, there's always concerns that customers are pulling in CapEx, but it sounds like you have pretty good visibility into July. But I guess last time, most folks also thought that your kind of attrition really was so low that you would be immune into any digestion, but that wasn't the case. So I guess I'm wondering, if things are different now with A100 and whatnot, but my question is how do you handicap your ability to this time, maybe get through any digestion on the CapEx side? And then I guess, second question, Colette, stock comp had been running like 220 a quarter, and the guidance implies that it goes to like 460 a quarter. So it goes up a lot. Is that all executive retention? And is that sort of the right level as you look into 2021?

實際上我有 2 個，我猜，Jensen，首先是給你的。就數據中心業務而言，最近情況非常強勁。顯然，總是有人擔心客戶會吸引資本支出，但聽起來你對 7 月份的可見性非常好。但我想上次，大多數人也認為你的消耗真的很低，以至於你對任何消化都免疫，但事實並非如此。所以我想我想知道，如果現在 A100 和諸如此類的情況有所不同，但我的問題是你如何限制你這次的能力，也許可以通過資本支出方面的任何消化？然後我猜，第二個問題，科萊特，股票組合每季度運行 220 點，而指導暗示它每季度達到 460 點。所以漲了很多。這就是所有的高管保留嗎？當您展望 2021 年時，這種水平是否正確？

Colette, did you want to handle that first? And then I'll do the...

科萊特，你想先處理這個嗎？然後我會做...

Sure. So let me help you out on the overall GAAP adjustments, so the delta between our GAAP OpEx and our non-GAAP OpEx. If you look at it for the full year and what we guided, we probably have about $1.55 billion associated with GAAP level expenses. Keep in mind, there is more in there than just our stock-based compensation. We have also incorporated the accounting that we will do for the overall Mellanox, and a really good portion of those costs are associated with the amortization of intangibles and also in terms of acquisition-related costs and deal fees and onetime items. So our stock-based compensation includes what we need for NVIDIA and also the onboarding of Mellanox. There is some retention with the overall onboarding of Mellanox. But for the most part, it is just working them in to the year for 3 quarters, which is influencing the stock-based compensation.

當然。因此，讓我幫助您了解整體 GAAP 調整，即我們的 GAAP 運營支出和非 GAAP 運營支出之間的差異。如果您查看全年以及我們的指導，我們可能有大約 15.5 億美元與 GAAP 水平的費用相關。請記住，除了我們的股票薪酬之外，還有更多內容。我們還納入了我們將為整個 Mellanox 進行的會計處理，其中很大一部分成本與無形資產的攤銷以及與收購相關的成本、交易費用和一次性項目相關。因此，我們基於股票的薪酬包括我們對 NVIDIA 的需求以及 Mellanox 的入職。 Mellanox 的整體入職有一些保留。但在大多數情況下，這只是讓他們在一年內工作了三個季度，這影響了基於股票的薪酬。

Tim, there are several differences between our condition then and our condition today. So the first -- the first difference is the diversity of workload we now accelerate. Back then, we were early in our inference. We were still early in our inference, and most of the data center acceleration was used for deep learning. And so today, the versatility stands from data processing to deep learning and the number of -- the number of different types of AI models that's being trained for deep learning is growing tremendously from detecting, from training video, from training a model to detecting unsafe video. The natural language understanding the conversational AI to now a gigantic movement towards deep recommender systems. And so the number of different models that are being trained is growing. The size of the models are gigantic. Recommendation systems are gigantic. They're training on models that are hundreds. The data sizes, hundreds of terabytes. Terabytes, hundreds of terabytes. And it would take tens of -- hundreds of servers to hold all of the data that is needed to train these recommender systems. And so the diversity of -- from data analytics to training all the different models to the influence of all different models. We didn't inference recurring elements at a time, which is probably the most important model today. Text language models, speech models are all recurrent, Euronet models. And so those models were early for us at the time. So number one is the diversity of workload. The second is the acceleration of -- to cloud computing. I think that accelerated cloud computing is a movement that is going to be a multiyear if not a decade-long transition. From where we are today, it's only $100 billion industry segment of the IT industry. It's going to be $1 trillion someday, and that movement is just starting. We're also much more diversified out of the clouds. At the time, cloud was largely where our acceleration went for deep learning. And today, hyperscale only represents about half. And so we've diversified significantly out of cloud, not out of cloud, but including vertical industries. And a lot of that has to do with edge AI and inference. And as I mentioned earlier, we're working with Walmart and BMW and USPS, and that's just the tip of the iceberg. And so I think the conditions are a little different. And then what I would say lastly is Ampere. I mean we are -- we've ramped a few weeks. Even though it was quite significant, it was a great ramp. The demand is fantastic. It is the best ramp we've ever had. The demand is the strongest we've ever had in data centers. And we're starting to ramp of a multiyear ramp. And so -- those are some of the differences. I think the conditions are very different.

蒂姆，我們當時的情況和今天的情況有幾個不同之處。所以第一個 - 第一個區別是我們現在加速的工作負載的多樣性。那時，我們還處於推斷的早期階段。我們的推理還處於早期階段，大部分數據中心加速都用於深度學習。所以今天，從數據處理到深度學習的多功能性和數量——從檢測到訓練視頻，從訓練模型到檢測不安全，正在為深度學習訓練的不同類型 AI 模型的數量正在大幅增長視頻。自然語言理解對話式人工智能現在是向深度推薦系統的巨大運動。因此，正在訓練的不同模型的數量正在增長。模型的大小是巨大的。推薦系統是巨大的。他們正在對數百個模型進行訓練。數據大小，數百 TB。太字節，數百太字節。並且需要數十——數百台服務器來保存訓練這些推薦系統所需的所有數據。所以多樣性——從數據分析到訓練所有不同的模型，再到所有不同模型的影響。我們沒有一次推斷重複元素，這可能是當今最重要的模型。文本語言模型、語音模型都是循環的、Euronet 模型。所以這些模型當時對我們來說還很早。所以第一是工作量的多樣性。二是加速——雲計算。我認為加速雲計算是一個多年甚至十年的轉變。從我們今天的情況來看，它只是 IT 行業中價值 1000 億美元的行業部分。總有一天會達到 1 萬億美元，而這一運動才剛剛開始。我們在雲端也更加多樣化。當時，雲主要是我們加速深度學習的地方。而今天，超大規模只佔一半左右。因此，我們在雲之外實現了多元化，不是在雲之外，而是包括垂直行業。其中很多與邊緣人工智能和推理有關。正如我之前提到的，我們正在與沃爾瑪、寶馬和 USPS 合作，而這只是冰山一角。所以我認為條件有點不同。然後我最後要說的是安培。我的意思是我們 - 我們已經增加了幾個星期。儘管它非常重要，但它是一個很棒的坡道。需求太棒了。這是我們有過的最好的坡道。需求是我們在數據中心中遇到過的最強烈的需求。而且我們開始增加多年的斜坡。所以 - 這些是一些差異。我認為條件非常不同。

Your next question comes from Harlan Sur with JPMorgan.

您的下一個問題來自摩根大通的 Harlan Sur。

Jensen, the team has showed the importance of networking, networking fabric and the Mellanox acquisition, like, for example, when you guys move from Volta DGX-1 to Volta DGX-2, you guys didn't change the GPU chipset. But by adding a custom networking fabric chip and more Mellanox network interface cards, among other things, you guys drove a pretty significant improvement in performance per GPU. But now when we think about scaling out compute acceleration to data center skilled implementation, how does Mellanox' Ethernet switching platforms differ from those provided by other large networking OEMs, some of whom have been your long-term partners? And then how does the Cumulus acquisition fit into the switching and networking strategy as well?

Jensen，該團隊已經展示了網絡、網絡結構和收購 Mellanox 的重要性，例如，當你們從 Volta DGX-1 遷移到 Volta DGX-2 時，你們並沒有改變 GPU 芯片組。但是通過添加自定義網絡結構芯片和更多 Mellanox 網絡接口卡等，你們推動了每 GPU 性能的顯著提高。但是現在，當我們考慮將計算加速擴展到數據中心熟練實施時，Mellanox 的以太網交換平台與其他大型網絡 OEM 提供的平台有何不同，其中一些是您的長期合作夥伴？那麼，收購 Cumulus 是否也適合交換和網絡戰略？

Yes. Great. Thanks a lot, Harlan. I appreciate the question. So DGX, you know this is our third-generation DGX and it's really successful. People love it. It's the most advanced AI instrument in the world. If you're a serious AI researcher, this is your instrument. And in the DGX, there are 8 A100s and there are 9 Mellanox NICs, the highest speed NICs they have. And so we have a great appreciation for high-performance networking. High-performance networking and high-performance computing go hand-in-hand. And the reason for that is because the problems we're trying to solve no longer fit in one computer, no matter how big it is. And so it has to be distributed. And when you distribute a computational workload of such intense scale, the communications overhead becomes one of its greatest bottlenecks, which is the reason why Mellanox is so valuable. There's reason why this company is so precious and really a jewel and one of a kind. And so -- and it's not just about the link speed. It's not mostly.

是的。偉大的。非常感謝，哈蘭。我很欣賞這個問題。所以 DGX，你知道這是我們的第三代 DGX，它真的很成功。人們喜歡它。它是世界上最先進的人工智能儀器。如果您是認真的 AI 研究人員，那麼這就是您的工具。而在 DGX 中，有 8 個 A100 和 9 個 Mellanox 網卡，這是他們擁有的最高速度的網卡。因此，我們非常欣賞高性能網絡。高性能網絡和高性能計算齊頭並進。原因是因為我們試圖解決的問題不再適合一台計算機，無論它有多大。所以它必須被分發。當您分配如此大規模的計算工作負載時，通信開銷成為其最大的瓶頸之一，這就是 Mellanox 如此有價值的原因。這家公司如此珍貴，真的是一顆寶石，獨一無二，這是有原因的。所以——這不僅僅是關於鏈接速度。不是大多數。

I mean we just have a deep appreciation for software. It's a combination of architecture and software and electronics design, chip design. And in that combination, Mellanox is just world-class. And that's the reason why they're in 60% of the world's supercomputers. That's why they're in 100% of the AI supercomputers. And their understanding of large-scale distributed computing is second to none. Now in the world of -- and I just talked about scale up. And you're absolutely right.

我的意思是我們只是對軟件有深刻的理解。它是架構、軟件和電子設計、芯片設計的結合。在這種組合中，Mellanox 是世界級的。這就是為什麼它們出現在世界上 60% 的超級計算機中的原因。這就是為什麼它們在 100% 的 AI 超級計算機中。而他們對大規模分佈式計算的理解更是首屈一指。現在在世界 - 我剛剛談到擴大規模。你是絕對正確的。

Now the question is why scale out? And the reason for that is this. This is the reason why they're doing so well. The movement towards disaggregated microservice applications where containers, microservice containers are distributed all over the data center and orchestrated so that the workload could be distributed across a very large hyperscale data center. That architecture -- and you probably know the 3 most important application in my estimation in the world today, number one, would be TensorFlow and PyTorch; number two would be Spark; and number three would be Kubernetes. And you could rank it however you desire.

現在的問題是為什麼要向外擴展？原因是這樣的。這就是他們做得這麼好的原因。向分散的微服務應用程序發展，其中容器、微服務容器分佈在整個數據中心並進行編排，以便工作負載可以分佈在一個非常大的超大規模數據中心。那個架構——你可能知道我估計當今世界上最重要的 3 個應用程序，排名第一的是 TensorFlow 和 PyTorch；第二名是 Spark；第三位是 Kubernetes。你可以隨心所欲地對其進行排名。

And these 3 applications, in the case of Kubernetes, it's a brand-new type of application where the application is broken up with a small pieces and orchestrated across an entire data center. And because it's broken up into small pieces and orchestrate across the entire data center, the networking between the compute nodes becomes the bottleneck again. And that's the reason why they're doing so well. By increasing the network performance by offloading the communications of the CPUs, you increase the throughput of a data center tremendously. And so it's the reason why they had a record quarter last quarter. It's the reason why they've been growing 27% per year. And their stock was back, their integration into the hyperscale cloud companies, they're low latency, they're incredibly low latency of their link makes them really unique, even whether it's Ethernet or InfiniBand in both cases.

而這 3 個應用程序，在 Kubernetes 的例子中，它是一種全新的應用程序類型，應用程序被分解為小塊，並在整個數據中心進行編排。而且由於它被分解成小塊並在整個數據中心進行編排，計算節點之間的網絡再次成為瓶頸。這就是他們做得這麼好的原因。通過卸載 CPU 的通信來提高網絡性能，您可以極大地提高數據中心的吞吐量。這就是他們上個季度創紀錄的原因。這就是他們每年增長 27% 的原因。他們的股票又回來了，他們與超大規模雲公司的整合，他們的延遲很低，他們的鏈接延遲非常低，這讓他們真的很獨特，即使在這兩種情況下都是以太網或 InfiniBand。

And so they're -- it's a really fantastic stack. And then lastly, Cumulus, we would like to integrate -- we would like to innovate in this world where the world is moving away from just a CPU as a compute node. The new computing unit, a software developer is writing a piece of software that runs on the entire data center. In the future, going forward, the computing, the fundamental computing unit is an entire data center. It's so incredible. It's just utterly incredible. You write an application, 1 human could write an application, and it would literally activate an entire data center. And in that world, we would like to be able to innovate from end to end, from networking storage, security. Everything has to be secured in the future so that we can reduce the attack surface down to practically nothing.

所以他們 - 這是一個非常棒的堆棧。最後，Cumulus，我們想要整合——我們想要在這個世界從僅僅作為計算節點的 CPU 轉移的世界中進行創新。新的計算單元，一個軟件開發人員正在編寫一個運行在整個數據中心上的軟件。未來，計算，基本計算單元是整個數據中心。太不可思議了。這簡直太不可思議了。你編寫一個應用程序，一個人可以編寫一個應用程序，它會真正激活整個數據中心。在那個世界中，我們希望能夠從端到端、網絡存儲、安全性等方面進行創新。未來一切都必須得到保護，以便我們可以將攻擊面減少到幾乎沒有。

And so networking storage, security are all completely offloaded, all incredibly low latency, all incredibly high performance and all the way to compute, all the way through the switch. And then the second thing is we'd like to be able to innovate across the entire stack. You know that NVIDIA is just supremely obsessed about software stacks. And the reason for that is because software creates markets. You can't create new markets like we're talking about, whether it's computational health care or autonomous driving or robotic or conversational AI or recommender systems or edge AI. All of that requires software stacks. It takes software to create markets. And so our obsession about software and creating open platforms for the ecosystem and all of our developer partners, Cumulus plays perfectly into that. They are -- they pioneered the open networking stack. And they pioneered, in a lot of ways, software-defined data centers. And so we're super, super excited about the team. And now we have the ability to innovate in a data center scale world from end to end and then from top to bottom of the entire stack.

因此，網絡存儲、安全性都完全卸載了，所有的延遲都非常低，性能非常高，一直到計算，一直到交換機。然後第二件事是我們希望能夠在整個堆棧中進行創新。您知道 NVIDIA 對軟件堆棧非常著迷。原因是軟件創造了市場。你無法像我們所說的那樣創造新市場，無論是計算醫療保健、自動駕駛、機器人或對話式人工智能、推薦系統或邊緣人工智能。所有這些都需要軟件堆棧。創造市場需要軟件。因此，我們對軟件的痴迷以及為生態系統和我們所有的開發合作夥伴創建開放平台，Cumulus 完美地發揮了作用。他們是——他們開創了開放網絡堆棧。他們在很多方面開創了軟件定義的數據中心。所以我們對這個團隊非常非常興奮。現在，我們有能力在數據中心規模的世界中從頭到尾進行創新，然後從整個堆棧的頂部到底部進行創新。

Your next question comes from William Stein with SunTrust.

您的下一個問題來自 SunTrust 的 William Stein。

Jensen, I'd like to focus on something you said. I think it was in one of your earlier responses, you said something about a very significant part of data centers are now accelerated with GPUs. I'm sort of curious how to interpret that. If we think about sort of the evolution of compute architecture going from almost entirely, let's say, [REX and REXs] CPUs to some future day where we have many more accelerators and maybe a much smaller number of CPUs relative to those. Maybe you can talk to us about where we are in terms of that architectural shift and where you think it goes sort of longer term, where we are in the position of that?

Jensen，我想重點談談你所說的。我認為這是在您之前的一個回復中，您說過數據中心的一個非常重要的部分現在使用 GPU 進行加速。我有點好奇如何解釋它。如果我們考慮計算架構的某種演變，從幾乎完全的 [REX 和 REXs] CPU 到未來的某個日子，我們將擁有更多的加速器，並且相對於這些加速器數量可能會少得多。也許你可以和我們談談我們在架構轉變方面的位置，你認為它會在更長期的情況下發展到什麼程度，我們處於什麼位置？

Yes. I appreciate the question. And this, for computer architecture geeks and people who follow history, you know well that in the entire history of time, there are only 2 computing architectures that has made it so far, which is one of them is x86. The other one's ARM in any reasonable way. And if you get an ARM computer, you get an x86 computer, you can program it. And in fact, there's no such thing as an accelerated computing platform until we came along.

是的。我很欣賞這個問題。而這一點，對於計算機架構極客和追尋歷史的人來說，你們很清楚，在整個時間歷史上，到目前為止，只有兩種計算架構取得了成功，其中之一就是 x86。另一個人的 ARM 以任何合理的方式。如果你得到一台 ARM 計算機，你得到一台 x86 計算機，你可以對它進行編程。事實上，在我們出現之前，沒有加速計算平台之類的東西。

And today, we're the only computing -- accelerated computing platform that you could really largely address. We're in every cloud. We're in every computer company. We're in every country. We have every single size, and we accelerate applications from computer graphics to video games to scientific computing to workstations to machine learning to robotics. This journey took 20-some-odd years. Inside our company, it took 20-some-odd years. And we've been focused on accelerated computing since the beginning of our company. And we made a general purpose. We made a general purpose really starting with an endeavor cost Cg, C for graphics, and then it became CUDA. And we've been working on accelerated computing for quite a long time.

而今天，我們是唯一可以真正在很大程度上解決的計算——加速計算平台。我們在每一朵雲中。我們在每個計算機公司。我們在每個國家。我們有各種規模，我們可以加速應用程序，從計算機圖形到視頻遊戲，從科學計算到工作站，從機器學習到機器人技術。這段旅程花費了 20 多年的時間。在我們公司內部，花了 20 多年的時間。自公司成立以來，我們一直專注於加速計算。我們制定了一個通用目的。我們真正從努力成本 Cg、C 用於圖形開始，然後它變成了 CUDA。很長一段時間以來，我們一直致力於加速計算。

And I think at this point, it's a foregone conclusion that accelerated computing has reached the tipping point and is well beyond it. The number of developers this year that support -- that we supported was almost 2 million developers around the world, and it's growing what appears to be exponentially. And so I think accelerated computing is now well established. NVIDIA-accelerated computing is well established. It's common sense, and people who are designing data centers expect to put accelerated computing in it. The question is how much? How much accelerated computing do you use? And what part of the date in your pipeline do you do it? And the big -- the gigantic breakthrough, of course, we know well now, and NVIDIA is recognized as one of the 3 pillars that ignited the modern AI, the big bang of modern AI.

而且我認為，在這一點上，加速計算已經達到臨界點並且遠遠超出了這一點已成定局。今年支持的開發人員數量——我們支持的開發人員在全球有近 200 萬開發人員，而且它正在呈指數級增長。所以我認為加速計算現在已經很成熟了。 NVIDIA 加速計算已經成熟。這是常識，設計數據中心的人希望在其中加入加速計算。問題是多少錢？您使用多少加速計算？您在管道中的哪一部分執行此操作？當然，巨大的突破，我們現在很清楚，NVIDIA 被公認為點燃現代 AI 的三大支柱之一，即現代 AI 的大爆炸。

And the other 2 pillar, of course, is deep learning algorithm and the abundance of data. And so the 3 -- these 3 ingredients came together and was -- people use NVIDIA accelerated computing largely for training. But over time, we expanded training to have a lot more models. And as I mentioned earlier, the single most important model of machine learning today is the recommender system. It's the most important model because it's the only way that you and I could use the Internet in any reasonable way. It's the only way that you and I could use a shopping website or a video web -- a video app or a music app or book or news or anything. And so it is the engine of the Internet from the consumer's perspective. On the company perspective, it is the engine of commerce. Without the recommender system, there's no way they could possibly make money.

而另外兩個支柱，當然是深度學習算法和豐富的數據。因此，這 3 個要素——這 3 個要素結合在一起——人們將 NVIDIA 加速計算主要用於培訓。但隨著時間的推移，我們擴展了訓練以擁有更多模型。正如我之前提到的，當今最重要的機器學習模型是推薦系統。它是最重要的模型，因為它是你和我以任何合理方式使用互聯網的唯一方式。這是你和我可以使用購物網站或視頻網絡的唯一方式——視頻應用程序或音樂應用程序或書籍或新聞或任何東西。因此，從消費者的角度來看，它是互聯網的引擎。從公司的角度來看，它是商業的引擎。沒有推薦系統，他們就不可能賺錢。

And so their accuracy in predicting user preferences is core to everything they do. You just go up and down the list of every company. And that engine is gigantic. It is just a gigantic engine. And from the data processing part of it, which is the reason why we went and spent 3 years on Spark and RAPIDS, which made Spark possible and all the work that we did on NVLink and all that stuff was really focused on big data analytics. The second is all of the training of the deep learning models and the inference. So the number of applications, the footprint, the accelerated computing has grown tremendously, and its importance has grown tremendously because of the applications are the most important applications of these companies. And so I think when I mentioned -- when I said that, that acceleration is still growing, it is. But the major workloads, the most important workloads of the world's most important companies are now -- solidly require acceleration. And so I'm looking forward to a really exciting ramp for Ampere for all the reasons that I just mentioned.

因此，他們預測用戶偏好的準確性是他們所做一切的核心。你只需在每家公司的名單上上下移動。那個引擎是巨大的。它只是一個巨大的引擎。從它的數據處理部分來看，這就是我們花了 3 年時間研究 Spark 和 RAPIDS 的原因，這使 Spark 成為可能，我們在 NVLink 上所做的所有工作都真正專注於大數據分析。第二個是深度學習模型的所有訓練和推理。因此，應用程序的數量、佔用空間、加速計算已經大幅增長，其重要性也大幅增長，因為應用程序是這些公司最重要的應用程序。所以我認為當我提到 - 當我這麼說時，加速仍在增長，確實如此。但是，世界上最重要的公司的主要工作負載，最重要的工作負載現在 - 確實需要加速。因此，出於我剛才提到的所有原因，我期待著 Ampere 的真正令人興奮的斜坡。

Your next question comes from John Pitzer with Crédit Suisse.

您的下一個問題來自瑞士信貸的 John Pitzer。

Just 2 quick ones. Colette, I hate to ask something as mundane as OpEx, but just given the full year guide, there's sort of a lot to unpack. And you talked about some of it like the raises. I mean I think you also probably have some COVID plus or minuses in that. I think there's an extra week this year as well. And then, of course, there's Mellanox and how you're thinking about investing in that asset.

只有2個快速的。 Colette，我不想問像運營支出這樣平凡的問題，但只要給出全年指南，就有很多東西要解開。你談到了其中的一些，比如加薪。我的意思是，我認為您可能也有一些 COVID 的優點或缺點。我認為今年也有額外的一周。然後，當然還有 Mellanox 以及您如何考慮投資該資產。

I guess I'm just kind of curious, when we look at the full year guide, is there something structural going on OpEx as you try to take advantage of all these opportunities? Or can we use it as sort of a guidepost to how you're thinking about revenue for the back half of the year as well? How do I understand that? And then, Jensen, just a quick one for you. It kind of makes sense to me that COVID is accelerating activity in sort of HPC and hyperscale and maybe even in certain verticals like health care. But in the other verticals, has the sort of shelter in place kind of hurt engagement? And could we actually come out of COVID with some pent-up demand in those vertical markets?

我想我只是有點好奇，當我們查看全年指南時，當您試圖利用所有這些機會時，OpEx 是否存在結構性問題？或者我們可以將其用作您如何考慮下半年收入的指南嗎？我怎麼理解？然後，Jensen，給你一個快速的。對我來說，COVID 正在加速某種 HPC 和超大規模的活動，甚至可能在醫療保健等某些垂直領域加速活動，這在我看來是有道理的。但在其他垂直領域，這種避難所是否會損害參與度？在這些垂直市場中，我們真的可以擺脫 COVID 帶來一些被壓抑的需求嗎？

Okay. Thanks, John, for the question. Let's start from the first perspective on the overall OpEx for the year. We've guided the non-GAAP at approximately $4.1 billion for the year. Yes, that incorporates 3 full quarters of Mellanox, Mellanox and its employees. We have about close to 3,000 Mellanox employees coming on board. You are correct. We have a 53rd week in this quarter -- excuse me, not this quarter, this year.

好的。謝謝，約翰，這個問題。讓我們從今年整體運營支出的第一個角度開始。我們將非 GAAP 指導為全年約 41 億美元。是的，這包括 Mellanox、Mellanox 及其員工的 3 個完整季度。我們有近 3,000 名 Mellanox 員工加入。你是對的。本季度我們有第 53 週——對不起，不是本季度，而是今年。

And that is -- has been outlined in SEC filings, and you should expect that as well. We pulled forward a little bit our focal by several months in order to take care of our employees. And then lastly though, we are investing in our business. We see some great opportunities. You've seen some good results from our investment, and there's more to do. We are hiring and investing in those businesses. So there's nothing different structurally, but just this onset of Mellanox and are investing together, I think, we'll produce long-term great results.

那就是 - 已在美國證券交易委員會的文件中概述，您也應該期待這一點。為了照顧我們的員工，我們將我們的焦點提前了幾個月。最後，我們正在投資我們的業務。我們看到了一些很好的機會。您已經從我們的投資中看到了一些好的結果，還有更多工作要做。我們正在招聘和投資這些業務。因此，在結構上沒有什麼不同，但只是 Mellanox 的開始和正在共同投資，我認為，我們將產生長期的良好結果。

And as usual, John, you know that we're investing into the IT industry's largest opportunities, cloud computing and AI. And then after these 2 opportunities is edge AI. And so we're looking down the fairway with some pretty extraordinary opportunities. But as usual, we're thoughtful about the rate of investment, and we're well-managed. And NVIDIA's leadership team are excellent managers, and you could count on us to continue to do that. Simona, what was John's question? Could you just give me one hint? I haven't...

和往常一樣，約翰，你知道我們正在投資 IT 行業最大的機會，即云計算和人工智能。然後在這兩個機會之後是邊緣人工智能。因此，我們正在尋找一些非常非凡的機會。但像往常一樣，我們對投資率深思熟慮，而且我們管理得很好。 NVIDIA 的領導團隊是優秀的管理者，您可以指望我們繼續這樣做。西蒙娜，約翰的問題是什麼？你能給我一個提示嗎？我沒有...

Just the idea of engagement levels in verticals, just with shelter in place. Has that hampered...

只是垂直的參與水平的想法，只是在適當的地方。有沒有阻礙...

Oh yes. Right. Yes, right. Yes, right. A few -- some of the industries have been affected. We already mentioned automotive industry. The automotive industry has been grounded to a halt. Manufacturing hasn't largely stopped. And you saw that in our guidance. We expect automotive to be down 40% quarter-to-quarter. It's not going to remain that way. It's going to come back. And nobody knows what level is going to come back to you and how long, but it's going to come back. And there's no question in my mind that the automotive industry, they're hunkered down right now, but they will absolutely invest in the future of autonomous vehicles.

哦是的。正確的。是的，沒錯。是的，沒錯。少數——一些行業受到了影響。我們已經提到了汽車行業。汽車行業已經陷入停頓。製造業並沒有在很大程度上停止。你在我們的指導中看到了這一點。我們預計汽車行業將環比下降 40%。它不會保持這種狀態。它會回來的。沒有人知道什麼水平會回到你身邊，多久會回來，但它會回來的。在我看來，汽車行業毫無疑問，他們現在處於低迷狀態，但他們絕對會投資於自動駕駛汽車的未來。

They have to, or they'll be extinct. It's not possible not to have autonomous capability in the future of everything that moves. Not so that it could just completely drive without you. That a nice benefit, too. But mostly because of safety and comfort and just a joy of what seems like the car is reading your mind. And of course, you're still responsible for driving it and -- but it just seems to be coasting down the road, reading your mind and helping you. And so I think the future of autonomous vehicles is a certainty. People recognize the incredible economics that the pioneer, Tesla, is enjoying. And the industry is going to go after it.

他們必須這樣做，否則他們將滅絕。未來所有移動的事物都不可能不具備自主能力。不是這樣它就可以在沒有你的情況下完全驅動。這也是一個不錯的好處。但主要是因為安全性和舒適性，而且只是因為汽車似乎在讀你的想法而感到高興。當然，你仍然有責任駕駛它 - 但它似乎只是在路上滑行，讀懂你的想法並幫助你。所以我認為自動駕駛汽車的未來是確定無疑的。人們認識到先驅特斯拉正在享受令人難以置信的經濟。該行業將追隨它。

The future car companies are going to be software-defined companies and technology companies. And they would love to have an economic that allows them to enjoy the installed base of their fleets. And so they're going to go after it. And so this is -- I'm certain that this is going to come back. And well, I have every confidence is going to come back. And let's see, the energy sectors are -- have been impacted. The retail sector has been impacted. There's -- those aren't large industries for us. The impact in some of the industries is accelerating their focus in robotics. Like, for example, on the one hand, BMW has obviously impacted in manufacturing, which is the reason why they're moving so rapidly towards robotics. They have to figure out a way to get robotics into their factories. So same thing with retail. You're going to see a lot more robotic support in retail, you're going to see a lot more robotic support in warehouses, in logistics. And so during this time, when the market -- when the industry is disrupted and impacted, it allows the market leaders to really lean into investing into the future. And so when they come back, they'll be coming back stronger than ever.

未來的汽車公司將成為軟件定義公司和技術公司。他們希望擁有一種經濟，使他們能夠享受其機隊的安裝基礎。所以他們會去追求它。所以這是 - 我確信這會回來。好吧，我完全有信心會回來。讓我們看看，能源部門 - 已經受到影響。零售業受到影響。有 - 這些對我們來說不是大行業。一些行業的影響正在加速他們對機器人技術的關注。例如，一方面，寶馬顯然對製造業產生了影響，這就是他們如此迅速地轉向機器人技術的原因。他們必須想辦法讓機器人技術進入他們的工廠。零售也是如此。你會在零售中看到更多的機器人支持，你會在倉庫和物流中看到更多的機器人支持。因此，在這段時間裡，當市場——當行業受到干擾和影響時，它讓市場領導者真正傾向於投資於未來。所以當他們回來時，他們會比以往任何時候都更強大。

And your next question comes from Matt Ramsay with Cowen.

您的下一個問題來自 Matt Ramsay 和 Cowen。

Two different topics, Jensen. Well, first of all, congrats on Ampere. It's a heck of a product. The first question...

兩個不同的話題，詹森。好吧，首先，恭喜安培。這真是一個產品。第一個問題...

Thank you, Matt. I'm so proud.

謝謝你，馬特。我很自豪。

The first question is it might have been a little bit hard to talk when the deal was pending about this topic, but now that it's closed, maybe you could talk a little bit about opportunities to innovate on and customize the Mellanox stack and the balance of having an industry standard. And the second one is E3 canceled, Computex moved around. At the same time, there's obviously stay-at-home gaming demand. Just how you think about gaming product, launch logistics? And any comments on there would be really helpful.

第一個問題是，在交易未決時可能有點難以談論這個話題，但現在它已經結束，也許你可以談談創新和定制 Mellanox 堆棧的機會以及平衡具有行業標準。而第二個是E3取消，Computex搬來搬去。與此同時，顯然還有待在家裡的遊戲需求。您如何看待遊戲產品、發布物流？任何關於那裡的評論都會非常有幫助。

Yes. Thanks a lot, Matt. I appreciate your questions. I'll go backwards because it's kind of cool. On the one hand, I do miss that we can't engage the developers face to face. It's just so much fun. GTC is doing all their work and the hundreds of papers that are presented, I learned so much each time. And frankly, I really enjoyed the analyst meeting that we have. And so there's all kinds of stuff that I missed about the physical GTC, but here's the amazing thing.

是的。非常感謝，馬特。我很欣賞你的問題。我會倒退，因為它有點酷。一方面，我確實懷念我們無法與開發人員面對面交流。這真是太有趣了。 GTC 正在做他們所有的工作和提交的數百篇論文，我每次都學到了很多。坦率地說，我真的很喜歡我們舉行的分析師會議。所以我錯過了關於物理 GTC 的各種東西，但這是令人驚奇的事情。

We had almost -- the GTC kitchen keynote. I did it from my kitchen just right behind me, and the kitchen keynote has been viewed almost 4 million times. And the video is incredible. And so I think our reach is -- could be quite great. And so I'm not too -- we've got an amazing marketing team, and we just -- we've got great people. They're going to find a way to reach our gamers. And whenever we launch something next, you know that gamers are going to be and our customers are going to be -- our end markets are going to be really excited to see it. And so I'm very confident that we're going to do just fine. Matt, what was the question before? I should never do backwards.

我們幾乎有——GTC 廚房主題演講。我是在我身後的廚房裡做的，廚房的主題演講已經被瀏覽了近 400 萬次。視頻令人難以置信。所以我認為我們的影響力可能非常大。所以我也不是——我們有一個了不起的營銷團隊，我們只是——我們有很棒的人。他們會想辦法接觸我們的遊戲玩家。每當我們接下來推出一些東西時，你知道遊戲玩家將會成為我們的客戶，我們的終端市場將會非常興奮地看到它。所以我非常有信心我們會做得很好。馬特，之前的問題是什麼？我永遠不應該倒退。

Just the industry standard versus customization of Mellanox opportunity.

只是行業標準與 Mellanox 機會的定制。

I see. Okay. Yes. There's -- we work so closely with Mellanox over the years. And on the day that we announced GTC, you could see the number of products that we have working together. The product synergies are really incredible, and the product synergies include a lot of software development that went in and a lot of architectural development that went in. DGX comes with 9 Mellanox, Matt, as I mentioned. If you look at our data center, we ship -- before we ship DGXs to the customers, we ship it to our own engineers. And the reason for that is because every single product in our company has AI in it.

我知道了。好的。是的。多年來，我們與 Mellanox 密切合作。在我們宣布 GTC 的那天，您可以看到我們合作的產品數量。產品協同效應確實令人難以置信，產品協同效應包括大量的軟件開發和大量的架構開發。正如我所提到的，DGX 附帶 9 Mellanox，Matt。如果您查看我們的數據中心，我們會發貨——在將 DGX 發貨給客戶之前，我們會將其發貨給我們自己的工程師。原因是我們公司的每一個產品都包含人工智能。

From Jarvis to Metropolis to Merlin to DRIVE to Clara to Isaac to -- right? All of our products has AI in it, and we're accelerating frameworks for all of the AI industry. And Ampere comes with a brand-new numerical format called Tensor Float 32. And TF32 is just a fantastic [medium medical] format and the performance is incredible. And we had to get it integrated in with the industry standard frameworks. And now Tensor Float comes standard with Tensor Flow -- with TF -- NVIDIA TF32, and PyTorch come standard with TF32. And so we need our own large-scale data center. And so the first customer we ship to was ourselves. And then we started shipping as quickly as we could to all of the customers. You saw that in our data center, in our supercomputer.

從 Jarvis 到 Metropolis 到 Merlin 到 DRIVE 到 Clara 到 Isaac 到——對吧？我們所有的產品都包含人工智能，我們正在加速所有人工智能行業的框架。並且 Ampere 帶有一種全新的數字格式，稱為 Tensor Float 32。而 TF32 只是一種出色的 [medium Medical] 格式，性能令人難以置信。我們必須將其與行業標準框架集成。現在，Tensor Float 標配了 Tensor Flow——帶有 TF——NVIDIA TF32，而 PyTorch 標配了 TF32。所以我們需要我們自己的大型數據中心。所以我們發貨的第一個客戶就是我們自己。然後我們開始盡快向所有客戶發貨。你在我們的數據中心，在我們的超級計算機中看到了這一點。

We have 170 -- 170 state-of-the-art, brand-new Mellanox switches. And almost 1,500 -- 200 gigabit per second Mellanox mix. And 15 kilometers of cables, fiber optic cables. And that is one of the most powerful supercomputers in the world today, and it's based on Ampere. And so we have a great deal of work that we did there together. We announced our first edge computer between us and Mellanox in this new card, we call it the EGX A100. It integrates Ampere and it integrates Mellanox' CX-6 Dx, which is designed for 5G telcos and edge computing. It's incredible security and has a single route of trust, and it's virtualized. And so basically, we -- this EGX A100, when you put it into a standard center x86 server, turns that server into a cloud computer in a box.

我們有 170 到 170 個最先進的全新 Mellanox 開關。幾乎每秒 1,500 - 200 Gb 的 Mellanox 混合。以及15公里的光纜、光纜。這是當今世界上最強大的超級計算機之一，它基於安培。所以我們在那裡一起做了很多工作。我們在這張新卡中宣布了我們和 Mellanox 之間的第一台邊緣計算機，我們稱之為 EGX A100。它集成了 Ampere，並集成了 Mellanox 的 CX-6 Dx，後者專為 5G 電信公司和邊緣計算而設計。它具有令人難以置信的安全性，具有單一的信任路徑，並且是虛擬化的。所以基本上，我們——這個EGX A100，當你把它放到一個標準的中心x86服務器中時，把那個服務器變成一個盒子裡的雲計算機。

The entire capability of a cloud, of a state-of-the-art cloud, which is cloud native, it's secure, it has incredible AI processing, it's now completely hyperconverged inside 1 box. The technology that made EGX A100 is really quite remarkable. And so you could see all the different product synergies that we have in working together. We could have done Spark acceleration without the collaboration with Mellanox. They worked on this piece of networking software called UCX. We worked on [nickel] together. It made possible the infrastructure for large-scale distributor computing. I mean just the list goes on and on and on. And so we -- the 2 teams have great chemistry. The culture -- it's a great culture fit. I love working with them. And right out of the chute, you saw all of the great product synergies that are made possible because of the combination.

雲的全部功能，最先進的雲，它是雲原生的，它是安全的，它具有令人難以置信的 AI 處理，它現在完全超融合在一個盒子內。製造 EGX A100 的技術確實非常了不起。因此，您可以看到我們在合作過程中的所有不同產品協同效應。如果沒有與 Mellanox 合作，我們本可以完成 Spark 加速。他們開發了一款名為 UCX 的網絡軟件。我們一起研究[鎳]。它使大規模分發計算的基礎設施成為可能。我的意思是，名單還在繼續。所以我們——兩支球隊有很好的化學反應。文化——這是一種很好的文化契合。我喜歡和他們一起工作。走出滑道，您就看到了由於這種組合而成為可能的所有偉大的產品協同作用。

That is all the time we have for questions. I'll turn the call back to Jensen Huang for closing remarks.

這就是我們提出問題的所有時間。我會將電話轉回給 Jensen Huang 進行結束致辭。

It's coming. Let me just -- thank you. We had a great and busy quarter. With our announcements, we highlighted several initiatives. First, computing is moving to data center scale where computing and networking go hand in hand. The acquisition of Mellanox gives us deep expertise and scale to innovate from end to end. Second, AI is the most powerful technology force of our time. Our Ampere generation offers several breakthroughs. It is the largest ever generational leap 20x in training and inference throughput; the first unified acceleration platform for data analytics, machine learning, deep learning, training and inference; and the first elastic accelerator that can be configured for scale-up applications like training to scale-out applications like inference.

它來了。讓我 - 謝謝。我們度過了一個美好而忙碌的季度。在我們的公告中，我們強調了幾項舉措。首先，計算正在向數據中心規模發展，計算和網絡齊頭並進。收購 Mellanox 為我們提供了深厚的專業知識和規模以進行端到端的創新。其次，人工智能是我們這個時代最強大的技術力量。我們的安培一代提供了多項突破。這是訓練和推理吞吐量的 20 倍最大的代際飛躍；第一個用於數據分析、機器學習、深度學習、訓練和推理的統一加速平台；以及第一個彈性加速器，可以配置用於擴展應用程序（如訓練）到擴展應用程序（如推理）。

Ampere is fast, it's universal and it's elastic. It's going to re-architect the modern data center. Third, we are opening large new markets with AI software application framework, such as Clara for health care, DRIVE for autonomous vehicles, Isaac for robotics, Jarvis for conversational AI, Metropolis for edge IoT, AERIAL for 5G and Merlin with the very important recommender systems. And then finally, we have built up multiple engines of accelerated computing growth. RTX computer graphics, artificial intelligence, and data center scale computing from cloud to edge. I look forward to updating you on our progress next quarter. Thanks, everybody.

安培速度快，通用且有彈性。它將重新構建現代數據中心。第三，我們正在用人工智能軟件應用框架打開巨大的新市場，例如用於醫療保健的 Clara、用於自動駕駛汽車的 DRIVE、用於機器人技術的 Isaac、用於對話式人工智能的 Jarvis、用於邊緣物聯網的 Metropolis、用於 5G 的 AERIAL 和非常重要的推薦人 Merlin系統。最後，我們建立了多個加速計算增長的引擎。從雲到邊緣的 RTX 計算機圖形、人工智能和數據中心規模計算。我期待著在下個季度向您通報我們的進展情況。謝謝大家。

This concludes today's conference call. You may now disconnect.

今天的電話會議到此結束。您現在可以斷開連接。