輝達 (NVDA) 2019 Q1 法說會逐字稿

Good afternoon, my name is Kelsey, and I am your conference operator for today. Welcome to NVIDIA's financial results conference call. (Operator Instructions)

下午好，我叫凱爾西，我是你們今天的會議接線生。歡迎參加英偉達財務業績電話會議。（操作說明）

I'll now turn the call over to Simona Jankowski, Vice President of Investor Relations, to begin your conference.

現在我將把電話交給投資者關係副總裁西蒙娜·揚科夫斯基，由她來開始你們的會議。

Thank you. Good afternoon, everyone, and welcome to NVIDIA's Conference Call for the First Quarter of Fiscal 2019. 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.

謝謝。各位下午好，歡迎參加英偉達2019財年第一季業績電話會議。今天與我一起參加電話會議的英偉達總裁兼執行長黃仁勳，以及執行副總裁兼財務長科萊特·克雷斯。

I'd like to remind you that our call is being webcast live on NVIDIA's Investor Relations website. It's also being recorded. You can hear a replay by telephone until May 16, 2018. The webcast will be available for replay until the conference call to discuss our financial results for the second quarter of fiscal 2019.

我想提醒各位，我們的電話會議正在英偉達投資者關係網站上進行網路直播。整個過程都被錄影了。您可以在 2018 年 5 月 16 日之前透過電話收聽重播。網路直播將提供回放，直至召開電話會議討論我們 2019 財年第二季度的財務業績。

The content of today's call is NVIDIA's property. It can't be reproduced or transcribed without our prior written consent. 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 differ 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 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 10, 2018, based on information currently available to us. Except as required by law, we assume no obligation to update any such statements.

本次電話會議的內容歸英偉達所有。未經我們事先書面同意，不得複製或轉錄。在本次電話會議中，我們可能會根據目前的預期發表一些前瞻性聲明。這些都存在著許多重大風險和不確定性，我們的實際結果可能與預期有重大差異。有關可能影響我們未來財務表現和業務的因素的討論，請參閱今天發布的收益報告中的披露資訊、我們最新的 10-K 表格以及我們可能向美國證券交易委員會提交的 8-K 表格報告。我們所有聲明均截至2018年5月10日，並基於我們目前掌握的資訊作出。除法律另有規定外，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。您可以在我們的財務長評論中找到這些非GAAP財務指標與GAAP財務指標的調節表，該評論已發佈在我們的網站上。

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

那麼，現在讓我把電話交給科萊特。

Thanks, Simona. We had an excellent quarter, with growth across all our platforms, led by gaming and data center. Q1 revenue reached a record $3.21 billion, up 66% year-on-year, up 10% sequentially, and above our outlook of $2.9 billion. Once again, all measures of profitability set records, with GAAP gross margins at 64.5%, operating margins at 40.4%, and net income at $1.24 billion. From a reporting segment perspective, Q1 GPU revenue grew 77% from last year, to $2.77 billion. Tegra Processor revenue rose 33% to $442 million.

謝謝你，西蒙娜。本季我們表現出色，所有平台均實現成長，其中以遊戲和資料中心業務表現最為強勁。第一季營收達到創紀錄的 32.1 億美元，年增 66%，季增 10%，高於我們先前 29 億美元的預期。各項獲利指標再次創下紀錄，GAAP毛利率為64.5%，營業利益率為40.4%，淨收入為12.4億美元。從報告分部來看，第一季 GPU 營收比去年同期成長 77%，達到 27.7 億美元。Tegra處理器營收成長33%，達到4.42億美元。

Let's start with our gaming business. Revenue was $1.72 billion, up 68% year-on-year and down 1% sequentially. Demand was strong and broad-based across regions and products. The gaming market remains robust and the popular Battle Royale genre is attracting a new wave of gamers to the GeForce platform. We also continued to see demand from upgrades, with about 35% of our installed base currently on our Pascal architecture. The launch of popular titles, like Far Cry 5 and FANTASY -- FINAL FANTASY XV, continued to drive excitement in the quarter. Gamers are increasingly engaging in social gameplay and gaming is rapidly becoming a spectator sport, while the production value of games continues to increase. This dynamic is fueling a virtuous cycle that expands the universe of gamers and drives a mix shift to higher end GPUs.

讓我們先從遊戲業務說起。營收為 17.2 億美元，年增 68%，季減 1%。各地區和各產品的需求都很強勁且廣泛。遊戲市場依然強勁，而受歡迎的「大逃殺」遊戲類型正在吸引新一波玩家加入 GeForce 平台。我們也持續看到升級需求，目前我們約有 35% 的已安裝設備採用 Pascal 架構。《孤島驚魂5》和《最終幻想15》等熱門遊戲的推出，繼續推動了本季的市場熱情。玩家們越來越熱衷於社群遊戲，遊戲也迅速成為一項觀賞性運動，同時遊戲的製作水準也不斷提升。這種動態正在推動良性循環，擴大遊戲玩家群體，並推動遊戲產品組合轉向高階GPU轉變。

At the recent Game Developers Conference, we announced our real-time ray-tracing technology, NVIDIA RTX. Ray tracing is movie-quality rendering technique that delivers lifelike lighting, reflections and shadows. It has long been considered the Holy Grail of graphics and we've been working on it for over 10 years. We look forward to seeing amazing, cinematic games that take advantage of this technology come to the market later this year, with the pipeline building into next year and beyond. And we expect RTX, as well as other new technologies like 4K and virtual reality, to continue driving gamers' requirements for higher GPU performance.

在最近的遊戲開發者大會上，我們發布了我們的即時光線追蹤技術 NVIDIA RTX。光線追蹤是一種電影級渲染技術，能夠呈現逼真的光照、反射和陰影效果。它一直被認為是圖形領域的聖杯，我們已經為此努力了 10 多年。我們期待今年稍晚能看到利用這項技術的精彩電影級遊戲上市，並期待明年及以後有更多這類遊戲問世。我們預計 RTX 以及 4K 和虛擬實境等其他新技術將繼續推動遊戲玩家對更高 GPU 效能的需求。

While supply was tight earlier in the quarter, the situation is now easing. As a result, we are pleased to see that channel prices for our GPUs are beginning to normalize, allowing gamers who had been priced out of the market last quarter to get their hands on the new GeForce GTX at a reasonable price. Cryptocurrency demand was again stronger than expected, but we were able to fulfill most of it with crypto-specific GPUs, which are included in our OEM business, at $289 million. As a result, we could protect the vast majority of our limited gaming GPU supply for use by gamers. Looking into Q2, we expect crypto-specific revenue to be about 1/3 of its Q1 level.

雖然本季初供應緊張，但目前情況正在緩解。因此，我們很高興地看到，我們 GPU 的通路價格開始趨於正常化，使得上個季度因價格過高而無法購買顯示卡的玩家能夠以合理的價格入手新款 GeForce GTX。加密貨幣需求再次超出預期，但我們能夠透過加密貨幣專用 GPU 來滿足大部分需求，這些 GPU 包含在我們的 OEM 業務中，價值 2.89 億美元。因此，我們可以保護絕大部分有限的遊戲GPU供應，供遊戲玩家使用。展望第二季度，我們預期加密貨幣相關收入約為第一季的三分之一。

Gaming notebooks also grew well, driven by an increasing number of thin and light notebooks based on our Max-Q design. And Nintendo Switch contributed strongly to year-on-year growth, reflecting that platform's continued success.

遊戲筆記型電腦也發展良好，這得益於越來越多基於我們 Max-Q 設計的輕薄筆記型電腦的推出。Nintendo Switch 對同比增速做出了巨大貢獻，反映了該平台的持續成功。

Moving to data center. We had another phenomenal quarter, with revenue of $701 million, up 71% year-on-year, up 16% sequentially. Demand was strong in all market segments, and customers increasingly embraced our GPUs and CUDA platform for high-performance computing and AI. Adoption of our Volta architecture remained strong across a wide range of verticals and customers. In the public cloud segment, Microsoft Azure announced general availability of Tesla V100 instances, joining Amazon, IBM and Oracle. And Google Cloud announced that the V100 is now publicly available in beta. Many other hyperscale and consumer Internet companies also continued their ramp of Volta, which delivers 5x the deep learning performance of its predecessor, Pascal. Volta has been chosen by every major cloud provider and server maker, reinforcing our leadership in AI deep learning.

遷移至資料中心。我們又迎來了一個業績斐然的季度，營收達到 7.01 億美元，年增 71%，季增 16%。各個細分市場的需求都很強勁，客戶越來越多地採用我們的 GPU 和 CUDA 平台進行高效能運算和人工智慧應用。我們的 Volta 架構在各個垂直行業和客戶群中都得到了廣泛認可。在公有雲領域，微軟 Azure 宣布正式推出 Tesla V100 實例，加入亞馬遜、IBM 和 Oracle 的行列。谷歌雲端宣布 V100 現在已公開提供測試版。許多其他超大規模和消費網路公司也持續推動 Volta 的部署，Volta 的深度學習效能是其前身 Pascal 的 5 倍。Volta 已被所有主要雲端服務供應商和伺服器製造商選中，鞏固了我們在人工智慧深度學習領域的領先地位。

In high-performance computing, strength from the broad enterprise vertical more than offset the ramp down of major supercomputing projects such as the U.S. Department of Energy's Summit System. We see a strong pipeline across a number of vertical industries from manufacturing to oil and gas, which has helped sustain the trajectory of high-performance computing next quarter and beyond.

在高效能運算領域，來自廣泛的企業垂直領域的強勁成長足以彌補美國能源部 Summit 系統等大型超級運算專案的逐步減少。我們看到從製造業到石油和天然氣等多個垂直行業都有強勁的需求，這有助於在下一季及以後保持高效能運算的發展動能。

Traction is also increasing in AI inference. Inference GPU shipments to cloud service providers more than doubled from last quarter. And our pipeline is growing into next quarter. We dramatically increased our inference capabilities with the announcement of the TensorRT 4 AI Inference Accelerator Software at our recent GPU Technology Conference in San Jose. TensorRT 4 accelerates deep learning inference up to 190 times faster than CPUs for common applications such as computer vision, neural machine translation, automatic speech recognition, speech synthesis and recommendation systems. It also dramatically expands the use cases prepared with the prior version. With TensorRT 4, NVIDIA's market reach has expanded to approximately 30 million hyperscale servers worldwide.

人工智慧推理領域的應用也日益增多。用於推理的 GPU 向雲端服務供應商的出貨量比上一季翻了一番還多。我們的業務拓展計劃將持續到下一季。我們在最近於聖荷西舉行的 GPU 技術大會上發布了 TensorRT 4 AI 推理加速器軟體，從而大幅提升了我們的推理能力。TensorRT 4 可將深度學習推理速度提升至 CPU 的 190 倍，適用於電腦視覺、神經機器翻譯、自動語音辨識、語音合成和推薦系統等常見應用。它還極大地擴展了先前版本所準備的應用場景。透過 TensorRT 4，NVIDIA 的市場覆蓋範圍已擴展到全球約 3,000 萬台超大規模伺服器。

At GTC, we also announced other major advancements in our deep learning platform. We doubled the memory of Tesla V100 to 32 GB DRAM, which is a key enabler for customers building large neural networks through larger data sets. And we announced a new GPU interconnect fabric called NVIDIA NVSwitch. (inaudible) 16 Pascal V100 GPUs at a speed of 2.4 terabytes per second, or 5x faster than the best PCIe switch. We also announced our DGX-2 system, which leverages these new technologies and its updated, fully-optimized software stack to deliver a 10x performance boost beyond last year's DGX. DGX-2 is the first single server capable of delivering 2 petaflops of computational power. We are seeing strong interest from both hyperscale and (inaudible) customers, and we look forward to bringing this technology to cloud customers later this year.

在 GTC 大會上，我們也宣布了深度學習平台的其他重大進展。我們將 Tesla V100 的記憶體容量增加了一倍，達到 32 GB DRAM，這對於客戶透過更大的資料集建立大型神經網路來說至關重要。我們也發布了名為 NVIDIA NVSwitch 的全新 GPU 互連架構。（聽不清楚）16 個 Pascal V100 GPU，速度為每秒 2.4 TB，比最好的 PCIe 交換器快 5 倍。我們還發布了 DGX-2 系統，該系統利用了這些新技術及其更新、完全優化的軟體堆疊，與去年的 DGX 相比，性能提高了 10 倍。DGX-2 是首台能夠提供 2 petaflops 運算能力的單一伺服器。我們看到來自超大規模客戶和（聽不清楚）客戶的強烈興趣，我們期待在今年稍後將這項技術帶給雲端客戶。

At our Investor Day in March, we updated our forecast for the data center and the rest of the market. We see the data center opportunity as very large, fueled by growing demand for accelerated computing and applications ranging from AI (inaudible) multiple market segments and vertical industries. We estimate the TAM at $50 billion by 2023, which extends our previous forecast of $30 billion by 2020. We see strong momentum in the adoption of our accelerated computing platform and the expansion of our development ecosystem to serve this rapidly growing market. About 8,500 attendees registered for GTC, up 18% from last year. CUDA downloads have continued to grow, setting a fresh record in the quarter. And our total number of developers is well over 850,000, up 72% from last year.

在三月的投資者日上，我們更新了對資料中心和整個市場的預測。我們認為資料中心的機會非常巨大，這得益於對加速運算和人工智慧等多個市場領域及垂直行業應用的日益增長的需求。我們預計到 2023 年，TAM 將達到 500 億美元，這比我們先前預測的到 2020 年達到 300 億美元有所擴大。我們看到，我們的加速運算平台得到了廣泛採用，我們的開發生態系統也在不斷擴展，以服務這個快速成長的市場。約有 8500 名與會者註冊參加 GTC，比去年增長了 18%。CUDA 下載量持續成長，本季創下新紀錄。我們的開發人員總數已超過 85 萬，比去年增加了 72%。

Moving to pro visualization. Revenue grew to $251 million, up 22% from a year ago and accelerating from last quarter, driven by demand for real-time rendering as well as emerging applications like AI and VR. Strength extended across several key industries, including public sector, health care and retail. Key wins in the quarter included Columbia University, using high-end Quadro GPUs for AI, and Siemens, using them for CT and ultrasound solutions.

邁向專業可視化。營收成長至 2.51 億美元，比去年同期成長 22%，比上一季增速加快，這主要得益於對即時渲染的需求以及人工智慧和虛擬實境等新興應用的發展。強勁勢頭遍及多個關鍵行業，包括公共部門、醫療保健和零售業。本季的主要成功案例包括哥倫比亞大學（使用高端 Quadro GPU 進行人工智慧開發）和西門子（使用它們進行 CT 和超音波解決方案開發）。

At GTC, we announced the Quadro GV100 GPU with NVIDIA RTX technology, capable of delivering real-time ray tracing to the more than 25 million artists and designers throughout the world. RTX makes computational intensive ray tracing possible in real time when running professional design and content creation applications. This allows media and entertainment professionals to see and interact with their creations with correct light and shadows and do complex renders up to 10x faster than a GPU -- a CPU alone. And the NVIDIA OptiX AI denoiser built into RTX delivers almost 100x the performance of CPUs for real-time noise-free rendering. This enabled customers to replace racks of servers in traditional render farms with GPU servers at 1/5 the cost, 1/7 the space, and 1/7 the power.

在 GTC 大會上，我們發布了採用 NVIDIA RTX 技術的 Quadro GV100 GPU，能夠為全球超過 2,500 萬名藝術家和設計師提供即時光線追蹤功能。RTX 技術使得在運行專業設計和內容創作應用程式時，能夠即時進行運算密集型的光線追蹤。這使得媒體和娛樂專業人士能夠以正確的光影效果查看和互動他們的作品，並能以比 GPU 快 10 倍的速度（僅使用 CPU）進行複雜的渲染。RTX 內建的 NVIDIA OptiX AI 降噪器可提供比 CPU 近 100 倍的即時無噪渲染效能。這使得客戶能夠以 1/5 的成本、1/7 的空間和 1/7 的功耗，用 GPU 伺服器替換傳統渲染農場中的伺服器機架。

Lastly, automotive. Revenue grew 4% year-on-year to a record $145 million. This reflects the ongoing transition from our infotainment business to our growing autonomous vehicle development and production opportunities around the globe.

最後，是汽車業。營收年增 4%，達到創紀錄的 1.45 億美元。這反映了我們正在從資訊娛樂業務向全球範圍內不斷增長的自動駕駛汽車開發和生產機會轉型。

At GTC and Investor Day, we made key product announcements on the advancement of autonomous vehicles and established a total addressable market opportunity of 60 billion by 2035. We believe that every vehicle will be autonomous one day. By 2035, this will encompass 100 million autonomous passenger vehicles and 10 million robo-taxis.

在 GTC 和投資者日上，我們發布了有關自動駕駛汽車發展的重要產品公告，並確定到 2035 年，總潛在市場機會將達到 600 億美元。我們相信，總有一天所有車輛都會實現自動駕駛。到 2035 年，這將包括 1 億輛自動駕駛乘用車和 1,000 萬輛無人駕駛計程車。

We also introduced NVIDIA DRIVE Constellation, a platform that will help car companies, carmakers, tier 1 suppliers, and others developing autonomous vehicles test and validate their systems in a virtual world across a wide range of scenarios before deploying on the road. Each year, 10 trillion miles are driven around the world. Even if test cars can eventually cover millions of miles, that's an insignificant fraction of all the scenarios that require testing to create a safe and reliable autonomous vehicle. DRIVE Constellation addresses this challenge by (inaudible) cars to safely drive billions of miles in virtual reality.

我們還推出了 NVIDIA DRIVE Con​​stellation 平台，將幫助汽車公司、汽車製造商、一級供應商以及其他開發自動駕駛汽車的公司在虛擬世界中測試和驗證其係統，涵蓋各種場景，然後再部署到道路上。每年全球車輛行駛里程達10兆英哩。即使測試車輛最終能夠行駛數百萬英里，但這與創造安全可靠的自動駕駛汽車所需的所有測試場景相比，也只是微不足道的一部分。DRIVE Con​​stellation 透過（聽不清楚）汽車在虛擬實境中安全行駛數十億英里來應對這一挑戰。

The platform has 2 different servers. The first is loaded with GPUs and simulates the environment that the car is driving in, as in a hyper-real video game. The second contains the NVIDIA DRIVE Pegasus Autonomous Vehicle Computer, which possesses the simulated data, as if it were coming from the sensors of a car driving on the road. Real-time driving command from the DRIVE Pegasus are fed back to the simulation for true hardware-in-the-loop verification. Constellation will enable autonomous vehicle industry for safety test and validate their AI self-driving systems in ways that are not practical or possible with on-road testing.

該平台擁有 2 台不同的伺服器。第一個系統配備了 GPU，可以模擬汽車行駛的環境，就像超現實電玩遊戲一樣。第二個模組包含 NVIDIA DRIVE Pegasus 自動駕駛汽車計算機，其中包含模擬數據，就像來自在道路上行駛的汽車的傳感器一樣。DRIVE Pegasus 的即時駕駛指令會回饋到模擬中，以實現真正的硬體在環驗證。Constellation 將使自動駕駛汽車行業能夠以道路測試不切實際或不可能的方式進行安全測試並驗證其人工智慧自動駕駛系統。

We also extended our product roadmap to include our next-generation DRIVE Autonomous Vehicle Computer. We have created a scalable AI car platform that spans the entire range of autonomous driving, from traffic jams, pilots, to level 5 robo-taxis. More than 370 companies and research institutions are now using NVIDIA's automotive platform. With this growing momentum, we remain excited about the intermediate and long-term opportunities for autonomous driving business.

我們也擴展了產品路線圖，將下一代 DRIVE 自動駕駛汽車電腦納入其中。我們創建了一個可擴展的 AI 汽車平台，涵蓋了自動駕駛的整個範圍，從交通擁堵、試點駕駛到 5 級無人駕駛計程車。目前已有超過 370 家公司和研究機構正在使用英偉達的汽車平台。隨著這一發展勢頭日益強勁，我們對自動駕駛業務的中長期機會仍然充滿信心。

Now moving to the rest of the P&L. Q1 GAAP gross margins were 64.5% and non-GAAP was 64.7%, records that reflect continued growth in our value-added platforms. GAAP operating expenses were $773 million. Non-GAAP operating expenses were $648 million, up 25% year-on-year. We continue to invest in key platforms driving our long-term growth, including gaming, AI and automotive. GAAP net income was a record $1.24 billion and EPS was $1.98, up 145% and 151% respectively from a year earlier. Some of the expenses (inaudible) by a tax rate of 5% compared to our guidance of 12%. Non-GAAP net income was $1.29 billion and EPS was $2.05, both up 141% from a year ago, reflecting the revenue strength as well as gross margins and operating margin expansion on slightly lower tax. Our quarterly cash flow from operations reached record levels at $1.45 billion. Capital expenditures were $118 million.

現在來看損益表的其餘部分。第一季 GAAP 毛利率為 64.5%，非 GAAP 毛利率為 64.7%，這些記錄反映了我們增值平台的持續成長。GAAP營運費用為7.73億美元。非GAAP營運費用為6.48億美元，年增25%。我們將繼續投資於推動我們長期成長的關鍵平台，包括遊戲、人工智慧和汽車。GAAP淨利創歷史新高，達12.4億美元，每股收益為1.98美元，分別比上年同期增加145%和151%。部分費用（聽不清楚）以 5% 的稅率計算，而我們先前的預期稅率為 12%。非GAAP淨利潤為12.9億美元，每股收益為2.05美元，均比上年同期增長141%，這反映了收入的強勁增長，以及在稅收略微降低的情況下毛利率和營業利潤率的擴張。我們的季度經營活動現金流達到創紀錄的14.5億美元。資本支出為1.18億美元。

With that, let me turn to the outlook for the second quarter of fiscal 2019. We expect revenue to be $3.1 billion plus or minus 2%. GAAP and non-GAAP gross margins are expected to be 63.6% and 63.5%, respectively, plus or minus 50 basis points. GAAP and non-GAAP operating expenses are expected to be approximately $810 million and $685 million, respectively. GAAP...

接下來，我將展望一下 2019 財年第二季的情況。我們預計營收為 31 億美元，上下浮動 2%。GAAP 和非 GAAP 毛利率預計分別為 63.6% 和 63.5%，上下浮動 50 個基點。GAAP 和非 GAAP 營運費用預計分別為約 8.1 億美元和 6.85 億美元。GAAP...

(technical difficulty)

（技術難題）

Capital expenditures are expected to be approximately $130 million to $150 million. Further financial details are included in the CFO Commentary and other information available on our IR website.

預計資本支出約 1.3 億美元至 1.5 億美元。更多財務詳情請參閱財務長評論以及我們投資者關係網站上的其他資訊。

In closing, I'd like to highlight a few upcoming events for the financial community. We'll be presenting at the JPMorgan Technology Conference next week on May 15, and at the Bank of America Global Technology Conference on June 5. We will also hold our Annual Meeting of Stockholders online on May 16.

最後，我想重點介紹一下金融界即將發生的一些事件。我們將於下週5月15日在摩根大通科技大會上演講，並於6月5日在美洲銀行全球科技大會上進行演講。此外，我們也將於5月16日在線舉行年度股東大會。

We will now open the call for questions. Simona and I are here in Santa Clara and Jensen is dialing in from the road. Operator, would you please poll for questions? Thank you.

現在開始接受提問。Simona 和我現在在聖克拉拉，Jensen 正在路上透過電話連線。操作員，請問能否進行一次投票，看看大家有什麼問題？謝謝。

(Operator Instructions) Your first question is from Stacy Rasgon with Bernstein Research.

（操作說明）您的第一個問題來自 Bernstein Research 的 Stacy Rasgon。

First, I had a question on gaming seasonality. It's usually down pretty decently in Q1. It was obviously flat this time as you were trying to fill up the channel. Now that's done. I was just wondering on with the supply dynamics -- supply-demand dynamics as well as like any thoughts on crypto might mean for typical -- the seasonality in the Q2 versus what would be typical or what would usually be down -- or usually be up pretty decently? How are you looking at that? And there's a question for Colette.

首先，我有一個關於遊戲季節性的問題。通常情況下，第一季降幅會相當可觀。很明顯，這次河道是平坦的，因為你試圖填滿河道。現在這件事已經辦完了。我只是想了解供應動態——供需動態，以及對加密貨幣的任何想法，這對於典型的情況意味著什麼——第二季度的季節性與典型的情況或通常會下降的情況相比如何——或者通常會上漲相當不錯的情況？你對此有何看法？還有一個問題想問科萊特。

Jensen, why don't you start on the question for Stacy, and I'll follow-up afterwards, after you speak.

Jensen，不如你先回答Stacy的問題，等你回答完後我再跟進。

Okay. Stacy, so let's see. Q1, as you probably know, Fortnite and PUBG are global phenomenons. The success of Fortnite and PUBG are just beyond, beyond comprehension, really. Those 2 games, a combination of Hunger Games and Survivor, has just captured imaginations of gamers all over the world. And we saw the uptick and we saw the demand on our GPUs from all over the world. Surely, there was scarcity as you know. Crypto miners bought a lot of our GPUs during the quarter, and it drove prices up. And I think that a lot of the gamers weren't able to buy into the new GeForces as a result. And so we're starting to see the prices come down. We monitor spot pricing every single day around the world. And the prices are starting to normalize. It's still higher than where they should be. And so obviously, the demand is still quite strong out there. But my sense is that there's a fair amount of pent-up demand still. Fortnite is still growing in popularity. PUBG is doing great. And then, we've had some amazing titles coming out. And so my sense is that the overall gaming market is just really -- is super healthy. And our job is to make sure that we work as hard as we can to get supply out into the marketplace. And hopefully, by doing that, the pricing will normalize and the gamers can buy into their favorite graphics card at a price that we hope they can get it at. And so I think there's a fair -- so I mean, the simple answer to your question is Fortnite and PUBG. And the demand is just really great. They did a great job.

好的。史泰西，那我們來看看。Q1，正如你可能知道的那樣，《要塞英雄》和《絕地求生》是全球現象級遊戲。Fortnite 和 PUBG 的成功簡直令人難以置信。這兩款遊戲，分別是《飢餓遊戲》和《倖存者》的結合體，已經擄獲了全世界遊戲玩家的心。我們看到了需求的成長，也看到了來自世界各地對我們GPU的需求。當然，如你所知，當時物資匱乏。本季加密貨幣礦工大量購買了我們的GPU，並推高了價格。因此，我認為很多遊戲玩家都無法接受新的GeForce顯示卡。因此，我們開始看到價格下降。我們每天都會監測全球現貨價格。價格開始趨於正常化。它仍然高於應有的水平。顯然，目前市場需求依然十分強勁。但我感覺目前仍然存在相當多的潛在需求。Fortnite 的人氣仍在持續成長。PUBG表現非常出色。然後，我們也推出了一些非常棒的遊戲作品。因此，我的感覺是，整個遊戲市場真的非常健康。我們的工作就是盡一切努力確保將產品推向市場。希望透過這種方式，價格能夠趨於正常，讓玩家能夠以我們希望他們能夠負擔的價格買到自己喜歡的顯示卡。所以我覺得這個問題的答案很公平——我的意思是，你問題的簡單答案是 Fortnite 和 PUBG。市場需求確實非常旺盛。他們做得很好。

Your next question is from Joe Moore with Morgan Stanley.

下一個問題來自摩根士丹利的喬·摩爾。

No wonder -- Colette had talked about the inference doubling in sales quarter-over-quarter with cloud. Can you just talk about where you're seeing the early applications for inference? Is that sort of as-a-service business? Or are you looking at internal cloud workloads? And just any color you can give us on where you guys are sitting in the inference space.

這也難怪——科萊特曾談到，借助雲端運算，銷售額有望實現季度環比翻倍。您能否談談目前看到的推理早期應用領域？這是那種「即服務」型業務嗎？還是您指的是內部雲端工作負載？你們可以用任何顏色來表示你們在推理空間中的位置。

Sure. Joe, so as you know, there are 30 million servers around the world. And they were put in place during the time when the world didn't have deep learning. And now with deep learning and with machine learning approaches, the accuracy of prediction, the accuracy of recommendation has jumped so much that just about every Internet service provider in the world that has a lot of different customers and consumers are jumping onto this new software approach. And in order to take this newer network -- and the software that's written by deep learning, these frameworks, are massive software. The way to think about these deep neural nets is, it has millions and millions and millions of parameters in it, and these networks are getting larger every year. And they're enormously complex. And the output of these neural nets had to be optimized for the computing platform that it targets. How you would optimize the neural network for a CPU or a GPU is very, very different. And how you optimize for different neural networks, whether it's image recognition, speech recognition, natural language translation, recommendation systems, all of these networks have different architectures, and the optimizing compiler that's necessary to make the neural network run smoothly and fast is incredibly complex. And so that's why we created TensorRT. That's what TensorRT is. TensorRT is an optimizing graph neural network compiler. And it optimizes for our -- each one of our platforms. And each -- even each one of our platforms has very different architectures. For example, we invented recently -- reinvented the GPU and it's called the Tensor Core GPU, and the first of its kind is called Volta. And so TensorRT 4.0 now supports, in addition to image recognition, all of the different types of neural network models. The answer to your question is internal consumption. Internal consumption is going to be the first users. Video recognition, detecting for inappropriate video, for example, all over the world, making recommendations from the videos that you search or the images that you're uploading, all of these types of applications are going to require an enormous amount of computation.

當然。喬，如你所知，全世界有3000萬台伺服器。而它們是在深度學習尚未出現的時代建立起來的。現在，隨著深度學習和機器學習方法的出現，預測的準確性、建議的準確性都得到了極大的提升，以至於世界上幾乎所有擁有眾多不同客戶和消費者的網路服務供應商都在採用這種新的軟體方法。為了使用這種新型網路——以及深度學習編寫的軟體，這些框架，都是龐大的軟體。理解這些深度神經網路的方式是：它包含數百萬、數百萬、數百萬個參數，而這些網路每年都在變得更大。它們極其複雜。而且，這些神經網路的輸出必須針對其目標運算平台進行最佳化。針對 CPU 和 GPU 最佳化神經網路的方法截然不同。如何針對不同的神經網路進行最佳化，無論是影像辨識、語音辨識、自然語言翻譯或推薦系統，所有這些網路都有不同的架構，而使神經網路流暢快速運作所必需的最佳化編譯器則極為複雜。所以這就是我們創建 TensorRT 的原因。這就是TensorRT。TensorRT 是一個最佳化型圖神經網路編譯器。它針對我們的每個平台進行了最佳化。而且，我們每個平台——甚至是我們每個平台——的架構都截然不同。例如，我們最近發明了——重新發明了 GPU，它被稱為 Tensor Core GPU，而同類產品中的第一個被稱為 Volta。因此，TensorRT 4.0 除了支援影像辨識之外，現在還支援所有不同類型的神經網路模型。你的問題的答案是內需。內部消費將是首批使用者。視頻識別，例如檢測全球範圍內的不當視頻，根據你搜索的視頻或你上傳的圖片進行推薦，所有這些類型的應用都需要大量的計算。

Next question is from Vivek Arya with Bank of America.

下一個問題來自美國銀行的維韋克·阿亞。

Jensen, I have 2 questions about the data center. One from a growth, and the second, from a competition perspective. So from the growth side, you guys are doing about, say, $3 billion or so annualized, but you have outlined a market that could be $50 billion. What needs to happen for the next inflection? Is it something in the market that needs to change? Is it something in the product set that needs to? How do you go and address that $50 billion market, right? Because you're only a few percent penetrated today in that large market. So what needs to change for the next inflection point? And then, on the competition side, as you are looking at that big market, how should we think about competition that is coming from some of your cloud customers, like a Google announcing a TPU 3 or perhaps others looking at other competing technologies? So any color on both sort of how you look at growth and competition would be very helpful.

Jensen，關於資料中心我有兩個問題。一個是從成長的角度，另一個是從競爭的角度。從成長方面來看，你們的年收入大約是 30 億美元左右，但你們勾勒出的市場規模可能達到 500 億美元。下一個轉折點需要具備哪些條件？市場上是否有什麼需要改變的地方？是產品系列中的某個部件需要改進嗎？你該如何開拓這價值 500 億美元的市場？因為在這個龐大的市場中，你目前的滲透率只有幾個百分點。那麼，下一個轉折點需要做出哪些改變呢？那麼，在競爭方面，當您審視這個龐大的市場時，我們應該如何看待來自一些雲端客戶的競爭，例如Google宣布推出 TPU 3，或者其他公司正在研究其他競爭技術？所以，如果您能就您如何看待成長和競爭這兩方面發表一些看法，那就太好了。

Thanks, Vivek. First of all, at its core, this is something we all know now, that CPU scaling has really slowed. And if you think about the several hundred billion dollars worth of computer equipment that's been installed in the cloud, in data centers all over the world, and as these applications for machine learning and high-performance computing approaches come along, the world needs a solution. CPU scaling has slowed. And so here's the approach that we pioneered 1.5 decades ago called GPU computing. And we've been determined to continue to advance it during this time because we saw this day coming and we really believed that it was going to end. I mean, you can't deny physics. And so we find ourselves in a great position today. And as Colette already mentioned, we have something close to 1 million developers on this platform now. It is incredibly fast, speeding up CPUs by 10, 20, 50, 100x, 200x sometimes, depending on the algorithm. It's everywhere. The software ecosystem is just super rich. And as Colette mentioned, that there is already almost 1 billion -- 1 million developers around the world, that's grown 70% year-over-year. And so I think at the core, it's about the fact that the world needs a computing approach going forward. With respect to the -- our ability to address the TAM, there are 3 major segments. There's more than that, but there's 3 major segments. One is, of course, training for deep learning. The other is inferencing, and TRT 4 is intended to do just that, to expand our ability to address all of the different types of algorithms, machine learning algorithms that are now coming -- that are running in the data centers. The third is high-performance computing, and that's molecular dynamics, to medical imaging, to earth sciences, to energy sciences. The type of algorithms that are being run in supercomputers all over the world is expanding. And we're doing more and more of our product designs in virtual reality. We want to simulate our products and simulate its capabilities in simulation in this computer rather than build it in the beginning. And then, the last category would be graphics virtualization. We've taken with GRID and our Quadro virtual workstation and now with Quadro -- with NVIDIA RTX, we turned the data center into a powerful graphic supercomputer. And so these are the various applications and segments of data center that we see. I think, in the case of training, we're limited by the number of deep learning experts in the world. And that's growing very quickly. The frameworks are making it easier. There's a lot more open source and open documentation on sharing of knowledge. And so the number of AI engineers around the world is growing super fast. The second is inference. And I've already talked about that. It's really limited by our optimizing compilers and how we can target these neural network models to run on our processors. And if we could do so, we're going to save our customers enormous amounts of money. We speed up applications, we speed up these neural network models 50x, 100x, 200x over a CPU. And so the more GPUs they buy, the more they're going to save. And high-performance computing, the way to think about that is, I think, at this point, it's very clear that going forward, supercomputers are going to get built with accelerators in them. And because of our long-term dedication to CUDA and our GPUs for acceleration, of all these codes and the nurturing of the ecosystem, I think that we're going to be -- we're going to do super well in the supercomputing world. And so these are the different verticals. With respect to competition, I think it all starts with the core. And the core is that the CPU scaling has slowed. And so the world needs another approach going forward. And surely, because of our focus on it, we find ourselves in a great position. Google announced GPU 3 and it's still behind our Tensor Core GPU. Our Volta is our first generation of a newly reinvented approach of doing GPUs. It's called Tensor Core GPUs. And we're far ahead of the competition. And -- but more than that, more than that, it's programmable. It's not one function. It's programmable. Not only is it faster, it's also more flexible. And as a result of the flexibility, developers could use it in all kinds of applications, whether it's medical imaging or weather simulations or deep learning or computer graphics. And as a result, our GPUs are available in every cloud and every data center, everywhere on the planet. And which developers need, so that accessibility so that they could develop their software. And so I think that on the one hand, it's too simplistic to compare a GPU to just one of the many features that's in our Tensor Core GPU. But even if you did, we're faster. We support more frameworks. We support all neural networks. And as a result, if you look at GitHub, there are some 60,000 different neural network research papers that are posted that runs on NVIDIA GPUs. And it's just a handful for the second alternative. And so it just kind of gives you a sense of the reach and the capabilities of our GPUs.

謝謝你，維韋克。首先，從本質上講，我們現在都知道，CPU 擴展性確實已經放緩了。想想看，價值數千億美元的電腦設備已經安裝在雲端和世界各地的資料中心，隨著機器學習和高效能運算等應用程式的出現，世界需要一個解決方案。CPU擴充速度已放緩。所以，這就是我們15年前首創的GPU計算方法。我們決心在這段時間繼續推進這項工作，因為我們預見到這一天的到來，我們真的相信它終將結束。我的意思是，你不能否認物理學。因此，我們今天處於非常有利的地位。正如科萊特已經提到的，我們這個平台現在已經有近 100 萬名開發者了。它的速度非常快，有時能將 CPU 的速度提高 10 倍、20 倍、50 倍、100 倍甚至 200 倍，具體取決於演算法。它無所不在。軟體生態系非常豐富。正如科萊特所提到的那樣，全球已有近 10 億——100 萬名開發人員，年增 70%。所以我認為，問題的核心在於，世界需要一種計算方法才能向前發展。就我們應對 TAM 的能力而言，主要有 3 個面向。不僅如此，還有三大主要部分。當然，其中之一就是深度學習訓練。另一個方面是推理，而 TRT 4 的目的正是為了實現這一點，擴展我們處理所有不同類型的演算法、機器學習演算法的能力——這些演算法現在正在資料中心運行。第三是高效能運算，其應用範圍涵蓋分子動力學、醫學影像、地球科學和能源科學等領域。世界各地超級電腦運行的演算法類型正在不斷擴展。我們正在越來越多地利用虛擬實境技術進行產品設計。我們希望在這台電腦上模擬我們的產品及其功能，而不是一開始就實際製造出來。最後，最後一類是圖形虛擬化。我們已經利用 GRID 和 Quadro 虛擬工作站，現在又利用 Quadro – NVIDIA RTX，將資料中心變成了一台強大的圖形超級電腦。因此，以上就是我們看到的資料中心的各種應用和組成部分。我認為，就培訓而言，我們受到全球深度學習專家數量的限制。而且這種趨勢發展得非常迅速。這些框架讓一切變得更簡單。關於知識共享，開源和開放文件的數量大大增加。因此，世界各地的人工智慧工程師數量正在快速增長。第二種方法是推理。我已經談過這個問題了。這實際上受限於我們的最佳化編譯器以及我們如何使這些神經網路模型在我們的處理器上運行。如果我們能夠做到這一點，我們將為客戶節省一大筆錢。我們讓應用程式運作速度加快，讓這些神經網路模型的運作速度比 CPU 快 50 倍、100 倍、200 倍。因此，他們購買的GPU越多，節省的費用就越多。至於高效能運算，我認為，目前很明顯，未來的超級電腦將會配備加速器。由於我們長期致力於 CUDA 和 GPU 的加速，以及對所有這些程式碼和生態系統的培育，我認為我們將在超級運算領域取得巨大成功。所以，這些就是不同的垂直領域。就競爭而言，我認為一切都始於核心。問題的核心在於 CPU 擴展速度已經放緩。因此，世界需要另一種發展方式。的確，正因為我們專注於此，我們才處於非常有利的地位。Google發布了GPU 3，但它仍然落後於我們的Tensor Core GPU。我們的 Volta 是我們採用全新設計概念打造的第一代 GPU。它被稱為 Tensor Core GPU。我們遠遠領先競爭對手。而且——但更重要的是，它還是可編程的。它不是一個單一的功能。它是可編程的。它不僅速度更快，而且更靈活。由於其靈活性，開發人員可以將其用於各種應用程序，無論是醫學成像、天氣模擬、深度學習還是電腦圖形學。因此，我們的 GPU 可以在地球上的每個雲端和每個資料中心使用。開發者需要這些功能，以便他們能夠開發自己的軟體。因此，我認為一方面，將 GPU 與我們的 Tensor Core GPU 中的眾多功能之一進行比較過於簡化。但即便你真的做到了，我們也更快。我們支援更多框架。我們支持所有神經網路。因此，如果你查看 GitHub，你會發現上面發布了大約 60,000 篇不同的神經網路研究論文，這些論文都是在 NVIDIA GPU 上運行的。而第二種選擇只需要很少的錢。因此，這可以讓你對我們GPU的覆蓋範圍和能力有所了解。

Your next question comes from Toshiya Hari with Goldman Sachs.

下一個問題來自高盛的 Toshiya Hari。

Jensen, I had a question regarding your decision to pull the plug on your GeForce partner program. I think most of us read your blog from last Friday, I think it was. So we understand the basic background. But if you can describe what led to this decision, and perhaps talk a little bit about the potential implications, if any, in terms of your ability to compete or gain share, that will be really helpful.

Jensen，關於你決定終止GeForce合作夥伴計劃，我有一個問題。我想我們大多數人都讀過你上週五的博客，我記得是那天。所以我們了解了基本背景。但是，如果您能描述一下導致您做出此決定的原因，並談論這可能會對您的競爭力或市場份額產生哪些潛在影響（如果有的話），那將非常有幫助。

Yes. Thanks for the question, Toshiya. At the core, the program was about making sure that gamers who buy graphics cards knows exactly the GPU brand that's inside. And the reason for that is because we want gamers to -- the gaming experience of a graphics card depends so much on the GPU that is chosen. And we felt that using one gaming brand, a graphics card brand and interchanging the GPU underneath causes it to be less -- causes it to be more opaque and less transparent for gamers to choose the GPU brand that they wanted. And most of the ecosystem loved it. And some of the people really disliked it. And so instead of all that distraction, we're doing so well, and we're going to continue to help the gamers choose the graphics cards like we always have, and things will sort out. And so we decided to pull the plug because the distraction was unnecessary and we had too much good stuff to go do.

是的。謝謝你的提問，Toshiya。該計劃的核心是確保購買顯示卡的玩家能夠確切地知道顯示卡內部的GPU品牌。原因在於，我們希望玩家－顯示卡的遊戲體驗很大程度取決於所選的GPU。我們覺得，使用同一個遊戲品牌、同一個顯示卡品牌，然後更換下面的GPU，會導致遊戲體驗更加不透明，玩家選擇自己想要的GPU品牌變得更加困難。而且，生態系統中的大多數生物都喜歡它。有些人真的很不喜歡它。所以，我們沒有受到任何干擾，反而做得非常好，我們將繼續像以往一樣幫助玩家選擇顯示卡，一切都會好起來的。因此，我們決定結束這段關係，因為這種幹擾沒有必要，而且我們還有很多好事情要做。

Next question is from C.J. Muse with Evercore ISI.

下一個問題來自 Evercore ISI 的 C.J. Muse。

This is (inaudible) calling in for C.J. Muse. So I had a question on HPC. TSMC, on their recent call, raised their accelerator attach rate forecast in HPC to 50% from mid-teens. So I would love to get further details on what exactly NVIDIA is doing to software services, et cetera, that's kind of creating this competitive positioning in HPC and AI, basically. And then, if I could ask a follow-up, basically, in benchmarks. So there's been some news on AI benchmarks, whether it's Stanford's DAWNBench, et cetera. So I would love to get your thoughts on, a, the current state of benchmarks for AI workloads. And b, the relative positioning of ASICs versus GPUs, especially as we move towards newer networks like RNN and GAN, et cetera.

這裡是（聽不清楚）呼叫 C.J. Muse。我有一個關於高效能運算的問題。台積電在最近的電話會議上，將高效能運算 (HPC) 加速器附加率預測從 15% 左右上調至 50%。所以我很想進一步了解英偉達在軟體服務等方面的具體舉措，這些舉措基本上是在高效能運算和人工智慧領域打造競爭優勢。然後，如果可以的話，我想問一個後續問題，主要是關於基準測試方面的問題。所以，關於人工智慧基準測試有一些新消息，例如史丹佛大學的 DAWNBench 等等。所以我很想聽聽您對以下方面的看法：a，人工智慧工作負載基準測試的現狀。b，ASIC 與 GPU 的相對定位，尤其是當我們轉向 RNN 和 GAN 等更新的網路時。

Yes. Thanks for the question. HPC. First of all, at the core, CPU scaling has stalled and it's reached the limits of physics. And the world needs another approach to go forward. We created the GPU computing approach 1.5 decades ago, and I think at this point, with the number of developers jumping on, the number of applications that's emerging, it's really clear that the future of HPC has accelerated. And our GPU approach, because of its flexibility, because of its performance, because of the value that we create, that as a result of the throughput of a data center, we save people so much money just in cables alone. Oftentimes, more than pays for the GPUs that they buy. And the reason for that is because the number of servers is reduced dramatically. And so I think the future of HPC is about acceleration, and the NVIDIA CUDA GPUs are really in a great position to serve this vacuum that's been created. With respect to benchmarks, you might have seen that earlier this week, we released 3 speed records. The fastest single GPU, the fastest single node -- single computer node. A definition of a computer node is something that fits in a box that runs one operating system, one node. And one instance, one cloud instance. We now have the fastest speed record for 1 GPU, 1 node and 1 instance. And so we love benchmarks. Nothing is more joyful than having a benchmark to demonstrate your leadership position. And in the world of deep learning, the number of networks is just growing so fast because the number of different applications that deep learning is able to solve is really huge. And so you need a lot of software capability and you're -- the versatility of your platform needs to be great. We also have a lot of expertise in the company and software. I mean, NVIDIA is really a full-stack computing company. From architecture, to system software, to algorithms, to applications, we have a great deal of expertise across the entire stack. And so we love these complicated benchmarking -- benchmarks that are out there, and I think this is a great way for us to simplify our leadership position. I think long term, the number of networks that are going to emerge will continue to grow. And so the flexibility of ASICs is going to be its greatest downfall. And if someone were to create a general-purpose parallel accelerating processor like ours, and had it designed to be incredibly good at deep learning, like recently what we did with our Tensor Core GPU, which is a reinvented GPU, and Volta is the first one, I -- it's going to be hard. It's going to be expensive, and we've been doing it for a long time. And so I think this is a -- it's a great time for us.

是的。謝謝你的提問。高效能運算。首先，從根本上講，CPU 擴展性已經停滯不前，達到了物理極限。世界需要另一種前進的方式。我們在 15 年前創造了 GPU 運算方法，我認為現在，隨著越來越多的開發者加入進來，以及越來越多的應用程式湧現，很明顯，高效能運算的未來已經加速發展。由於我們的 GPU 方案具有靈活性、高效能以及我們所創造的價值，資料中心的吞吐量大幅提升，僅有電纜一項就能為用戶節省大量資金。通常情況下，收益會超過他們購買顯示卡的費用。原因在於伺服器數量大幅減少。所以我認為高效能運算的未來在於加速，而NVIDIA CUDA GPU完全有能力填補這一空白。關於基準測試，您可能已經注意到，本週早些時候，我們發布了 3 項速度記錄。速度最快的單一GPU，速度最快的單一節點－單一電腦節點。電腦節點的定義是：裝在一個盒子裡，運行一個作業系統（即一個節點）的東西。一個實例，一個雲端實例。我們現在擁有 1 個 GPU、1 個節點和 1 個執行個體的最快速度記錄。所以我們都喜歡基準測試。沒有什麼比擁有一個能夠證明你領導地位的標竿更令人高興的了。在深度學習領域，網路數量成長如此之快，是因為深度學習能夠解決的不同應用領域數量非常龐大。因此，你需要強大的軟體功能，而且你的平台必須具有很強的多功能性。我們在公司和軟體方面也擁有豐富的專業知識。我的意思是，NVIDIA 真是一家全端運算公司。從架構、系統軟體、演算法到應用程序，我們在整個技術堆疊上都擁有豐富的專業知識。因此，我們喜歡這些複雜的基準測試——現有的基準測試，我認為這對我們簡化領導地位來說是一個很好的方法。我認為從長遠來看，將會湧現出更多的網路。因此，ASIC的靈活性將成為其最大的弱點。如果有人要開發像我們這樣的通用平行加速處理器，並且將其設計得非常擅長深度學習，就像我們最近開發的 Tensor Core GPU 一樣，它是一款重新設計的 GPU，而 Volta 是第一款，我——這將非常困難。這會很貴，而且我們已經這樣做了很長時間了。所以我覺得現在是——現在是我們的絕佳時機。

Next question is from Blayne Curtis with Barclays.

下一個問題來自巴克萊銀行的布萊恩‧柯蒂斯。

Jensen, I wanted to ask on the inference side [about edge inference]. And beyond autos, when you look at sizing that TAM, what are the other big areas that you think you can penetrate with GPUs and inference, besides auto?

Jensen，我想問推理方面的問題（關於邊緣推理）。除了汽車領域之外，在評估 TAM 規模時，除了汽車領域之外，您認為還可以利用 GPU 和推理技術滲透到哪些其他重要領域？

Yes, Blayne. The largest inference opportunity for us is actually in the cloud and the data center. That's the first great opportunity. The reason for that is there's just an explosion in the number of different types of neural networks that are available. There are image recognition, there's video sequencing, there's video -- there's recommender systems. There's speech recognition, speech synthesis, natural language understanding. There's just so many different types of neural networks that are being created. And creating one ASIC that can be adapted to all of these different types of networks is just a real challenge. And by the time that you create such a thing, it's called a Tensor Core GPU, which is what we created. And so I think the first opportunity for us in -- large-scale opportunity will be in the data center and the cloud. The second will be in vertical markets. The vertical market that you mentioned is self-driving cars. And we see a great opportunity in autonomous vehicles, both in the creation of autonomous vehicles. And I mentioned that before, between now and the time that we ramp our AV computers we call DRIVE, we're going to be selling a whole lot of servers so that the companies could develop their neural network models for their self-driving cars as well as simulating the virtual reality -- in virtual reality, there are various test drives as well as testing their neural network and their self-driving car stack against billions and billions of miles of saved-up prerecorded videos. And so in the vertical markets, we're going to see inference both in the data center for developing the self-driving car stack as well as in the self-driving cars themselves. Now in the self-driving cars, the ASPs for level 2 could be a few hundred dollars to a level 5 self-driving car, taxi or driverless taxi being a few thousand dollars. And I expect that driverless taxis will start going to market about 2019, and self-driving cars probably somewhere between 2020 and 2021. And I think the size of the market is fairly well modeled. And the simple way to think about that is, I believe that every single -- everything that moves someday will be autonomous or have autonomous capabilities. And so the 100 million cars, the countless taxis, all the trucks, all the agriculture equipment, all the pizza delivery vehicles, you name it, everything is going to be autonomous. And the market opportunity is going to be quite large. And that's the reason why we're so determined to go create that market.

是的，布萊恩。對我們來說，最大的推理機會實際上在於雲端和資料中心。這是第一個絕佳的機會。原因在於，目前可用的不同類型神經網路的數量呈現爆炸性成長。有圖像辨識、影片排序、影片——還有推薦系統。包括語音辨識、語音合成和自然語言理解。目前正在開發的神經網路類型實在太多了。而要打造一款能夠適應所有這些不同類型網路的 ASIC 晶片，確實是一個巨大的挑戰。當你創造出這樣的東西時，它就被稱為 Tensor Core GPU，而這正是我們所創造的。所以我認為，我們面臨的第一個大規模機會將在資料中心和雲端運算領域。第二部分將面向垂直市場。您提到的垂直市場是自動駕駛汽車。我們看到自動駕駛汽車領域蘊藏著巨大的機遇，無論是在自動駕駛汽車的研發或其他方面。我之前提到過，從現在到我們推出名為 DRIVE 的 AV 電腦期間，我們將銷售大量的伺服器，以便各公司能夠開發其自動駕駛汽車的神經網路模型，並模擬虛擬實境——在虛擬實境中，可以進行各種試駕，還可以用數十億英里的預錄影片來測試其神經網路和自動駕駛汽車系統。因此，在垂直市場中，我們將看到推斷技術既應用於開發自動駕駛汽車堆疊的資料中心，也應用於自動駕駛汽車本身。現在，在自動駕駛汽車領域，2 級自動駕駛汽車的平均售價可能只有幾百美元，而 5 級自動駕駛汽車、計程車或無人駕駛計程車的平均售價可能高達數千美元。我預計無人駕駛計程車將於 2019 年左右開始進入市場，而自動駕駛汽車可能會在 2020 年至 2021 年之間進入市場。我認為市場規模的模型已經相當準確了。簡單來說，我認為所有會動的東西，將來有一天都會是自主的，或是有自主能力。因此，1億輛汽車、無數的計程車、所有的卡車、所有的農業設備、所有的披薩外送車，你能想到的，一切都將自動化。市場機會將會非常大。正因如此，我們才如此堅定地要去創造這個市場。

Your next question is from Tim Arcuri with UBS.

下一個問題來自瑞銀集團的提姆·阿庫裡。

I actually wanted to go back to the question about seasonality for gaming in June. Normal seasonal sounds like it's up mid-teens for June in gaming. But obviously, the comps are skewed a little bit because of the channel restock and the crypto stuff. So does the guidance for June assume that gaming is better or worse than that mid-teens normal seasonal?

其實我六月想回到遊戲季節性這個問題來討論一下。遊戲產業的正常季節性波動聽起來像是六月會漲到十幾個百分點。但很顯然，由於通路補貨和加密貨幣相關因素，這些比較數據有些偏差。那麼，6 月的指導意見是假設遊戲產業的表現比往年同期（十幾分）好還是差呢？

We are expecting the -- we are expecting Q2 to be better than seasonality, if I understand your question. It should be -- we're expecting Q2 to be better than Q1 and we're expecting Q2 to be better than seasonality. Did that answer your question?

如果我理解你的問題沒錯的話，我們預期第二季的情況會比季節性因素的影響更好。應該如此——我們預計第二季會比第一季好，而且我們預計第二季會比季節性因素的影響更好。這樣回答了你的問題嗎？

Your next question is from Atif Malik with Citi.

下一個問題來自花旗銀行的阿提夫‧馬利克。

I have a question for Colette. Colette, first thank you for breaking out crypto sales in the OEM line and guide for us. I have a question on your gross margins. Your gross margins have been expanding on product mix, despite component pricing headwinds on the DRAM side. When do you expect component pricing to become a tailwind to your gross margin?

我有個問題想問科萊特。Colette，首先感謝你為我們說明 OEM 產品線中的加密貨幣銷售情況和指導。我有一個關於貴公司毛利率的問題。儘管DRAM元件價格面臨不利因素，但產品組合的多元化仍擴大了毛利率。您預計零件價格何時才能對毛利率產生正面影響？

Thanks so much for the question. When you think about our gross margins, just over this last quarter, as you know, we were working on stabilizing the overall supply that was out there in the market for consumer GPUs. We benefited from that with a higher gross margin as we filled and completed that. You've seen us absorb a significant amount of the component pricing changes that we have seen, particularly around the memory. We're not here to be able to forecast, generally, when those pricing of those components will stabilize. But we believe in terms of the value added that our platforms provide, the components are an important part of finishing that. But I think we have a tremendous amount more value that we are adding in terms of the software on top of our platforms, which is enabling our gross margins.

非常感謝你的提問。考慮到我們上個季度的毛利率，如您所知，我們一直在努力穩定市場上消費級GPU的整體供應。由於訂單已完成，我們的毛利率也因此提高了。您已經看到我們吸收了相當一部分組件價格變化的影響，尤其是在內存方面。我們無法預測這些零件的價格何時才能穩定下來。但我們相信，就我們平台所提供的附加價值而言，這些元件是實現這一目標的重要組成部分。但我認為，我們在平台之上的軟體方面創造了更大的價值，這提高了我們的毛利率。

Your next question is from Chris Caso with Raymond James.

下一個問題來自 Raymond James 的 Chris Caso。

My question is the progress on the deployment of Volta into the cloud service providers. You talked in your prepared remarks about 5 deployments, including the Google beta. Can you talk about the -- how soon we can expect to see some of those remaining deployments? And of those already launched, how far are they along? What -- I guess, to say proverbially, what inning are we in, in these deployments?

我的問題是Volta在雲端服務供應商中的部署進度。您在準備好的演講稿中談到了 5 個部署項目，其中包括 Google 測試版。您能否談談－我們大概什麼時候能看到剩餘的部署工作？已經啟動的專案進度如何？我想，用一句俗語來說，我們現在處於這些部署的哪個階段？

Yes. So first of all, Volta is a reinvented GPU. Volta is the world's first GPU that has been designed to be incredibly good at deep learning. We call it the Tensor Core GPU. It has still retained all of the flexibilities of all -- everything that CUDA has ever run is backwards compatible with everything that runs on CUDA. But it has new architectures designed to be incredibly good at deep learning. We call it a Tensor Core GPU. And that's the reason why it has all of the benefits of our GPU, but none of the ASICs can catch up to it. And so Volta is really a breakthrough. We're going to be very successful with Volta. Every cloud will have it. The initial deployment is for internal consumption. Volta has been shipping to the cloud service -- cloud providers, the Internet service companies, for the vast majority of last quarter, as you guys know. And they're using it internally. And now they're starting to open up Volta for external consumption, their cloud customers. And they are moving as fast as they can. And my expectation is that you're going to see a lot more coming online this quarter.

是的。首先，Volta 是一款重新設計的 GPU。Volta 是世界上第一款專為深度學習而設計的 GPU。我們稱之為 Tensor Core GPU。它仍然保留了所有靈活性——CUDA 曾經運行過的所有內容都向後相容於所有在 CUDA 上運行的內容。但它採用了全新的架構，旨在實現極其出色的深度學習。我們稱之為 Tensor Core GPU。這就是為什麼它具備我們GPU的所有優點，但沒有任何ASIC晶片能夠趕上它的原因。因此，伏特電池確實是一項突破。我們與Volta的合作一定會非常成功。每朵雲都會有它。初始部署僅供內部使用。如你所知，上個季度的大部分時間裡，Volta 一直向雲端服務（雲端供應商、網路服務公司）出貨。他們內部也在使用它。現在他們開始向外部用戶（他們的雲端客戶）開放 Volta。他們正以最快的速度前進。我預計本季會有更多產品上線。

Your next question comes from Mark Lipacis with Jefferies.

下一個問題來自傑富瑞集團的馬克·利帕西斯。

I had a question about the DGX family of products. Our own fieldwork is indicating very positive reception for DGX. And I was wondering, Jensen, if can you help us understand, the high-growth we've seen in the data center business, to what extent is that being driven by the DGX? And what, when DGX-2 starts to ramp in the back half of the year, is that -- is it something that kind of layers on top of DGX? Does DGX-2 layer top of DGX? Are they (inaudible) segments, that you're kind of segmenting the market with these different products? Any color on how to think about those 2 products are -- should be -- would be helpful.

我有一個關於DGX系列產品的問題。我們自己的實地調查顯示，DGX 的市場反應非常正面。Jensen，我想知道，您能否幫助我們理解，我們在資料中心業務中看到的高速成長在多大程度上是由 DGX 推動的？那麼，當 DGX-2 在下半年開始增產時，它是不是會在 DGX 的基礎上疊加一些東西呢？DGX-2 是否位於 DGX 的頂層？這些（聽不清楚）細分市場，是指你們用這些不同的產品來分割市場嗎？關於如何看待這兩款產品（現在和將來）的任何想法都將有所幫助。

Colette, could you give me a brief version of that? It was kind of crackling on my side.

科萊特，你能簡要地跟我講一下嗎？我這邊好像有劈啪聲。

I'm going to ask the operator if they could ask for the question again because it was also, on our side, a little crackly.

我打算問問接線生能不能再問這個問題，因為我們這邊也聽到了一些雜音。

Yes. Mark, your line is open. Please restate your question.

是的。馬克，你的線路已接通。請重新表述您的問題。

Okay. Can you hear me better now?

好的。現在你能聽得更清楚些嗎？

Yes. Much better, Mark.

是的。好多了，馬克。

Okay. Sorry about that. I'm at the airport. So the question was on the DGX family of products. Our own fieldwork indicates very positive reception. I was wondering, Jensen, if you could help us understand the high-growth you've seen in the data center market. How much is DGX contributing to that? And then, when DGX-2 starts to ramp in the second half of the year, how do we think about DGX-1? Does it replace DGX -- the original DGX? Or you're going after different segments? Or do they layer on top of one another? Any color on that would be helpful.

好的。抱歉。我在機場。所以問題是關於DGX系列產品的。我們自己的實地調查顯示，反應非常正面。Jensen，我想請教一下，您能否幫助我們理解您在資料中心市場所看到的快速成長現象？DGX對此貢獻有多大？那麼，當 DGX-2 在下半年開始增產時，我們該如何看待 DGX-1 呢？它會取代 DGX——原版的 DGX 嗎？還是你們瞄準的是不同的市場區隔？還是它們層層疊放呢？如果能加上任何顏色就更好了。

I see. Thank you. DGX-2 and DGX-1 will both be in the market at the same time. And DGX is a few hundred million dollar business. It's -- it was introduced last year. So its growth rate is obviously very high. It's designed for enterprises where they don't -- they need to have their computers on-premise, but they don't want to build a supercomputer and they don't have the expertise to do so. And they would like to pull a supercomputer out of a box, plug it in, and start doing supercomputing. And so DGX is really designed for enterprises. It's designed for car companies. It's designed for health care companies doing life sciences work or medical imaging work. We recently announced a project called Project Clara, which basically takes medical imaging equipment, virtualizes them, containerizes the software, and turns it into a -- and most medical imaging equipment today are computational and they -- a lot of them run on NVIDIA CUDA anyways. And so we put that -- we can put that into the data center, we can virtualize their medical instruments and it gives them the opportunity to upgrade the millions of instruments that are out in the marketplace today. And so DGX is really designed for enterprises and we're seeing great success there. It's really super easy to use and it comes with direct support from HPC and AI researchers at NVIDIA. And the answer to your question at the end is, both of them will be in the marketplace at the same time.

我懂了。謝謝。DGX-2 和 DGX-1 將同時上市。DGX是一家價值數億美元的公司。它是——它是去年推出的。所以它的成長率顯然非常高。它是為那些不需要——他們需要將計算機部署在本地，但他們不想建造超級計算機，也沒有建造超級計算機的專業知識的企業而設計的。他們希望把一台超級電腦從箱子裡拿出來，插上電源，就可以開始進行超級運算。因此，DGX 實際上是為企業設計的。它是為汽車公司設計的。它是為從事生命科學或醫學影像工作的醫療保健公司設計的。我們最近宣布了一個名為「Clara 項目」的項目，該項目基本上是將醫療成像設備虛擬化，將軟體容器化，並將其變成——而且如今大多數醫療成像設備都是計算設備，它們——很多設備無論如何都在 NVIDIA CUDA 上運行。因此，我們可以將這些技術——我們可以將它們放入資料中心，我們可以將他們的醫療儀器虛擬化，這使他們有機會升級目前市場上數百萬台儀器。因此，DGX 實際上是為企業設計的，而且我們在這方面取得了巨大的成功。它使用起來真的非常簡單，而且還能獲得 NVIDIA 高效能運算和人工智慧研究人員的直接支援。最後，你問題的答案是，它們將同時進入市場。

Next question is from Mitch Steves with RBC Capital Markets.

下一個問題來自加拿大皇家銀行資本市場的米奇·史蒂夫斯。

I'm actually going to go with a more nitty-gritty question just on the financial side, just to make sure I'm understanding this right. So the OEM beat was pretty material, given a lot of crypto revenue. Is it still the case that OEM is materially lower gross margin than your corporate average at this time?

我其實想問一個更具體、更偏向財務方面的問題，以確保我理解正確。考慮到加密貨幣帶來的大量收入，OEM廠商的業績表現相當出色。目前OEM業務的毛利率是否仍明顯低於貴公司的平均？

Sure. I'll take that question. Generally, our OEM business can be a little bit volatile. Because remember, OEM business incorporates our mainstream GPUs as well as our Tegra integrated. So we have development platforms that we sell on some of the Tegra piece of it. But they are slightly below, and I think you can go back and refer to our discussion at Investor Day as there's a slide there that talks about those embedded pieces and them being below. So yes, you're correct. Again, a very small part of our business right now.

當然。我來回答這個問題。一般來說，我們的OEM業務波動性較大。因為請記住，OEM 業務涵蓋了我們的主流 GPU 以及我們的 Tegra 整合顯示卡。所以我們有一些基於 Tegra 技術的開發平台，我們會在平台上銷售這些平台。但它們略低於預期，我認為您可以回顧一下我們在投資者日上的討論，因為那裡有一張幻燈片談到了這些嵌入式部分以及它們低於預期的情況。是的，你說得對。再次強調，這目前只占我們業務的一小部分。

Your next question comes from Christopher Rolland with Susquehanna.

下一個問題來自薩斯奎哈納的克里斯多福羅蘭。

So your competitor thinks that just 10% of their sales were from crypto, or like $150 million, $160 million. And you guys did almost $300 million there. And perhaps I think there could actually be some in gaming as well, which would imply that you guys have 2/3 or more of that market? So I guess, what's going on there? Is there a pricing dynamic that's allowing you to have such share there? Or do you think it's your competitors that don't know what's actually being sold to miners versus gamers? Why such implied share in that market?

所以你的競爭對手認為他們只有 10% 的銷售額來自加密貨幣，大約是 1.5 億美元、1.6 億美元。你們在那裡賺了將近3億美元。或許遊戲領域也存在這樣的市場，這意味著你們可能佔據了該市場三分之二甚至更多的份額？所以，我猜，那裡到底發生了什麼事？是否存在某種定價策略，使得你們能夠佔據如此大的市場份額？還是你認為你的競爭對手不了解礦工和遊戲玩家實際購買的產品是什麼？為什麼該市佔率如此之低？

Well, we try to as transparently reveal our numbers as we can. And so the thing that we -- our strategy is to create a SKU that allows the crypto miners to fulfill their needs, and we call it CMP, and to the -- as much as possible, fulfill their demand that way. Sometimes it's just not possible because the demand is too great. And -- but we try to do so. And we try to keep the miners on the CMP SKUs as much as we can. And so I'm not exactly sure how other people do it, but that's the way we do it.

我們會盡可能透明地公佈我們的數據。因此，我們的策略是創建一個 SKU，讓加密貨幣礦工能夠滿足他們的需求，我們稱之為 CMP，並盡可能以這種方式滿足他們的需求。有時，由於需求太大，根本無法實現。但我們努力做到這一點。我們會盡可能讓礦工們繼續使用 CMP SKU。所以我不太清楚其他人是怎麼做的，但我們就是這麼做的。

Your next question is from Craig Ellis with B. Riley.

你的下一個問題來自 Craig Ellis 和 B. Riley。

Jensen, I just wanted to come back to an announcement that you made at GTC with ray tracing. Because the technology looked like it was very high fidelity, and I think you noted at that time that it was very computationally intensive. So the question is, as we think about the gaming business and the potential for ray tracing to enter that platform group, what does it mean for dynamics that we've seen in the past, for example, the ability to really push the high end of the market with high-end capability, 1070. Ti launched late last year. It was very successful. Does this give you further flexibility for those types of launches as you bring exciting and very high-end technology to market?

Jensen，我只是想再談談你在 GTC 上宣布的光線追蹤技術。因為這項技術看起來非常逼真，而且我想你當時也注意到它對運算能力要求很高。所以問題是，當我們思考遊戲產業以及光線追蹤技術進入該平台領域的潛力時，這對我們過去看到的動態意味著什麼，例如，利用高端功能（如 1070）真正推動高端市場的能力。Ti 於去年底發布。非常成功。這是否能讓您在將令人興奮的高端技術推向市場時，擁有更大的彈性來進行此類產品發布？

Yes. I appreciate it. NVIDIA RTX is the biggest computer graphics invention in the last 15 years. It took us 1 decade to do. We literally worked on it continuously for 1 decade. And to put it into perspective, it's basically film rendering, cinematic rendering, except it's in real time. It merges the style of computer graphics, rasterization, and light simulation, what people call ray tracing, as well as deep learning and AI, merged it into one unified framework so that we can achieve cinematic rendering in real time. What it currently takes is a server about a few hours, depending on the scene, it might take as long as a full day, take a few hours to render one frame. So it takes a server, one node of a server, several hours to render one frame. And in order to render 30 frames per second, just imagine the number of servers you need. If it takes several hours per frame, and you need to render 30 frames per second in order to be real-time, it basically takes a high-performance computer, a supercomputer, a render farm, that's why they call it a render farm, it's a full data center designed just for rendering. And now we've created NVIDIA RTX which makes it possible to do in real time. We demonstrated RTX on 4 Quadro GV100s. It takes 4 of our latest generation Volta Tensor Core GPUs to be able to render 30 frames per second, the Star Wars cinematic that people enjoyed. And so the amount that we saved, we basically took an entire data center, reduced it into one node. And we're now doing it in real time. And so the amount of money that we can save people who create movies, people who do commercials, people who use film rendering to create the game content, almost every single game is done that way. There's quite a bit of offline rendering to create the imagery and the textures and the lighting. And then, of course, architectural design and car design, the number of applications, the number of industries that are built on top of modern computer graphics is really quite large. And I'm certain that NVIDIA RTX is going to impact every single one of them. And so that's our starting point, is to dramatically reduce the cost of film rendering, dramatically reduce the time that it takes to do it, and hopefully, more GPU servers will get -- will be purchased. And of course, better content will be created. Long-term, we've also now plotted the path towards doing it in real time. And someday, we will be able to put RTX into a GeForce gaming card and the transformation to the revolution to the gaming industry will be quite extraordinary. So we're super excited about RTX.

是的。謝謝。NVIDIA RTX 是近 15 年來電腦繪圖技術最偉大的發明。我們花了十年才完成。我們為此持續工作了整整十年。簡單來說，它基本上就是電影渲染，電影級渲染，只不過是即時渲染的。它融合了電腦圖形、光柵化和光照模擬（人們稱之為光線追蹤）的風格，以及深度學習和人工智慧，並將它們融合到一個統一的框架中，以便我們能夠即時實現電影級渲染。目前伺服器渲染一幀需要幾個小時，具體時間取決於場景，有時甚至需要一整天，有時則需要幾個小時。所以一台伺服器，或者說伺服器上的一個節點，要幾個小時才能渲染出一幀畫面。為了達到每秒渲染 30 幀的目標，想想你需要多少台伺服器。如果每幀需要幾個小時才能渲染完成，而為了實現即時渲染，又需要每秒渲染 30 幀，那麼基本上就需要一台高效能電腦、一台超級電腦、一個渲染農場。這就是為什麼它被稱為渲染農場，它是一個專門為渲染而設計的完整資料中心。現在我們創造了 NVIDIA RTX，它使得即時實現這一目標成為可能。我們在 4 個 Quadro GV100 顯示卡上示範了 RTX 技術。我們需要 4 個最新一代 Volta Tensor Core GPU 才能以每秒 30 幀的速度渲染出人們喜愛的《星際大戰》電影。因此，我們節省下來的資源，基本上相當於把整個資料中心縮減到一個節點上。我們現在正在實時進行這項工作。因此，我們可以為電影製作人、廣告製作人、使用電影渲染技術製作遊戲內容的人節省大量資金，幾乎所有遊戲都是這樣製作的。為了創建影像、紋理和光照效果，需要進行大量的離線渲染。當然，建築設計和汽車設計等應用領域，以及建立在現代電腦圖形技術之上的行業數量，都非常龐大。我確信NVIDIA RTX將會對它們中的每一個產生影響。因此，我們的出發點是大幅降低電影渲染的成本，大幅縮短渲染時間，希望能夠購買更多的 GPU 伺服器。當然，未來還會創造出更好的內容。從長遠來看，我們現在也規劃了實現即時傳輸的路徑。總有一天，我們將能夠把 RTX 技術應用到 GeForce 遊戲顯示卡中，這將為遊戲產業帶來非同尋常的變革。所以我們對RTX感到非常興奮。

Your next question is from Stacy Rasgon with Bernstein.

下一個問題來自伯恩斯坦的史黛西·拉斯貢。

This is a question for Colette. I want to follow up, again, on the seasonality, understanding the prior comments. Normal seasonal for Q2 for gaming would be up in the double digits. Given your commentary on the crypto declined into Q2, given your commentary on just the general drivers around data center and the Volta ramps, I can't bring that together with the idea of gaming being above seasonal within the guidance -- the context of your guidance envelope. So how should I reconcile those things? How are you actually thinking about seasonality for gaming into Q2 within the context of the scenarios that are currently contemplated in your guidance for next quarter?

這是問科萊特的問題。我想再次就季節性問題進行補充說明，並理解先前的評論。正常情況下，遊戲產業第二季的季節性成長幅度會達到兩位數。鑑於您對加密貨幣在第二季度下跌的評論，以及您對資料中心和 Volta 產能提升等一般驅動因素的評論，我無法將這些與遊戲業務在指導範圍內高於季節性水平的想法聯繫起來——這是在您的指導範圍之內的。那我該如何調和這些事情呢？在您對下一季業績展望中設想的各種情景下，您是如何考慮第二季遊戲產業的季節性因素的？

Sure, Stacy. Let me see if I can bridge together Jensen and then some comments here. Unfortunately, they are moving quite fast to the next question, so I wasn't able to add on. But let me see if I can add on here and provide a little bit of clarity in terms of the seasonality. Remember, in Q1, we outgrew seasonality significantly. We left Q4 with very low inventory in terms of the channel. We spent Q1 working on establishing a decent amount of inventory available. We wanted to concentrate on our miners separately, and then you can see we did that in terms of Q1 by moving that to OEM and moving that to cryptocurrency-only boards. So we left Q1 at this point with healthy overall channel inventory levels as far as where we stand. So that then takes you now to Q2. But if we overshot in terms of seasonality in terms of Q1, we don't have to do those channel-fill dynamics again as we get into Q2. But we do have demand out there for our gamers that we can now address very carefully with the overall inventory that we now have available. So putting together Q1 and Q2 together, yes, we are within normal seasonality, again, for our guidance. And we'll see how we'll finish in terms of the quarter. But you should be in that range. So yes, from a normal seasonality, at a year-to-date inclusive of Q2, yes, we're on that overall seasonality. Always keep in mind, generally, our H2s are usually higher than our overall H1s, and that's what you should think about our overall guidance. Gaming is still strong. We have to comment that our overall drivers that have taken us to this place over the last 3 to 5 years with phenomenal growth and our ability to grow that overall market is still here and all of those things are together. We've just had a few quarters in terms of making sure that we get the overall channel correct and put our miners separately. I hope that clarifies in terms of where we are, in terms of gaming seasonality.

當然可以，斯黛西。讓我看看能不能把詹森的話和這裡的一些評論連結起來。很遺憾，他們很快就要問下一個問題了，所以我沒能補充。不過，讓我看看我能否補充一些內容，並就季節性方面做一些更清晰的解釋。請記住，在第一季度，我們的成長已經顯著超過了季節性波動的影響。第四季末，我們的通路庫存非常低。我們第一季致力於建立充足的庫存。我們希望將礦機單獨集中起來，然後您可以看到，我們在第一季透過將礦機轉移到 OEM 和加密貨幣專用闆卡上實現了這一點。因此，就我們目前的情況來看，第一季末的整體通路庫存水準良好。那麼，現在就進入第二題了。但是，如果我們在第一季季節性方面做得過火了，那麼到了第二季度，我們就不必再進行這些管道填充動態調整了。但是，我們確實存在對遊戲玩家的需求，我們現在可以利用現有的庫存非常謹慎地滿足這些需求。因此，將第一季和第二季的數據結合起來來看，是的，我們目前處於正常的季節性波動範圍內，僅供參考。我們拭目以待，看看本季最終的業績如何。但你的體重應該在這個範圍內。所以，是的，從正常的季節性來看，從今年迄今為止（包括第二季）來看，是的，我們目前處於整體季節性狀態。請記住，一般來說，我們的 H2 值通常高於我們的整體 H1 值，這就是您應該對我們的整體指導的看法。遊戲產業依然強勁。我們必須指出，過去 3 到 5 年來，推動我們取得如此驚人成長的整體動力以及我們拓展整體市場的能力依然存在，所有這些因素都緊密相連。我們花了幾個季度的時間來確保整個通道運作正常，並將我們的礦工單獨放置。我希望這能澄清我們目前在遊戲季節性方面的位置。

Your last question comes from Will Stein with SunTrust.

您的最後一個問題來自 SunTrust 銀行的 Will Stein。

The question relates to the supply chain challenges that you've talked so much about in the gaming end market. I'm wondering if there's something particular to that end market that is making the shortages concentrated there? Or are, in fact, other end markets in particular that the data center end market also somewhat restricted from what growth they might have achieved if there weren't the shortages that are out there? And maybe talk about the pace of recovery of those. That be really helpful.

這個問題與您在遊戲終端市場中多次提到的供應鏈挑戰有關。我想知道終端市場是不是有什麼特殊之處，導致短缺現象集中在那裡？或者，事實上，其他終端市場，特別是資料中心終端市場，是否也因為目前存在的短缺問題而受到一定程度的限制，無法實現它們本來可以達到的成長？或許還可以談談這些地區的復甦速度。那真的很有幫助。

Let me start off here, and I'll have Jensen finish up on the last part of that question. But overall, our data center business did phenomenal. Volta is doing extremely well. And even now with 32-bit, we're seeing tremendous adoption throughout. Again, remember it's very different than the overall consumer business. You have significant amount of time for qualification, and that is moving extremely fast based on a lot of other industries and their ability to qualify. So no, there is not a supply challenge at all in terms of our data center. And our overall growth in data center, we're extremely pleased with in terms of how the quarter came out. I'll turn it over to you, Jensen, and you can answer the rest of the part of it.

我先來回答這個問題，剩下的部分就交給 Jensen 來解答。但總體而言，我們的資料中心業務表現非常出色。Volta表現非常出色。即使現在採用的是 32 位元技術，我們也看到了其在各個領域的廣泛應用。再次提醒，這與整體消費業務非常不同。您有充足的時間進行資格認證，而且相對於其他許多行業的資格認證速度而言，這個速度已經非常快了。所以，就我們的資料中心而言，根本不存在供應方面的挑戰。對於本季資料中心的整體成長情況，我們感到非常滿意。詹森，剩下的部分就交給你回答吧。

Yes. The reason why miners love GeForce is because miners are everywhere in the world. One of the benefits of cryptocurrency is that it's not any sovereign currency. And it's, in the digital world, it's distributed. And GeForce is the single largest distributed supercomputing infrastructure on the planet. Every gamer has a supercomputer in their PC. And GeForce is so broadly distributed, it's available everywhere. And so GeForce is really a good candidate for any new cryptocurrency or any new cryptography algorithm that comes along. We try the best we can to go directly to the major miners, and the major miners. And they represent the vast majority of the demand. And to the best of our ability, serve their needs directly and we call that CMP, and that's why it's not called GeForce. They're called CMP. And we can serve those miners directly, hopefully, to take some of the demand pressure off of the GeForce market. Because ultimately, what we would like is, we would like the market for GeForce pricing to come down so that the gamers could benefit from the GeForces that we built for them. And the gaming demand is strong. I mean, the bottom line is, Fortnite is a home run. The bottom line is, PUBG is a home run. And the number of gamers that are enjoying these games is really astronomic, as people know very well. And it's a global phenomenon. These 2 games are equally fun in Asia as it is in Europe, as it is in the United States. And because you team up and this is a Battle Royale, you'd rather play with your friends. So it's incredibly social. It's incredibly sticky. And more and more -- more gamers that play, more of their friends join, and more of their friends join, more gamers that play. And so it's this positive feedback system, and the guys at Epic did a fantastic job creating Fortnite, and it's just a wonderful game genre that people are really enjoying. And so I think at the core of it, gaming is strong and we are looking forward to inventory normalizing in the channel so that pricing could normalize in the channel, so that gamers can come back to buy the GeForce cards that has now been in short supply for over a quarter. And so the pent-up demand is quite significant, and I'm expecting the gamers to be able to buy new GeForces pretty soon.

是的。礦工喜歡GeForce顯示卡的原因是，礦工遍布世界各地。加密貨幣的優點之一在於它不屬於任何主權貨幣。在數位世界中，它是分散式的。GeForce 是全球規模最大的分散式超級運算基礎設施。每個遊戲玩家的電腦裡都裝著一台超級電腦。而且 GeForce 的發行範圍非常廣泛，幾乎隨處可見。因此，GeForce 非常適合任何新的加密貨幣或任何新的加密演算法。我們盡最大努力直接聯繫主要的礦工和主要礦工。他們代表了絕大多數的需求。我們會盡最大努力直接滿足他​​們的需求，我們稱之為 CMP，這就是為什麼它不叫 GeForce 的原因。它們被稱為CMP。我們希望能夠直接服務這些礦工，從而減輕 GeForce 市場的部分需求壓力。因為歸根究底，我們希望GeForce顯示卡的價格能夠下降，讓玩家能夠從我們為他們打造的GeForce顯示卡中受益。遊戲需求強勁。我的意思是，總而言之，《要塞英雄》取得了巨大的成功。總之，PUBG 非常成功。眾所周知，喜歡玩這些遊戲的玩家數量非常龐大。而且這是一種全球現象。這兩款遊戲在亞洲、歐洲和美國都同樣有趣。因為這是組隊遊戲，而且是大逃殺模式，所以你更願意和朋友一起玩。所以它具有很強的社交性。它非常粘。而且玩家越來越多，他們的朋友也越來越多，他們的朋友也越來越多，玩家也越來越多。所以這是一個正回饋系統，Epic Games 的開發者在打造 Fortnite 方面做得非常出色，這確實是一個很棒的遊戲類型，人們真的很喜歡它。所以我認為，從根本上來說，遊戲市場依然強勁，我們期待通路庫存恢復正常，價格也能恢復正常，這樣玩家就可以回來購買已經短缺超過一個季度的 GeForce 顯示卡了。因此，被壓抑的需求相當大，我預計遊戲玩家很快就能買到新的 GeForce 顯示卡了。

Unfortunately, we have ran out of time. I will now turn it back over to Jensen for any closing remarks.

很遺憾，我們時間不夠了。現在我將把發言權交還給詹森，請他作總結陳詞。

Let's see here. Is it my turn again?

讓我看看。又輪到我了嗎？

It is.

這是。

Okay. We had another great quarter. Record revenue, record margins, record earnings, growth across every platform. Data center achieved another record, with strong demand for Volta and AI inference. Gaming was strong. We're delighted to see prices normalizing and we can better serve pent-up gamer demand.

好的。我們又度過了一個非常棒的季度。營收、利潤率和獲利均創歷史新高，所有平台均實現成長。資料中心再創佳績，對 Volta 和 AI 推理的需求強勁。遊戲業務表現強勁。我們很高興看到價格趨於正常，這樣我們就能更好地滿足遊戲玩家積壓已久的需求。

At the heart of our opportunity is the incredible growth of computing demand of AI, just as traditional computing has slowed. The GPU computing approach that we've pioneered is ideal for filling this vacuum. And our invention of the Tensor Core GPU has further enhanced our strong position to power the AI era. I look forward to giving you another update next quarter. Thank you.

我們面臨的機會的核心在於人工智慧運算需求的驚人成長，而傳統運算的需求卻在放緩。我們首創的GPU計算方法非常適合填補這一空白。我們發明的 Tensor Core GPU 進一步鞏固了我們在人工智慧時代中的強大地位。我期待下個季度再向您報告最新進展。謝謝。

This concludes today's conference call. Thank you, guys, for joining. You may now disconnect.

今天的電話會議到此結束。謝謝各位的參與。您現在可以斷開連線了。