加速量子運算研發 [S73197]
Accelerate R&D in Quantum Computing [S73197]
https://register.nvidia.com/flow/nvidia/gtcs25/vap/page/vsessioncatalog/session/1727998010163001XX9c

Eric Kessler, Amazon Braket 應用科學資深經理, AWS

人們希望有一天量子電腦能夠解決甚至最強大的傳統處理器都無法解決的問題。但這些計算機無法獨立運作。相反，量子電腦將成為高效能運算架構的一個組成部分，改變當今一些最具挑戰性的工作流程，例如藥物發現、投資組合優化等等。了解 AWS 和 NVIDIA 如何透過 AWS 的量子服務 Amazon Braket 和 NVIDIA 的 CUDA-Q 平台合作實現這一共同願景。透過提供客戶所需的工具，無論他們處於量子旅程的哪個階段，他們都可以輕鬆存取 Braket 上的各種量子計算硬件，從而進行學習和創新。我們將介紹最近的進展，將量子處理器的運行性能提高 10 倍以上，使研究人員能夠在運行演算法時最大限度地提高效率。最後，作為了解量子堆疊上下長期經典運算資源需求的第一步，以便為客戶提供生產級混合量子基礎設施，我們將描述我們最近的工作，以使 NVIDIA 的 CUDA-Q 框架適用於 Braket 上的所有量子設備。

Eric Kessler,Sr. Manager Applied Science, Amazon Braket, AWS

Quantum computers are one day expected to solve problems out of reach of even the most powerful classical processors. But these computers won't operate in isolation. Instead, quantum computers will form an integral part of high performance computing architectures that transform some of today’s most challenging workflows, such as drug discovery, portfolio optimization, and many others. Learn how AWS and NVIDIA are collaborating to bring this shared vision to life through AWS’ quantum service, Amazon Braket, and NVIDIA’s CUDA-Q platform. By providing the tools needed by customers, wherever they are in their quantum journey, they can learn and innovate with easy access to a diverse range of quantum computing hardware on Braket. We'll describe recent progress to improve the runtime performance on quantum processors by over 10-fold, enabling researchers to maximize efficiency when running their algorithms. Finally, as a first step toward understanding the longer-term classical compute resource requirements up and down the quantum stack to deliver a production-grade, hybrid quantum infrastructure for customers, we’ll describe our recent work to enable NVIDIA's CUDA-Q framework for all quantum devices on Braket.

Important: Near capacity, highly suggest arriving early. Attendees are let in on a first-come, first-served basis.

Key Takeaways:
Quantum computing is hybrid
Customer needs are evolving quickly
AWS and NVIDIA are ideally placed to address these requirements

重要提示：由於座位數量有限，強烈建議提前到達。與會者將以先到先得的原則入場。

關鍵要點：
量子計算是混合的
客戶需求快速發展
AWS 和 NVIDIA 是滿足這些要求的理想選擇
主題：模擬/建模/設計 - 量子計算
行業細分：所有行業
技術等級：技術 - 中級
目標受眾：研究：非學術
所有目標受眾類型：企業主管
3 月 19 日，星期三
凌晨 4:00 - 凌晨 4:40 中部標準時間

非常感謝您的友好介紹。Nick，大家在後面聽得清楚嗎？非常感謝大家參加這場會議。正如Nick剛才提到的，我的名字是Eric Kessler。我目前在Amazon領導科學團隊，負責AWS（Amazon Web Services）的量子計算服務——Amazon Braket。我們在五年前推出了Amazon Braket，當時的目標是讓量子計算資源像AWS上的其他計算資源和選項一樣易於獲取。在接下來的30到40分鐘裡，我將分享我們這段時間學到的經驗教訓、當前產業的現況，以及我們如何看待量子計算未來在科學計算架構（scientific compute stack）中的角色。我特別高興能參加這次的GTC（GPU Technology Conference），因為我們與NVIDIA在挑戰和機遇的看法上有許多共鳴，尤其是在量子計算架構（quantum computing stack）及其整合方式的問題上。過去一年，我們開始與NVIDIA團隊合作，很高興能分享一些我們的成果。當然，AWS與NVIDIA的關係遠遠超出我們在量子計算領域的小規模合作。我想借此機會，向在場的聽眾簡單介紹一下我們更廣泛的合作關係。

Thank you very much for the kind introduction. Nick, can you all hear okay in the back of the room? So thank you very much for coming to this session. As Nick said, my name is Eric Kessler. I'm running the science team at Amazon, working on the quantum computing service by AWS—Amazon Braket. We launched Amazon Braket five years ago with the goal back then to make quantum computing resources as accessible as other compute resources and compute options on AWS. In the next 30 to 40 minutes, I will talk about what we have learned over that time, where the industries are today, and how we see quantum computing fitting into the scientific compute stack in the future. I'm particularly excited to be here at GTC because we share a lot of views on the challenges and opportunities in this area, around the quantum computing stack and how it fits together with our friends from NVIDIA. We’ve started collaborating with the NVIDIA teams over the last year, and we’re happy to share some of the things that we have done. But of course, the relationship between AWS and NVIDIA goes a lot deeper than our small collaboration in quantum computing. And I wanted to spend a little bit of time with this audience to talk a little about the broader relationship.

AWS與NVIDIA的合作關係，大家可能都知道，已經持續了很長一段時間，我們致力於在雲端提供GPU加速的解決方案（GPU-accelerated solutions）。我們的共同創新涵蓋了從基礎設施軟體到服務的各個層面，提供了圍繞機器學習（machine learning）、高效能計算（high performance computing）和人工智慧（artificial intelligence）等主題的全棧解決方案（full stack solutions）。我特意回顧了我們多年來聯合推出的時間線，事實上，我們的合作至少可以追溯到15年前。

AWS and NVIDIA, as many of you know, have collaborated for a long time to deliver GPU-accelerated solutions in the cloud. Our joint innovations span topics from infrastructure software to services, offering full stack solutions around machine learning, high performance computing, and artificial intelligence. I actually went back to look at the timeline of our joint launches over the years, and as a matter of fact, our collaboration goes back at least 15 years at this point.

2010年，AWS率先將NVIDIA GPUs引入雲端，推出了CG1實例類型（CG1 instance type），即叢集GPU實例（cluster GPU instance）。我特意回顧了2010年那篇相關的部落格文章，至今仍可在網上找到。或許這對在場的聽眾來說有些意外，但我發現有趣的是，這篇文章中完全沒有提到機器學習（machine learning）或人工智慧（AI）。這是我稍後會回到的重點。隨後，我們成為首個將NVIDIA V100 GPUs引入的主要雲端供應商，推出了P3實例（P3 instances）。接著在2020年，我們又推出了搭載A100 Tensor GPUs的P4實例（P4 instances）。就在去年，我們推出了最新的NVIDIA H100和H200 GPUs，並作為首個主要雲端供應商將其投入生產環境。P5實例（P5 instances）提供了驚艷的性能，成本比前一代P4低40%，速度卻快了4倍。我們將探討一個具體例子，展示如何利用H200 GPUs的尖端計算能力，與AstraZeneca、IOQ和NVIDIA合作，加速量子計算（quantum computing）的展示。但我想回顧這段更廣泛合作關係的原因在於，今天，顯而易見地，數千名客戶在AWS上使用NVIDIA GPUs，驅動雲端中最先進的人工智慧和高效能計算（HPC, high performance computing）工作負載，涵蓋醫療生命科學、汽車、媒體娛樂以及公共部門等多個領域。例如，福斯汽車研究集團（Volkswagen Research Group）在AWS上使用NVIDIA GPUs進行空氣動力學模擬（aerodynamic simulation）；牛津大學（University of Oxford）正在進行人工智慧的前沿研究；Netflix則利用AWS本地區域（Local Zones）和NVIDIA GPUs打造低延遲的遠端藝術工作站（remote artistry stations）。這種大規模的市場採用真是令人振奮。

In 2010, AWS was the first to bring NVIDIA GPUs to the cloud with the CG1 instance type, the cluster GPU instance. I actually looked at that blog post from 2010, and you can still find it online. Perhaps surprising for this audience, I found it interesting that there wasn’t a single mention of machine learning or AI in that blog post. That’s a point I’ll come back to. We were also the first major cloud provider to bring the NVIDIA V100 GPUs with our P3 instances. Then, in 2020, we introduced the A100 Tensor GPUs with the P4 instances. Just last year, we launched the latest NVIDIA H100 and H200 GPUs, bringing them to market in production as the first major cloud provider. The P5 instances provide absolutely amazing performance—40% less expensive and 4 times faster than the previous generation of P4. We’ll look at a specific example of how we’re using the state-of-the-art computational power of H200 GPUs to accelerate quantum computing demonstrations in a collaboration with AstraZeneca, IOQ, and NVIDIA. But the broader reason I wanted to revisit this relationship is that today, thousands of customers are obviously using NVIDIA GPUs on AWS, powering the most advanced AI and HPC workloads in the cloud across a variety of sectors, from healthcare and life sciences to automotive, media and entertainment, and the public sector. For example, the Volkswagen Research Group is using NVIDIA GPUs on AWS for aerodynamic simulations; the University of Oxford is conducting cutting-edge research in artificial intelligence; and Netflix is building remote artistry stations using AWS Local Zones and NVIDIA GPUs for low-latency artistry work. This mass market adoption is fantastic.

當然，GPU（Graphics Processing Unit）並非像其他技術一樣一開始就是大眾市場技術。GPU的成功故事實際上始於圖形渲染（graphics rendering）這一小眾應用領域，正如2010年那篇部落格文章所描述的。隨後，科學界意識到，用於加速圖形渲染的相同運算也可以應用於多種科學計算（scientific computing），從流體動力學（fluid dynamics）到基因組學（genomics），當然還有當時新興的人工智慧（artificial intelligence）和機器學習（machine learning）領域。經過多次突破，尤其是大型語言模型（large language models）的出現，GPU獲得了如今驚人的大眾市場採用。我為什麼要講這些？或許你們已經很熟悉這些內容。我講這個故事，是因為我認為量子電腦（quantum computers）在概念上將遵循相似的發展軌跡。早期量子設備的首批使用案例將是物理模型的渲染（rendering of physics models）。我這裡使用的「渲染」一詞當然帶點開玩笑的意思，但實際上確實很相似。在這些首批使用案例中，你會試圖模擬量子力學模型（quantum mechanical models），這些模型可以自然地在這些設備上實現。你定義模型，然後量子電腦生成一些可觀測的東西，比如測量結果，或者提供一種不同的方式來探索你所定義的模型。例如，當你觀察類似我們在Amazon Braket上提供的QRO設備這樣的模擬量子電腦（analog quantum computers），這正是你會做的事情。你以特定方式定義系統參數，使其模擬你感興趣的模型，也就是你感興趣的哈密頓量（Hamiltonian）。然後，你讓系統自然演化，最後觀察結果。這樣你就以不同的方式接觸到你所定義的模型，這正是「渲染」的一種定義。從這些初期的量子模擬（quantum simulations）出發，不難想像其在科學計算中的更廣泛應用，例如高能物理學（high energy physics）、凝聚態理論（condensed matter theory）等等。

Of course, GPUs didn’t start like any other technology. GPUs didn’t begin as a mass market technology. The GPU success story actually started in the niche application area of graphics rendering, just as the 2010 blog post described. Then the scientific community realized that the same operations used to accelerate graphics rendering could be applied to a variety of scientific computing fields, from fluid dynamics to genomics, and of course, the then-emerging fields of artificial intelligence and machine learning. Through multiple breakthroughs—most prominently the emergence of large language models—GPUs achieved the incredible mass market adoption they enjoy today. So why am I telling you this? You probably already know all this. I’m sharing this story because I believe that, conceptually, quantum computers will follow a similar trajectory. The first use cases of early quantum devices will involve the rendering of physics models. I’m using the term "rendering" a bit tongue in cheek, of course, but it’s actually quite similar. In these initial use cases, you try to simulate quantum mechanical models that can be natively implemented on these devices. You define the model, and the quantum computer produces something observable—a measurement or something that gives you a different way to access the model you defined. For example, when you look at analog quantum computers like the QRO device we have on Amazon Braket, that’s exactly what you do. You define the system parameters in a way that mimics a model you’re interested in—the Hamiltonian you’re interested in. Then you let the system evolve and observe it at the end. This gives you a different way to access the model you’ve defined, which is a kind of rendering. From these initial quantum simulations, it’s not a far step to envision broader applications in scientific computing, such as in high energy physics, condensed matter theory, and so on.

當然，我們聚集在此的原因，是因為我們都希望量子計算能實現大眾市場採用（mass market adoption），並對各行各業產生大規模影響。但哪個會成為突破性的應用（breakout application）？我認為我們還不知道。可能是材料科學（materials）、化學（chemistry），也可能是人工智慧優化（AI optimization）。我們對此當然各有看法，但答案仍然很大程度上是未知的。這正是我們希望達到的目標。但我也不想貶低那些基礎應用，至少從我的角度來看是如此。以我的科學背景來說，我認為我們不應低估或忽視量子電腦（quantum computers）成熟後，將為我們理解和操控自然帶來的能力上的根本轉變。量子電腦在短期內首先是科學設備（scientific devices），有點像望遠鏡（telescope）。望遠鏡讓我們能夠在大尺度上探索自然，理解行星和銀河的運動。同樣地，量子電腦將讓我們以全新的方式理解微觀世界的物理（microscopic nature）。這也是為什麼，我認為無論需要多長時間，我們都必須追求這項技術。這是一種全新的計算技術（compute technology），不僅僅是速度更快，而是概念上的不同，是理解我們所處世界的新工具。我相信我不是唯一有這種看法的人。在美國的超級計算資源分配（supercomputing allocation）中，30%正是用於我提到的這類量子物理相關研究。由此，全球已有超過30個國家級量子計畫（national quantum initiatives）啟動，還有無數區域性計畫，推動量子計算（quantum computing）的研究與發展。這些計畫中，科學使用案例（scientific use cases）位居核心，其中許多正與Amazon Braket和AWS合作，為當地量子計算研究社群提供量子計算和傳統計算資源（classical computing resources），通過Amazon Braket讓他們使用量子電腦。

Of course, the reason we’re all here is that we hope there will be mass market adoption and large-scale impact across a variety of industries. But which one will be the breakout application? I don’t think we know yet. It could be materials, chemistry, or AI optimization. We all have our perspectives on this, of course, but the answer remains largely open. This is where we want to end up. But I also don’t want to downplay the applications at the foundational level, at least not from my perspective. Coming from a scientific background, I think we shouldn’t underestimate or underrepresent the fundamental shift in our ability to understand and manipulate nature that quantum computers will provide once they mature. Quantum computers, first and foremost, are scientific devices in the near term, a bit like a telescope. Just as a telescope allowed us to explore nature at a macroscopic scale and understand how planets and galaxies move, quantum computers will enable us to understand the microscopic nature and physics at the microscopic scale in a completely new way. That’s why, in my opinion, we must pursue this technology, no matter how long it takes. This is a fundamentally new compute technology—not just a little faster, but conceptually different, a new tool to understand the world we live in, ultimately. And I don’t think I’m the only one with this perspective. When you look at supercomputing allocation in the US, 30% of it is spent on exactly these types of quantum physics-related efforts I’ve mentioned in these foundational areas. As a result, over 30 national quantum initiatives have been launched by governments around the world, along with countless more regional initiatives, to drive research and development in quantum computing. The scientific use cases are front and center in these initiatives, and many of them are collaborating with Amazon Braket and AWS to provide quantum computing and classical computing resources to local communities of quantum computing researchers, giving them access to quantum computers through Amazon Braket.

同時，AWS提供可擴展、彈性的傳統計算資源（classical compute），在一個可擴展且安全的研發環境中，並具備AWS提供的所有管理功能，例如使用者存取管理（user access management）和預算控制（budget controls）。雖然我對量子電腦（quantum computers）將推動的科學創新感到非常興奮，但我們最終相信，隨著時間推移，量子電腦將對多種技術和產業產生廣泛影響。在未來，我們也相信量子電腦將成為AWS基礎計算結構（compute fabric）的一部分，我們很可能會在AWS資料中心（data centers）中運作這些設備。從某種意義上說，量子計算的長期挑戰是讓量子計算變得「平凡」。我們希望量子電腦最終只是AWS上的另一種加速器（accelerator），與GPU（Graphics Processing Unit）、AI/ML加速器（AI/ML accelerators）、FPGA（Field-Programmable Gate Array）等並存。但當然，這是一個漫長的旅程，目前技術尚未達到這個階段。許多客戶現在問我們：我們能做什麼？當前的技術能做到什麼？量子計算是威脅還是機會？我們應該何時參與？如果要參與，該如何參與？摩根大通（JP Morgan Chase）的Marco Pistoria說過：「如果一家公司現在對市場不採取任何行動，當量子優勢（quantum advantage）來臨時，可能就太晚了。」我很喜歡Marco的這句話，但顯然「任何行動」這個詞需要一些具體內容。如果不是「任何行動」，那麼客戶在此階段應該做些什麼呢？事實上，我認為沒有統一的答案，這取決於每家公司的獨特情況、目標以及所在產業。我們在思考企業客戶的不同參與方式時，試圖沿著客戶在產業中典型的研發旅程（R&D journey）來結構化。從左邊的發現使用案例（use cases）和普遍理解技術開始，到親手實作、研究演算法（algorithms）並在量子電腦上進行基準測試（benchmarking），再到不僅實作量子演算法，還推動技術前沿（state of the art），開發新的客製化演算法，發表研究成果等等，一直到右邊，為這些演算法最終在生產環境中運行奠定基礎並預測其前景。我們的客戶在這些階段中都有參與。

At the same time, AWS provides scalable, elastic classical compute in a scalable and secure research environment, with all the administrative features like user access management and budget controls that AWS offers. As excited as I am about the scientific innovation that quantum computers will drive, we ultimately believe that, in the fullness of time, quantum computers will have a broad impact on a variety of technologies and industries. In that future world, we also believe quantum computers will become part of the fundamental compute fabric of AWS, and we’ll likely operate these devices in AWS data centers. In some sense, the long-term challenge for quantum computing is to make it boring. We want quantum computers to simply be another accelerator on AWS, alongside GPUs, AI/ML accelerators, FPGAs, and so on. But of course, that’s a long journey, and the technology isn’t there today. Many of our customers are asking us now: What can we do? What’s possible with the technology today? Is quantum computing a threat or an opportunity? When should we engage? And if so, how should we engage? Marco Pistoria from JP Morgan Chase said, “If a company doesn’t do anything about the market right now, when quantum advantage comes along, it might be too late.” I love this quote from Marco, but obviously, the word “anything” leaves some work to be done. If not anything, then what’s the something customers should do at this stage of the industry? The truth is, I don’t think there’s a single answer to this. It depends on each company’s unique situation, ambitions, and the industry they’re in. When we think about the different types of engagement from enterprise customers, we try to structure it along the typical R&D journey of a customer in industry—from discovering use cases and generally understanding the technology on one end, to getting hands-on, implementing and researching algorithms, and benchmarking them on quantum computers, to not only implementing quantum algorithms but also pushing the state of the art, developing custom new algorithms, publishing results, and so on, all the way to laying the groundwork to envision and predict what it will look like for these algorithms to eventually run in production. We have customers engaging at all these stages.

我們已經談過JP Morgan Chase，對吧？他們擁有一個相當規模的研究組織，致力於推動原創且優秀的研究，特別是在量子金融（Quantum Finance）這個新興領域的成果。你會發現他們在這個領域中更偏向領先的位置。但我們也有其他客戶，他們可能會在探索階段（Discovery Session）參與，覺得這項技術對他們來說還稍微早了一些，於是退回到觀察的角色，或許一年後再重新參與。這完全沒問題，對吧？我們希望無論客戶處於旅程的哪個階段，都能提供支持，不管他們是想成為思想領袖（Thought Leaders），還是僅僅想降低自身的風險（De-risk）。因此，作為第一步，我們最近推出了一個客戶諮詢計畫，稱為「量子啟航」（Quantum Embark）。這個計畫旨在幫助客戶了解這項技術是否以及何時會影響他們的業務，並協助他們找出與其產業最相關的應用案例（Use Cases）。這是一個為期12週的量子準備計畫（Quantum Readiness Program），客戶無需事先承諾。我們提供了三個模組，可以整套參與，也可以單獨選擇，包括應用案例探索（Use Case Discovery）、培訓（Training）和賦能（Enablement）。通常，我們會進行一次深入研究（Deep Dive），選取一組已發表的成果或資料，試著重現這些成果，並一步步展示實現最先進研究成果所需的步驟。這個計畫的核心是讓量子計算（Quantum Computing）變得可操作，並對產業現狀保持高度透明，幫助客戶判斷這項技術何時會影響他們的業務，同時也為他們提供足夠的資訊，讓他們能向內部利益相關者（Stakeholders）說明情況。無論是「我們需要加倍努力，朝卓越中心（Center of Excellence）邁進，採取下一步」，還是「現在還太早，我們先退一步觀望」，這兩種結果我們都希望透過提供的資訊來支持。

英文原文

We've already spoken about JP Morgan Chase, right? So they have a sizable research organization that is pushing original, excellent research into novel output in quantum finance. You would locate them more on the right. But we have other customers that engage, maybe in the discovery session, see that the technology is still a bit too early for us, go back to a monitoring position and re-engage maybe a year from now. And that's totally fine, right? Like we want to support customers no matter where in the journey they are, whether they want to be thought leaders or whether they want to just de-risk themselves. So for that first step, to apply that first step, we recently launched a customer advisory program, which we call Quantum Embark. And that is really about helping customers understand if and when the technology will impact their business and help them understand the most relevant use cases in their vertical. It's a 12-week quantum readiness program with no upfront commitment. We have these three modules that you can take as a package or pick and choose a la carte around use case discovery, training, and enablement. And then typically, we have a deep dive where we take a set of published results or something and try to reproduce that and follow along the steps of what it takes to implement a state-of-the-art research result. So this program is really all about making quantum computing actionable and being very transparent about the state of the industry, helping customers understand if and when the technology will impact their business. But also arm them with the information that they need to inform their own stakeholders internally, whether it is, "Hey, we need to double down. We want to go to the center of excellence, take the next step," or whether it is "It's too early. We take a step back." Both of these are outcomes that we want to support with information that we can provide.

接著，下一步就是要親手實踐，對吧？學習如何編程量子電腦（Quantum Computers），實作演算法（Algorithms），在模擬器（Simulators）上進行基準測試（Benchmarking），當然還有在真正的量子電腦上測試。這個階段的重點在於發現機會並降低風險（Reducing Risk）。我說這項技術還處於非常早期的階段，我想這不是什麼秘密。它是一項早期技術（Early-Stage Technology），有多種競爭的範式（Paradigms）、競爭的技術平台（Technology Platforms）和編程模型（Programming Models）。因此，這很大程度上是關於如何降低風險，避免技術鎖定（Technology Lock-in），同時也減少實驗的成本（Cost of Experimentation）。

英文原文

And then in the next step, it's all about getting hands-on, right? Learning how to program quantum computers, implementing algorithms, benchmarking them on simulators, but also on actual quantum computers. Of course, that stage is all about spotting opportunities and reducing risk. Right? I don’t think I’m telling any secrets when I say that the technology is still very early. It’s an early-stage technology with multiple competing paradigms, multiple competing technology platforms, and programming models. So it’s very much about trying to reduce the risk or avoid technology lock-in as well as reducing the cost of experimentation, right?

我們希望賦能我們的團隊，讓他們能夠嘗試不同的技術和方法，並加速學習當前存在及新興的各種技術。正因如此，我們推出了「亞馬遜量子服務」（Amazon Braket）。正如我所說，「亞馬遜量子服務」是我们在「亞馬遜網路服務」（AWS）上的量子計算服務（Quantum Computing Service），目標是讓量子計算變得更容易接觸。我們提供了多種量子硬體（Quantum Hardware）的選擇，就像AWS一貫的模式，一切都是按使用付費（Pay-as-you-go）。客戶無需事先承諾，也不需要簽署昂貴的長期合約。我們致力於提供一致的使用者體驗（Consistent User Experience），我特別強調「一致」這個詞，因為我們並不想標準化量子電腦（Quantum Computers）。我們非常欣賞每種技術的獨特性及其獨特的能力。我們認為，要真正突破界限，研究人員需要接觸到這些設備的所有不同功能。現在標準化還為時過早，但我們希望以一致的方式展現這些功能，例如一致的應用程式介面（API）、一致的定價模式（Pricing Model），以及整體一致的使用者體驗。

Empowering your teams to try out different technologies, different approaches, and enable faster learning with the different technologies that exist and emerge now. And for that reason, this is what we try to provide with Amazon Braket, right? Amazon Braket, as I said, is our quantum computing service on Amazon, on AWS, with the goal to making quantum computing accessible, we provide a choice of different types of quantum hardware, like AWS in general, everything is paid as you go. There’s no upfront commitment, no need to sign long-term contracts, long yeah the expensive long-term contracts, as well as we’re trying to strive for a consistent user experience. And I’m emphasizing that point consistent. Because we don’t want to standardize right quantum computers. We very much appreciate all the uniqueness and the unique capabilities of every technology. And we think that in order to really push the boundaries, researchers need access to all the different capabilities that these devices have. And it’s way too early to standardize it, but we want to expose these capabilities in a consistent way right with a consistent API, consistent pricing model, consistent user experience, in general.

目前，「亞馬遜量子服務」（Amazon Braket）提供了來自四家不同供應商的量子電腦的存取。這包括「IonQ」的離子阱量子電腦（Ion Trap Quantum Computers），涵蓋「Aria」和「Forte」系統類別。我們還有來自「Rigetti」和「IQM」的超導量子電腦（Superconducting Quantum Computers），以及來自「QuEra」的模擬中性原子量子電腦（Analog Neutral Atom Quantum Computers）。我們已經稍微提到過「QuEra」的設備，它是一種模擬設備，用來模擬特定的哈密頓量（Hamiltonian）。而其他三者則是基於量子電路模型（Quantum Circuit Model）。我們試圖透過我們的產品涵蓋主要的技術和主要範式（Paradigms）。當然，我們會經常更新我們的量子設備組合（Portfolio of Quantum Devices）。今天我特別興奮地宣布，我們在今天稍早推出了「IonQ」的最新設備「Forte Enterprise」，這是「Forte」系統類別中的最新裝置。每週五天，每天提供15小時的存取權限。它擁有36個量子位元（Qubits），我想它的中位雙量子位錯誤率（Two-Qubit Error Rate）大約是0.5%。現場有專家可以糾正我，但我今早最後一次檢查時，實際上是0.4%左右。它支援全對全連接性（All-to-All Connectivity）和量子錯誤緩解（Quantum Error Mitigation）。我們對這台設備感到非常興奮。

And today, Amazon Braket provides access to quantum computers from four different providers. This is IonQ and ion trap quantum computers from the Aria and Forte system classes. We have superconducting quantum computers from Rigetti and IQM as well as analog neutral atom quantum computers from QuEra. So the QuEra device we already spoke a little bit about is an analog device to emulate certain Hamiltonian, whereas the other three are based on the quantum circuit model. So we’re trying to cover the main technologies and the main paradigms with our offerings here. And of course, this frequent new updates to our portfolio of quantum devices. I’m particularly excited to announce today that we launched earlier today, the latest device from IonQ, the Forte Enterprise, the latest device in the Forte and in the Forte system class, with 15 hours per day, access over 5 days a week. 36 qubits, I think it’s roughly 0.5% median two-qubit error rate. We have the experts in the room, so you can correct me. But last minute I checked this morning, it was around 0.4, actually, all-to-all connectivity support for quantum error mitigation. So we’re super excited about this device.

下一步，就是要推動界限，對吧？
Okay so in the next step, it’s now all about pushing the boundaries, right?

我們追求的不僅是實作現有的演算法（Algorithms），而是要推進技術的前沿（State of the Art），針對特定領域的應用進行原創研究（Original Research）。我們思考這一點的方式之一，是透過一個我們推出的計畫，稱為「亞馬遜量子直通計畫」（Amazon Braket Direct）。一般來說，雲端服務（Cloud）傾向於讓使用者自行操作，對吧？它的普遍哲學是推出應用程式介面（APIs），然後讓開發者自由發揮，讓他們實驗、做自己想做的事情。如果開發者需要主動聯繫「亞馬遜網路服務」（AWS），那就表示我們可能做錯了什麼。但量子計算（Quantum Computing）目前的情況並非如此。創新是透過協作（Collaboration）實現的，創新發生在放寬限制（Guardrails）的地方，讓你嘗試不同的東西。因此，我們希望透過這個計畫來認可產業的現狀，讓客戶能夠全力以赴，將這些早期設備（Early-Stage Devices）推向極限。這個計畫提供專屬存取權（Dedicated Access），讓客戶可以獨占使用我們提供的不同量子電腦，特別是針對最複雜的工作負載（Workload），你是唯一有權存取該設備的人。我們也提供了與「亞馬遜量子服務」（Amazon Braket）的科學家以及來自不同硬體供應商（Hardware Providers）的專家，例如「IonQ」或「QuEra」，互動的方式。此外，我們還將其視為一個創新沙盒（Innovation Sandbox），用來加速設備新功能的推出。我們希望讓硬體供應商能在更受控的環境中試驗新功能（New Capabilities）。例如，我們曾與「QuEra」密切合作，為「橡樹嶺國家實驗室」（Oak Ridge National Lab）和「普渡大學」（Purdue）的學者啟用了一組實驗性功能（Experimental Capabilities），幫助他們克服研究上的障礙。這涉及一組功能，像是原子陣列（Atomic Array）上光圈場（Field of Aperture）的大小等細節，這些細節本身不那麼重要，但重點是這些功能尚未完全成熟。它們有許多限制和注意事項，這對已經在「里德伯原子計算」（Rydberg Atom Computing）領域具有專業知識的專家客戶來說完全沒問題。透過與「QuEra」及我們團隊的直接互動，我們幫助他們理解這些複雜之處，但這些功能或許還沒準備好向所有AWS客戶公開。

英文原文

Like advancing, not only implementing existing algorithms, but advancing the state of the art, doing original research into applications in a particular article. And one way that we’re thinking about this is through a program that we launched, that we call Amazon Braket Direct. And you see the cloud, in general, tends to be hands-off, right? The general philosophy is to launch APIs and set your developers free. Let them experiment, let them do what they want. If a developer needs to reach out to AWS it’s a signal that we did something wrong. This is not quite where quantum computing is at today. Innovation happens through collaboration. Innovation happens where the guardrails are loosened, where you try out different things. And so we wanted to acknowledge this state of the industry through this program that really takes allows customers to pull out the brakes and push these early-stage devices to their limits, right? And it has, we provide dedicated access through this program to the different quantum computers that we have for the most complex workload where you only have access to. You are the only person having access to the device. We have ways to engage our scientists at Amazon Braket as well as experts from the different hardware providers, such as IonQ or from QuEra as well. And then we also think about it as an innovation sandbox, as a way to bring out new capabilities on devices faster than before where we want to enable also our hardware providers to try out new capabilities in a more controlled environment. One example where we did that was where we worked closely with QuEra to enable a set of experimental capabilities to unblock researchers from Oak Ridge National Lab and Purdue. So this is a set of features around, you know, how big is the field of aperture on the atomic array that you can implement the details? Not really matter that much, but the point is that these features weren’t quite ready for prime time. There were lots of limitations and caveats and things we needed to be aware of, which is totally fine for expert customers who is already an expert in the field of Rydberg atom computing. And through the direct interaction with QuEra and our team, we were able to help them understand. And all these intricacies, but maybe it wasn’t ready to expose to all of AWS customers.

這些實驗性功能（Experimental Capabilities）讓我們能夠將技術交到這些客戶手中。最終，他們成功模擬了一種特定物理模型中的量子相變（Quantum Phase Transitions），也就是「切斯特重整化模型」（Chest Resettlement Model）。順帶一提，這正是我之前描述的那種物理渲染理念和應用案例（Use Case），也正是「橡樹嶺國家實驗室」（Oak Ridge National Lab）非常感興趣的主題之一。我們還有另一個機制來幫助客戶突破界限，那就是我們所謂的「亞馬遜高級解決方案實驗室」（Amazon Advanced Solutions Lab）。這個實驗室是由「亞馬遜網路服務」（AWS）內涵蓋量子計算（Quantum Computing）、高效能運算（High Performance Computing）、人工智慧（AI）及運籌學（Operations Research）的專家團隊組成。他們與客戶進行定制的研發合作（Custom R&D Collaborations），針對客戶可能遇到的特定領域問題（Domain Problems）打造解決方案。我們的主要目標是幫助客戶了解量子計算的現況，並以協作的方式開發新的量子計算方法、新演算法（Algorithms）、新協議（Protocols）。同時，我們也幫助他們理解，對於同樣的問題，他們今天在AWS上能使用的最先進（State of the Art）甚至超越最先進的傳統方法（Classical Methods）可以做到什麼。因此，我們協助客戶發展知識產權（IP）和專業知識（Know-how），展望量子計算未來的可能性，同時探索今天在AWS上已經能實現的成果。最後，我想稍微談談我們認為未來雲端量子計算工作負載（Quantum Computing Workload）將面臨的一些挑戰，以及我們必須解決的問題。當然，我想大家都明白，最重要的議題是量子硬體（Quantum Hardware）。目前我們還看不到量子生產工作負載（Quantum Production Workloads）的原因，歸根結底是因為硬體的品質、錯誤率（Error Rates）和規模尚未達到我們發明這些已知演算法所需的水平。所以我們非常興奮地看到，去年在不同技術領域中，量子錯誤校正（Quantum Error Correction）取得了許多突破。我們特別為大約一個月前發表的成果感到興奮，這是我們在《自然》（Nature）期刊中強調的第一款晶片，命名為「貓量子位」（Cat Qubits），也被稱為「雙聲子」（Bosonic）或「振盪器量子位」（Oscillator Qubits）。這個名稱因此而來，它實現了更高效的硬體實現（Hardware-Efficient Implementation），採用了串聯的「貓量子位編碼」（Cat Qubit Codes）。這種方式比「表面編碼」（Surface Code）高出多達10倍的實現效率。

英文原文

So experimental capabilities allowed us to bring this in the hands of these customers. And ultimately, they were able to simulate quantum phase transitions in a particular type of physics model, the Chest Resettlement Model, which is, by the way, exactly this kind of physics rendering idea and use case that I was describing before and exactly one of the topics that Oak Ridge is very interested. Another mechanism that we have to enable customers to push the boundaries is what we call the Amazon Advanced Solutions Lab. The Amazon Advanced Solutions Lab is a team of specialists across AWS from quantum computing, high performance computing, AI, operations research that engages in custom R&D collaborations with customers in order to build solutions for specific domain problems that the customers might have, right? And so we want to help customers primarily understand where quantum computing is at and deliver, in a collaborative way, new approaches to quantum computing, new algorithms, new protocols, but also help them understand what they can do on these very same problems with state-of-the-art and beyond state-of-the-art classical methods on AWS today, right? So helping customers develop IP and know-how, trying to envision what is possible with quantum computing in the future, as well as what we can do already today with AWS. And so finally, I wanna talk a little bit about what some of the challenges are, that we think future quantum computing workload in the cloud will face and that we’ll have to address. And of course, I think everybody is aware of that. The most important topic is the quantum hardware, right? Ultimately, the reason why we don’t see quantum production workloads is because the hardware is not at the quality, at the error rates, at the scale that we would need to invent these algorithms that we know. So we’re very excited to have seen last year, a lot of breakthroughs in the field of quantum error corrections on different technologies. And we’re especially excited about our results that we published about a month ago on highly efficient error correction with our first chip that we highlighted in this Nature publication that we call a slot. It’s based on cat qubits or bosonic or oscillator qubits. Hence, the name, and allows for more efficient hardware-efficient implementation of concatenated cat qubit codes, right? That allows for up to 10 times more efficient implementation than the surface code.

是的，量子硬體（Quantum Hardware）固然重要，但它並不是唯一重要的因素。當我們審視一個典型的高效能運算（HPC, High Performance Computing）類型工作負載（Workload）時，問題在於量子計算（Quantum Computing）會在這個技術堆疊（Stack）中位於何處？這是一個示意圖，可能是計算化學（Computational Chemistry）工作負載，或其他高效能運算類型的計算。頂層是所有周邊運算，包括使用者介面（User Interface）、使用者互動（User Interactions）、應用程式介面互動（API Interactions）、資料輸入與輸出（Data In and Data Out），以及負責協調不同運算部件的排程器（Scheduler）。但真正的核心在底層，對吧？這裡是進行大量數值運算（Numerical Heavy Lifting）的地方。通常，每個工作負載都有不同的步驟組成，有些步驟可能相互依賴，具有線性依賴性（Linear Dependency），有些則可以並行處理（Parallelized）。因此，這裡會出現不同的運算模式（Compute Profiles）。這也是我們在「亞馬遜網路服務」（AWS）中通常試圖優化的地方，為這些高效能運算工作負載提供可擴展性（Scalability）、彈性（Elasticity）和協調（Orchestration）。顯然，量子計算不會取代這個圖表中的所有部分。相反，我們相信量子計算最終將加速這個運算流程中非常特定的部分（Specific Pieces）。但其餘的傳統運算（Classical Compute）並不會消失。因此，從某種意義上說，所有的量子計算在本質上都是混合的（Hybrid）。它將始終嵌入在一個經典-量子處理流程（Classical-Quantum Processing Pipeline）中，由量子電腦加速特定的運算部分。這是我們和「NVIDIA」團隊開始思考未來架構願景（Architectural Vision）的動機之一。雖然距離實現還有很長的路要走，但我們希望思考這些想法帶來的需求是什麼？我們如何為雲端中嵌入可擴展傳統運算流程的量子處理器（QPUs, Quantum Processing Units）建立一個架構願景？作為第一步，我們開始與「NVIDIA」團隊合作，將「CUDA Quantum」與「亞馬遜量子服務」（Amazon Braket）整合。「CUDA Quantum」是許多人熟知的由「NVIDIA」開發的程式設計框架（Programming Framework），它將量子處理器、圖形處理器（GPU, Graphics Processing Unit）和中央處理器（CPU, Central Processing Unit）的資源整合在單一開發框架中。我們去年一同合作推出了一個整合版本，讓它在「亞馬遜量子服務」上可用，提供基於「NVIDIA」實例（GPU Instances）的可擴展傳統模擬器（Classical Simulators），同時讓使用者能夠存取我們在「亞馬遜量子服務」上提供的量子電腦。這是我們邁出的第一步。但我們也希望開始研究真實世界的例子，並更深入了解這些工作負載的運算需求（Computational Requirements）。我們很高興能與「阿斯特捷利康」（AstraZeneca）展開合作。

英文原文

Yeah so as important as quantum hardware is, it is not the only thing that is important. Right? Because when we look at a typical HPC type workload, the question is, where will quantum computing sit in that stack? So this is kind of a schematic of, you know, maybe a computational chemistry workload, or some other HPC type computation. We have all the peripheral computations at the top from the user interface, the user interactions, API interactions, data in and data out, the scheduler to orchestrate the different compute pieces. But then the music plays at the bottom here, right? Where the numerical heavy lifting happens. And typically, every workload has some components of different steps. Some of them might be interdependent. They have a linear dependency, some of them can be parallelized out. So there’s different compute profiles that happen here. And you know this is what we try to optimize generally at AWS to provide the scalability, elasticity, and orchestration for these types of HPC workloads. And quantum computing is obviously not gonna replace all of this diagram. Instead, we believe that quantum computing will eventually accelerate very specific pieces of this computational flow. But the rest of the classical compute will not go away, right? So in a sense, all quantum computing is hybrid in this sense. Right? So it will always be embedded in a kind of classical-quantum processing pipeline where quantum computers accelerate very specific pieces of the computation. Right? And so this is one of the topics, this perspective, motivated us and the NVIDIA teams to start thinking about what is the architectural vision about this future, right? And you know obviously, there’s still a long time to go, but we want to think about what requirements arise from those ideas, right? And how do we build an architectural vision for QPUs embedded in scalable classical compute pipelines in the cloud? Yeah, as a first step, we started to work with NVIDIA team to integrate CUDA Quantum and Amazon Braket. So CUDA Q, obviously, as many of you know, is a programming framework that is developed by NVIDIA that combines QPU, GPU, and CPU resources in a single developer framework. And we worked together last year to launch an integration, make it available on Amazon Braket to provide, you know, scalable classical simulators based on NVIDIA instances, GPU instances in the cloud, but also make accessible the quantum computers that we have available on Amazon Braket. So that is that was the first step that we took. But we also wanted to start looking into real-world examples and trying to understand more the computational requirements around these workloads. And we were very happy to start working in a collaboration together with AstraZeneca.

我們與「IonQ」和「NVIDIA」的朋友們合作，打造了一個概念驗證展示（Proof of Concept Demonstration），用來模擬帶有過渡金屬催化劑（Transition Metal Catalysts）的「鈴木交叉偶聯反應」（Suzuki Cross-Coupling Reaction）。這聽起來有很多專業術語，但簡單來說，這是一種與醫藥科學（Medicinal Science）和藥物開發（Drug Development）高度相關的反應類型。我們採用了一種特殊的量子演算法來實現它，叫做「量子-經典-量子蒙特卡羅」（QCQMC, Quantum Classical Quantum Monte Carlo），這個名字確實有點拗口。所以我就簡稱它為「QCQMC」，用來估算這個反應的能量障礙（Reaction Barrier）。我們在下一頁投影片中會再多談談這個演算法的運作方式。我們使用了「Q#」語言來進行這個展示，整個流程（Pipeline）都是用「Q#」編程的，然後在「亞馬遜量子服務」（Amazon Braket）上使用「IonQ Forte」量子電腦，再利用「AWS平行叢集」（AWS Parallel Cluster）上的「NVIDIA H200」圖形處理器（GPUs）進行可擴展的後處理（Scalable Post-Processing），來模擬這個反應。「QCQMC」是經典演算法「量子蒙特卡羅」（QMC, Quantum Monte Carlo）的一個量子變體。儘管「QMC」名字裡有個「Q」，但它其實是一個完全經典的演算法（Fully Classical Algorithm）。我這裡有一個非常簡化的示意圖來說明它的不同組成部分，你可以看到，這又回到了我之前展示的那個圖表。因為這個過程中有不同的運算步驟，比如根據分子計算活性空間（Active Space）。這個活性空間可能是已知的，你可以直接從現有資料中取得，但原則上，你可以用「密度泛函理論」（DFT, Density Functional Theory）或「哈特里-福克方法」（Hartree-Fock）來計算這個活性空間。接著，你需要計算試驗波函數（Trial Wave Function），就像一個初始猜測（Ansatz）。然後，你試著通過虛時間演化（Imaginary Time Evolution）來演化這個試驗波函數的特定分解（Decomposition）。細節不是那麼重要，但這一步會變得非常需要運算資源。事實上，如果你想模擬高度相關的材料或系統（Highly Correlated Systems），準備一個有效的試驗波函數在經典電腦上會變得越來越困難。因此，「QCQMC」演算法的想法是，我們不需要在經典電腦上計算這個試驗波函數，而是可以在量子電腦上直接準備它。因為我們知道，量子電腦可以有效創建具有糾纏（Entanglement）的狀態，也就是這種高度相關的狀態，然後我們只需測量它們。舉個例子，這就像在「QMC」演算法中，我們挑出了一個小步驟，而不是計算它，我們直接在量子電腦上準備並測量它，而流程中的其他部分幾乎保持不變。這正是我們所做的。我們使用了「IonQ Forte」來為三個不同的分子準備這個試驗狀態（Trial State），運行了大約60,000次單次測量電路（Single-Shot Match Case Circuits）。在量子計算完成後，我們使用了「NVIDIA H200 Tensor Core GPUs」（這裡原文有誤，已更正），在「AWS平行叢集」上的叢集中進行處理，每個分子使用40個「P5」實例，每個實例配備多個圖形處理器。這是一個相當大的運算規模，展示了我們如何在雲端中擴展運算（Scale Out）。目前這項工作的結果仍在等待中。

英文原文

And our friends from IonQ and NVIDIA to build a proof of concept demonstration, to model Suzuki cross-coupling reaction with transition metal catalysts. So there’s a lot of big words, but it’s essentially, it’s a reaction type that is very relevant for medicinal science and drug development. And the way that we wanted to implement it was with a special type of quantum algorithm, which is QCQMC, Quantum Classical Quantum Monte Carlo, the pretty mouthful. So I’ll stick with QCQMC to estimate this reaction barrier. And we’re gonna talk on the next slide a little bit more what this algorithm how this algorithm works. And we did this demonstration with Q dot Q, so the entire pipeline is programmed as in Q dot Q and then using IonQ Forte on Amazon Braket, and then using post-processing, scalable post-processing on NVIDIA H200 GPUs on AWS Parallel Cluster to simulate this reaction. So QCQMC is a quantum variant of a classical algorithm, so QMC, Quantum Monte Carlo, despite the Q in a name, is actually a fully classical algorithm. I have a very schematic view on the different components, but you see, it’s kind of goes back to this graph that I showed before, right? Because you have different compute steps in this process from, you know, calculating the active space, depending on the molecule. This might be known. You might be able to pull this off the shelf. But in principle, you can do DFT or Hartree-Fock to calculate this active space, then you compute the trial wave function. It’s like an ansatz. And then you try to evolve that particular decomposition of the trial wave function with imaginary time evolution. Details don’t matter so much, but this becomes very compute intensive. And as a matter of fact, this creation of a trial wave function, if you want to simulate highly correlated materials, highly correlated systems, preparing a trial function that is effective, becomes harder and harder on classical computers. So the idea of the QCQMC algorithm is, well, we don’t need to compute this trial wave function. We can actually just prepare it on a quantum computer, because we know we can efficiently create states with entanglement, the type of highly correlated state on a quantum computer. And we just measure them. Right? So there’s an example. There’s like one small step in this QMC algorithm that we kind of carve out. And instead of calculating it, you just prepare on a computer, measure it, and then everything in the pipeline stays pretty much the same. And yeah so this is exactly what we did. So we used IonQ 41 to prepare this trial state for three different molecules, running roughly 60,000 single-shot match case circuits. And then after this quantum computation was complete, we used, can really H, so it should actually be H200 Tensor Core GPUs. This is a typo here, in a cluster on AWS using AWS Parallel Cluster, 40 P5 instances with GPUs, per instance, per molecule, so pretty sizable. Computation illustrates how we need to scale out in the cloud. And so the results of this work are still pending.

我們仍在努力準備發表這些結果，所以我不想透露太多，但我還是想快速分享一些初步結果（Preliminary Results）。我認為其中一個重要的結論可能是，整個流程（Pipeline）運作順利，我們成功讓這個演算法收斂（Converge）。據我們所知，這是迄今為止最大規模的「量子-經典-量子蒙特卡羅」（QCQMC, Quantum Classical Quantum Monte Carlo）展示，結合了「亞馬遜量子服務」（Amazon Braket）上的量子處理器（QPU, Quantum Processing Unit）資源，以及「NVIDIA」在「亞馬遜網路服務」（AWS）上提供的可擴展傳統運算（Scalable Classical Compute）。事實上，這個後處理步驟（Post-Processing Step）之前被認為是非常困難的。抱歉，我指的是流程末端的這部分。之前人們認為這是一個嚴重的瓶頸（Bottleneck），因為雖然它不像指數級增長那樣嚴重，但它的擴展性（Scaling）很差，且需求非常大。透過一些演算法上的進展，「IonQ」團隊成功降低了這部分的估計需求，再加上圖形處理器（GPUs, Graphics Processing Units）的加速，我們將這個傳統後處理的時間縮減了好幾個數量級（Orders of Magnitude）。不過，完整的結果還請大家留意即將發表的論文，我很樂意在論文發表後再多聊聊這個話題。但我想這張投影片的主要訊息大家應該已經明白了：未來的量子計算工作負載（Quantum Computing Workloads）將嵌入在可擴展且彈性的高效能運算處理流程（HPC Processing Pipelines）中。從某種意義上說，單獨的量子計算可能並不存在，總會與可擴展的傳統運算相結合。這是我們需要思考的挑戰之一：如何賦能使用者運行這些大規模展示（Large-Scale Demonstrations）？使用者體驗（User Experience）該如何設計？我們如何讓這一切變得無縫（Seamless）？我們很興奮能與您以及我們的合作夥伴「NVIDIA」、「IonQ」、「阿斯特捷利康」（AstraZeneca）一起繼續研究這個主題。是的，我想再提供幾個參考資料。我提到了一些內容，比如「亞馬遜高級解決方案實驗室」（Amazon Advanced Solutions Lab）、我們與「CUDA Quantum」的整合、「量子啟航」（Quantum Embark）作為在「亞馬遜量子服務」上接觸量子計算的第一步，以及底部的「亞馬遜量子直通計畫」（Amazon Braket Direct）。我沒有提到我們的數位批次（Digital Batch），但這些都是相關的參考。

英文原文

We are still working on publishing the results. So I don’t wanna take away too much, but I still wanted to give a quick view on some preliminary results. And I think one of the take-home messages, maybe, is that, you know, the full entry and pipeline worked, were able to bring this algorithm to converge and to our knowledge. This is the largest implementation of or the largest demonstration of QCQMC to date combining QPU resources on Amazon Braket with scalable classical compute from NVIDIA on AWS and as a matter of fact, this post-processing step was previously thought to be quite prohibited, right? Although it is, I’m sorry, this just this piece at the end there. Right? The second part, because previously, it was thought that this is a severe bottleneck, because while it’s not scaling exponentially, it had a very poor scaling and large requirements. And through some algorithm advances, the IonQ team was able to bring that estimate down. And then plus the acceleration with GPUs were able to reduce the time of this classical post-processing by several orders of magnitude. But again, for the full results, please keep an eye open for me in the paper lands. And then I’m happy to talk more about that. But I guess you know the main messages, I guess you have figured out at this, is that I want to make with this slide is that future quantum computing workloads will be embedded in scalable and elastic HPC processing pipelines. In a sense, the notion of standalone quantum computing probably doesn’t exist, but there’s always some kind of combination of scalable classical compute. And this is one of the challenges that we need to think about how we empower users to run those large-scale demonstrations. What is the user experience around it? How can we make that a seamless experience? And we’re excited to, you know, work with you and our collaborators from NVIDIA, IonQ, AstraZeneca to continue working on this topic. Yes, it was that I just wanted to give you a couple of references. So some of the things I spoke about the Amazon Advanced Solutions Lab, our integration with CUDA Quantum, Quantum Embark as the first step in engaging in quantum computing at Amazon Braket, Direct on the bottom. Here. I did not mention our Digital Batch.

這是一個在「亞馬遜網路服務」（AWS）上提供的量子計算（Quantum Computing）認證課程（Certified Learning），您可以參加。好了，就到這裡。謝謝大家。我很樂意回答幾個問題。好的，謝謝你，Eric。事實上，我們還有幾分鐘時間，可以回答幾個問題，如果有人有問題的話，請舉手，有人會拿麥克風給您測試一下。是的，我有一個具體問題，關於AWS基本上是在簡化量子計算的存取這件事。AWS如何過濾不良行為者（Bad Actors）？有沒有進行一些盡職調查（Due Diligence）？能否詳細說明一下？你說的不良行為者是指什麼？例如犯罪分子，如果犯罪分子有興趣破解加密（Encryption）的話？哦，我明白了，你是指我們的服務。是的，你知道，我想說，破解「RSA」加密的應用案例（Use Case），可能是最困難的一個，需要最先進的量子電腦，我認為這還是很遙遠的未來。這可能還不是我們目前特別關注的問題。當然，「亞馬遜量子服務」（Amazon Braket）只是AWS的另一項服務，我們對它的安全性（Security）和控制措施（Controls）與其他服務是一樣的。但現在，我認為這個應用案例對量子硬體（Quantum Hardware）的要求太高了，所以我們並不太擔心。一般來說，是的，「肖爾演算法」（Shor’s Algorithm）原則上可以用大量的量子位元（Qubits）和非常高品質的量子位元來破解「RSA」。不幸的是，我們大多數的網路加密（Internet Encryption）都基於「RSA」。但這並不是說沒有其他加密協議（Encryption Protocols），據我們所知，有些協議即使面對量子電腦也是無法破解的。最近，「美國國家標準與技術研究院」（NIST）發布了第一個「後量子密碼學」（Post-Quantum Cryptography）的標準。許多這些演算法已經在AWS上可用，所以這是一個逐步實現的過程。但我想我們還有時間再回答一個簡單的問題。我很好奇，你們是否計畫擴展對其他硬體（Hardware）的支援，超越目前提供的「IonQ」、「QuEra」等等？當然有這個打算。我們正在與許多供應商（Providers）接觸，可能幾乎涵蓋了這個領域的所有供應商，合作程度各有不同。當時機成熟，且我們雙方都看到機會時，我們絕對會引入其他設備（Devices）和新興技術（Emerging Technologies）。好的，再次感謝你，Eric。我認為這是一個很好的結束，請大家鼓掌。接下來在這個房間裡，我可以告訴大家下一場是什麼。下一小時，我們有來自John Linford的課程，主題是加速端到端公司計算加速工程工作流程（Computational Accelerated Engineering Workflows）。今天下午晚些時候，還會有更多量子相關的課程。如果您喜歡量子計算，我們在週四上午10點有「量子日」（Quantum Day），還有Jensen的爐邊談話（Fireside Chats）。再次感謝你，Eric。希望你在接下來的「GTC」（GPU Technology Conference）中過得愉快。

英文原文

That is a course, certified learning on quantum computing, on AWS that you can take. And yeah with that. Thank you. And I’m happy to answer a couple of questions. Okay, thank you, Eric. In fact, for that, and we have got a few minutes just for a couple of questions if anyone’s got any. So please raise your hand if someone will bring you a microphone testing. Yeah, so I have a question specific to the fact that AWS is essentially trivializing access to quantum computing so how AWS filters, bad actors. Is there some due diligence done? Can you elaborate? What do you mean with bad actors? Like for example, criminals, if the criminals are interested in breaking the encryption something? Oh I see I see your services. Yeah, well you know I think the use case about breaking RSA essentially, I think, is the hardest use case that requires the most advanced quantum computers that I find is deep deep in the future. But it may be a problem that is not quite as much on our radar yet. But of course Amazon Braket is just another AWS services, and we have the same security and controls around Amazon Braket like with any other service. But right now, I think that use case has such high requirements to the quantum hardware that we’re not too concerned. And then generally speaking, yes, it’s true. Shor’s algorithm can, in principle, break RSA with a lot of qubits and with a lot of very high quality qubits. And unfortunately, most of our internet encryption is based on RSA. It’s not that there aren’t encryption protocols that are the best of our knowledge also unbreakable for quantum computers. Right? So the recently NIST released the first standard in post-quantum cryptography. Many of these algorithms are already available on AWS, so it’s a matter of implementing them over time, but yeah that’s got time, I think, for just one more. Fairly quick question. I’m just curious if you are planning to expand your support to other hardware, beyond what you offer. Now, I on Cuban, we get, et cetera. Yeah, absolutely. So you know we’re talking to a lot of providers in the probably we talked to all of the providers in the field in varying degrees of engagement. And when it’s the right time and we mutually see an opportunity, we will absolutely onboard other devices, other technologies as they emerge. Okay. Well once again, thanks so much, Eric. I think that’s a great note to end on a round of applause, please. And so in this room, I can tell you what’s coming up next. At the next hour. We’ve got a session from John Linford on accelerating end-to-end company computational accelerated engineering workflows. And then later in the afternoon, there’s some more quantum sessions, too. And if you like quantum computing, we’ve got that Quantum Day on Thursday at 10:00 AM Jensen’s Fireside Chats. But again, thank you so much, Eric. And I hope you will have a great rest of your GTC.