Jon Salisbury’s Post

How big is the bet from #Microsoft for AI? On november 2023, the #Top500 list spoted in the 3rd place an Azure Cloud supercomputer, a Microsoft supercomputer called Eagle that delivered 561 petaflops of performance, with 1.12 million computing cores, including Nvidia's H100 GPU and Intel's Xeon Platinum 8480C CPU cores. The supercomputer provides access to the #GPT-4 and other generative models hosted on Azure cloud. Runs on top of Ubuntu 22.04. https://www.top500.org/ #GENAI #GPT4 #AzureCloud #AzureSupercomputer #Supercomputer #HPC #AI

Silicon Valley-based GPU cloud company Lambda Labs has launched Nvidia HGX H100 and Quantum-2 InfiniBand Clusters for AI model training. The Lambda 1-Click Clusters are targeted at AI developers that need access to high-end AI hardware, but don't want - or can't afford - to rent them for long periods. - Lambda provides computation to accelerate human progress. We're a team of Deep Learning engineers building the world's best GPU cloud, clusters, servers, and workstations. - The decentralized principles of structuring TON Datacenters Clusters are essential because they form the basis for highly resilient and flexible network scalability. - Datacenter Dynamics helping those who plan, build, operate and supply digital infra, make informed decisions faster. | lambdalabs.com | tonprovider.org | datacenterdynamics.com | #TONProvider #Lambdalabs #DatacenterDynamics #TONEdgeSevices https://lnkd.in/e7shmwG2

AI chip Sets - when giants marry Yesterday, Reuters and the Financial Times broke news that AWS, a a major customer of Nvidia and for sure a lighthouse tower of customers, had shifted an order for GPU from an older to newer chip set (Hopper to Blackwell). This caused much discussion and uncertainty in financial markets. So, both companies had to communicate that the announcement only affects a certain, special project and doesn’t mean a new strategic course for the cloud hyperscaler. Learning: in markets consisting of only a few participants, each and every coughing causes sweaty palms. On a sidenote: "Pricing details for the new Grace Blackwell solutions highlight their premium status. Analysts estimate that a GB200 chip could cost as much as $70,000, according to media reports. That might even be lower than actual pricing. A full NVL72 server rack equipped with the new hardware could cost up to $3 million, indicating a significant investment for companies looking to use the latest AI processors." #ai #nvidia #cloud #aws #gpu #blackwell

Microsoft is making significant strides by integrating AMD's advanced MI300X AI chips into its Azure cloud services... This strategic move is poised to offer a robust alternative to Nvidia's dominant GPUs, addressing the growing demand for high-performance AI hardware. With this integration, Azure aims to enhance its AI capabilities, providing users with more efficient and powerful computing resources. The collaboration between Microsoft and AMD marks a pivotal moment in the AI and cloud computing landscape, highlighting their joint commitment to innovation and technological advancement. In addition to the AMD collaboration, Microsoft is set to unveil its new Cobalt 100 processors. These processors promise substantial performance improvements, further strengthening Azure's infrastructure. The Cobalt 100 processors are designed to handle complex AI workloads, ensuring that Azure remains at the forefront of cloud computing technology. Microsoft's dual approach of incorporating AMD's AI chips and launching its own advanced processors underscores its dedication to providing cutting-edge solutions in the rapidly evolving AI and cloud computing sectors. #Technology #AI #CloudComputing #Innovation #ChangeYourPassword

Read how AI infrastructure at scale powered by Amazon Web Services (AWS) and NVIDIA is driving research and #innovation. Staggering numbers! https://lnkd.in/emDbVbqH

Question: Should you buy or lease your #NVIDIA H100/H200 GPU? >>> Don’t worry, we did the math for you and calculated the breakeven point. Let us know what you think in the comments below.⬇️ Click below for the full article: https://lnkd.in/gNF-a3V5 Those using NVIDIA #H100 GPUs often find themselves faced with a dilemma. Because choosing to rent from AWS or investing in #colocation can have a significant impact on a company’s bottom line. But how can companies make savings here, without impacting performance? In this article, we’ll discuss the cost implications and breakeven point of decisions like these, and we’ll talk about one solution that’s rapidly growing in popularity: the #hybridIT approach. #AI #GPU #ArtificialIntelligence #MachineLearning #BigData #OpenAI #AWS #CloudComputing Chris Hinkle, Nicole Varela, Robert West, MBA, TRG Datacenters, Joel St. Germain, Michael Price

Amazon’s Moonshot Plan to Rival NVIDIA in #AI #Chips The #cloud computing giant won’t dislodge the incumbent anytime soon but is hoping to reduce its reliance on the chipmaker. https://lnkd.in/drpMEnF7

We're committed to enabling an AI-accelerated future in a sustainable and scalable way. With the momentum we've seen this week at #MSIgnite, we're thrilled to be working with the ecosystem to make this vision a reality with new engineering innovations and developments. This includes our close collaboration with Microsoft on Azure Cobalt 100, which leverages the flexibility and power efficiency of Arm Neoverse Compute Subsystems to establish a robust and flexible infrastructure for mission critical applications. Microsoft also announced they will be the first cloud vendor to make instances based on NVIDIA's Grace Blackwell platform available - this includes 72 Arm Neoverse V2 CPU cores to help power the most demanding AI training and inference workloads. Catch up on what you need to know from #MSIgnite in our latest blog: https://lnkd.in/gNJfD4hg

Project DIGITS: NVIDIA's AI powerhouse for desktops, featuring the Grace Blackwell Superchip, 1 petaflop of compute, and 128 GB unified memory. Ideal for AI developers, researchers, and tech innovators. No more reliance on cloud—bring high-performance AI directly to your workspace!

Gavin Baker on this week’s All In Pod said the best GPU cluster that Microsoft has right now is still smaller than Elon’s xAI cluster Colossus (to the best of his knowledge). The thing about the massive Nvidia GPU purchases from major public CSPs like AWS and Azure is that not all of them are configured in dedicated GPU clusters for internal consumption. Baker mentioned the incredible network design challenges involved in building Colossus, where 100,000 H100s have to be synchronized with low latency and fast response. Previously many thought the upper limit was around 30,000 GPUs. But xAI did what even Jensen didn’t think was possible. Right now there’s no clear numbers on dedicated GPU cluster capacities of major AI providers, but my guess is that xAI, Meta, and Google (TPU) are top 3 for now.