Microsoft and NVIDIA proceed to ship on the promise of AI


At NVIDIA GTC, Microsoft and NVIDIA are saying new choices throughout a breadth of answer areas from main AI infrastructure to new platform integrations, and trade breakthroughs. Right this moment’s information expands our long-standing collaboration, which has paved the way in which for revolutionary AI improvements that clients are actually bringing to fruition.

Microsoft and NVIDIA collaborate on Grace Blackwell 200 Superchip for next-generation AI fashions

Microsoft and NVIDIA are bringing the ability of the NVIDIA Grace Blackwell 200 (GB200) Superchip to Microsoft Azure. The GB200 is a brand new processor designed particularly for large-scale generative AI workloads, knowledge processing, and excessive efficiency workloads, that includes up to an enormous 16 TB/s of reminiscence bandwidth and as much as an estimated 45 occasions the inference on trillion parameter fashions relative to the earlier Hopper technology of servers.

Microsoft has labored intently with NVIDIA to make sure their GPUs, together with the GB200, can deal with the most recent giant language fashions (LLMs) skilled on Azure AI infrastructure. These fashions require huge quantities of information and compute to coach and run, and the GB200 will allow Microsoft to assist clients scale these assets to new ranges of efficiency and accuracy.

Microsoft can even deploy an end-to-end AI compute cloth with the just lately introduced NVIDIA Quantum-X800 InfiniBand networking platform. By making the most of its in-network computing capabilities with SHARPv4, and its added assist for FP8 for modern AI methods, NVIDIA Quantum-X800 extends the GB200’s parallel computing duties into huge GPU scale.

Azure will likely be one of many first cloud platforms to ship on GB200-based situations

Microsoft has dedicated to bringing GB200-based situations to Azure to assist clients and Microsoft’s AI companies. The brand new Azure instances-based on the most recent GB200 and NVIDIA Quantum-X800 InfiniBand networking will assist speed up the technology of frontier and foundational fashions for pure language processing, laptop imaginative and prescient, speech recognition, and extra. Azure clients will have the ability to use GB200 Superchip to create and deploy state-of-the-art AI options that may deal with huge quantities of information and complexity, whereas accelerating time to market.

Azure additionally provides a variety of companies to assist clients optimize their AI workloads, similar to Microsoft Azure CycleCloud, Azure Machine Studying, Microsoft Azure AI Studio, Microsoft Azure Synapse Analytics, and Microsoft Azure Arc. These companies present clients with an end-to-end AI platform that may deal with knowledge ingestion, processing, coaching, inference, and deployment throughout hybrid and multi-cloud environments.

Microsoft Azure AI solution stack

Delivering on the promise of AI to clients worldwide

With a strong basis of Azure AI infrastructure that makes use of the most recent NVIDIA GPUs, Microsoft is infusing AI throughout each layer of the expertise stack, serving to clients drive new advantages and productiveness beneficial properties. Now, with greater than 53,000 Azure AI clients, Microsoft supplies entry to the most effective choice of basis and open-source fashions, together with each LLMs and small language fashions (SLMs), all built-in deeply with infrastructure knowledge and instruments on Azure.

The just lately introduced partnership with Mistral AI can be an awesome instance of how Microsoft is enabling main AI innovators with entry to Azure’s cutting-edge AI infrastructure, to speed up the event and deployment of next-generation LLMs. Azure’s rising AI mannequin catalogue provides, greater than 1,600 fashions, letting clients select from the most recent LLMs and SLMs, together with OpenAI, Mistral AI, Meta, Hugging Face, Deci AI, NVIDIA, and Microsoft Analysis. Azure clients can select the most effective mannequin for his or her use case.

“We’re thrilled to embark on this partnership with Microsoft. With Azure’s cutting-edge AI infrastructure, we’re reaching a brand new milestone in our growth propelling our revolutionary analysis and sensible purposes to new clients in all places. Collectively, we’re dedicated to driving impactful progress within the AI trade and delivering unparalleled worth to our clients and companions globally.”

Arthur Mensch, Chief Govt Officer, Mistral AI

Basic availability of Azure NC H100 v5 VM collection, optimized for generative inferencing and high-performance computing

Microsoft additionally introduced the overall availability of Azure NC H100 v5 VM collection, designed for mid-range coaching, inferencing, and excessive efficiency compute (HPC) simulations; it provides excessive efficiency and effectivity.

As generative AI purposes develop at unbelievable velocity, the basic language fashions that empower them will develop additionally to incorporate each SLMs and LLMs. As well as, synthetic slim intelligence (ANI) fashions will proceed to evolve, centered on extra exact predictions fairly than creation of novel knowledge to proceed to reinforce its use instances. Their purposes embody duties similar to picture classification, object detection, and broader pure language processing.

Utilizing the strong capabilities and scalability of Azure, we provide computational instruments that empower organizations of all sizes, no matter their assets. Azure NC H100 v5 VMs is yet one more computational software made typically accessible as we speak that can do exactly that.

The Azure NC H100 v5 VM collection relies on the NVIDIA H100 NVL platform, which provides two lessons of VMs, starting from one to 2 NVIDIA H100 94GB PCIe Tensor Core GPUs linked by NVLink with 600 GB/s of bandwidth. This VM collection helps PCIe Gen5, which supplies the very best communication speeds (128GB/s bi-directional) between the host processor and the GPU. This reduces the latency and overhead of information switch and permits sooner and extra scalable AI and HPC purposes.

The VM collection additionally helps NVIDIA multi-instance GPU (MIG) expertise, enabling clients to partition every GPU into as much as seven situations, offering flexibility and scalability for numerous AI workloads. This VM collection provides as much as 80 Gbps community bandwidth and as much as 8 TB of native NVMe storage on full node VM sizes.

These VMs are perfect for coaching fashions, working inferencing duties, and creating cutting-edge purposes. Be taught extra concerning the Azure NC H100 v5-series.

“Snorkel AI is proud to companion with Microsoft to assist organizations quickly and cost-effectively harness the ability of information and AI. Azure AI infrastructure delivers the efficiency our most demanding ML workloads require plus simplified deployment and streamlined administration options our researchers love. With the brand new Azure NC H100 v5 VM collection powered by NVIDIA H100 NVL GPUs, we’re excited to proceed to can speed up iterative knowledge growth for enterprises and OSS customers alike.”

Paroma Varma, Co-Founder and Head of Analysis, Snorkel AI

Microsoft and NVIDIA ship breakthroughs for healthcare and life sciences

Microsoft is increasing its collaboration with NVIDIA to assist remodel the healthcare and life sciences trade by way of the combination of cloud, AI, and supercomputing.

Through the use of the worldwide scale, safety, and superior computing capabilities of Azure and Azure AI, together with NVIDIA’S DGX Cloud and NVIDIA Clara suite, healthcare suppliers, pharmaceutical and biotechnology corporations, and medical system builders can now quickly speed up innovation throughout the whole scientific analysis to care supply worth chain for the advantage of sufferers worldwide. Be taught extra.

New Omniverse APIs allow clients throughout industries to embed huge graphics and visualization capabilities

Right this moment, NVIDIA’s Omniverse platform for creating 3D purposes will now be accessible as a set of APIs working on Microsoft Azure, enabling clients to embed superior graphics and visualization capabilities into present software program purposes from Microsoft and companion ISVs.

Constructed on OpenUSD, a common knowledge interchange, NVIDIA Omniverse Cloud APIs on Azure do the combination work for patrons, giving them seamless bodily primarily based rendering capabilities on the entrance finish. Demonstrating the worth of those APIs, Microsoft and NVIDIA have been working with Rockwell Automation and Hexagon to indicate how the bodily and digital worlds may be mixed for elevated productiveness and effectivity. Be taught extra.

Microsoft and NVIDIA envision deeper integration of NVIDIA DGX Cloud with Microsoft Material

The 2 corporations are additionally collaborating to carry NVIDIA DGX Cloud compute and Microsoft Material collectively to energy clients’ most demanding knowledge workloads. Because of this NVIDIA’s workload-specific optimized runtimes, LLMs, and machine studying will work seamlessly with Material.

NVIDIA DGX Cloud and Material integration embody extending the capabilities of Material by bringing in NVIDIA DGX Cloud’s giant language mannequin customization to deal with data-intensive use instances like digital twins and climate forecasting with Material OneLake because the underlying knowledge storage. The combination can even present DGX Cloud as an choice for patrons to speed up their Material knowledge science and knowledge engineering workloads. 

Accelerating innovation within the period of AI

For years, Microsoft and NVIDIA have collaborated from {hardware} to programs to VMs, to construct new and revolutionary AI-enabled options to deal with complicated challenges within the cloud. Microsoft will proceed to develop and improve its world infrastructure with essentially the most cutting-edge expertise in each layer of the stack, delivering improved efficiency and scalability for cloud and AI workloads and empowering clients to attain extra throughout industries and domains.

Be part of Microsoft at NVIDIA CTA AI Convention, March 18 by way of 21, at sales space #1108 and attend a session to be taught extra about options on Azure and NVIDIA.

Be taught extra about Microsoft AI options


Leave a comment