AI workloads are scaling sooner than any single infrastructure method can assist — with extra fashions, new agent-driven workloads and surging compute demand driving the necessity for better specialization throughout the stack. To fulfill this want, Microsoft continues to evolve Azure’s infrastructure, together with increasing its AI fleet with AMD’s most superior AI and high-performance computing (HPC) options.
Our method to AI infrastructure is designed to assist the breadth of how AI methods are constructed and run. We carefully work with business innovators like AMD in addition to our personal purpose-built silicon and methods to supply prospects with a complete, open and heterogenous platform to realize one of the best efficiency, price and power effectivity outcomes.
Constructing on our shut collaboration with AMD, Microsoft is bringing AMD’s newest Helios AI platform and next-generation EPYC datacenter processors to Azure. These applied sciences will energy three upcoming Azure choices: HDv2 VMs for knowledge processing, HXv2 VMs for digital design automation (EDA) and ND MI455X v7 VMs for AI inference workloads.
Expanded infrastructure for inference, AI knowledge methods and chip design
Constructed for AI knowledge methods — Azure HDv2
CPU infrastructure is important to the efficiency and effectivity of recent AI methods. AI accelerators depend upon high-density, power-efficient CPU compute to course of knowledge, coordinate workloads and hold pipelines operating at scale. With out this, coaching jobs don’t have sufficient knowledge to be taught from, and brokers don’t have sufficient capability to carry out duties on behalf of shoppers. Azure HDv2 digital machines are one in every of our newest choices designed from the bottom as much as get rid of these bottlenecks and empower huge agentic workload adoption.
Co-designed with AMD, HDv2 VMs broaden Azure’s portfolio of purpose-built options for probably the most demanding CPU workloads from AI prospects, together with knowledge preparation, search, reinforcement studying and agent coordination at scale. That includes practically 500 bodily 6th Gen AMD EPYC CPU cores, 4 terabytes of RAM, 32 terabytes of native NVMe storage and 400 Gb Azure Increase networking, HDv2 VMs are constructed for the workload wants of our most demanding AI prospects.
Optimized for silicon design and technical computing — Azure HXv2
The AI period has created large want and alternative for companies growing the silicon merchandise that energy this infrastructure. Because of this, Azure HX digital machines, launched in partnership with AMD in 2023 and that includes AMD’s distinctive 3D V-cache know-how, have seen vital adoption amongst silicon design companies working to carry extra succesful and environment friendly AI silicon to market. In the present day, we’re saying the following step in our workload optimized journey for these prospects, HXv2.
HXv2 digital machines construct on and prolong the strengths of HX. They each proceed the differentiation Azure affords for RTL simulation workloads by once more using 3D V-cache know-how, whereas providing vital enhancements to single threaded efficiency and reminiscence. HXv2 VMs will function 176 AMD 6th Gen EPYC CPU cores with a clock frequency of greater than 5 GHz, 50% extra addressable cache per core and VM sizes with practically 2 or 4 terabytes of RAM, serving to prospects optimize their workloads to reminiscence wants.
Azure HXv2 can also be designed to assist a broader vary of technical computing workloads together with scientific simulation, engineering evaluation and different distributed reminiscence functions. The considerably elevated per VM and per core efficiency, and the inclusion of 800 Gb InfiniBand, allow large-scale MPI-based simulations and make HXv2 a great match for all kinds of HPC prospects.
AMD, a number one HX-series buyer, highlights this impression immediately:
“Engineering groups are pushing the bounds of simulation, chip design and scientific computing. At AMD, we expertise these calls for firsthand as we design future AMD EPYC CPUs and AMD Intuition GPUs. Azure HX is a vital platform for scaling advanced EDA workloads, and we’re enthusiastic about Azure HXv2, which is designed to ship even better efficiency and scalability. We stay up for persevering with our collaboration with Microsoft as we assist advance infrastructure for the world’s most demanding engineering and scientific workloads.”
— Mark Papermaster, Government Vice President and CTO, AMD
The HXv2 additionally leverages Microsoft’s long-standing collaboration to optimize Synopsys AI-powered EDA options on Azure:
“As AI compute continues to push the bounds of semiconductor design, our collaboration with Microsoft on the Azure HX-series demonstrates a shared imaginative and prescient for enabling prospects to ship next-generation AI methods with precision and scale in accelerated design cycles. These methods have enabled Synopsys prospects to reliably and effectively leverage cloud-based compute, extending EDA workloads past conventional infrastructure constraints to allow them to meet formidable improvement schedules whereas maximizing design high quality and delivering dramatic efficiency positive aspects.”
— Shankar Krishnamoorthy, Chief Product Improvement Officer, Synopsys
Manufacturing-scale AI inference — ND MI455X v7
ND MI455X v7 is designed for the reasoning, search and agentic workloads behind trendy AI providers. Powered by the AMD Helios rackscale answer, it expands Azure’s infrastructure choices for large-scale inference and is designed to ship robust efficiency and effectivity for demanding AI workloads.
Collectively, these new capabilities broaden Azure capabilities whereas giving prospects extra flexibility to decide on the suitable compute for every distinctive AI workflow: from inference, to knowledge methods, to chip design. Buyer selection is a core design precept constructed immediately into Microsoft Azure, and we’re excited to carry AMD’s most superior improvements at manufacturing scale.
To be taught extra about Azure’s high-performance computing and AI infrastructure capabilities, go to Azure.com.
Scott Guthrie is answerable for a set of hyperscale cloud computing options and providers together with Azure, Microsoft’s cloud computing platform, generative AI options, knowledge platforms and data and cybersecurity. These platforms and providers assist organizations throughout the globe resolve pressing challenges — and remodel for the long run.
