This post has been republished via RSS; it originally appeared at: Microsoft Tech Community - Latest Blogs - .
Last month we first announced ND MI300X v5 virtual machines (VMs) and talked about how they were the culmination of a long-term partnership between Microsoft and AMD. We also talked about how Microsoft is working with AMD to build powerful cloud systems for AI with open-source software at Ignite, our annual conference for developers and IT professionals. Today, we are excited to share the latest milestone in our journey: the Preview of ND MI300X v5 VMs. This announcement was made earlier today by Microsoft's CTO, Kevin Scott, and AMD's CEO, Lisa Su, at the AMD Data Center Technology Premiere event.
ND MI300X v5 VMs are designed to tackle the most challenging AI and HPC workloads. They are powered by the AMD Instinct MI300X accelerator, which provides each VM with a staggering 1.5 TB of high bandwidth memory (HBM) and 5.2 TB/s of memory bandwidth. Moreover, these VMs are connected by NVIDIA Quantum-2 CX7 InfiniBand, which delivers 3.2Tb/s of scale out bandwidth per VM. This enables you to scale up to thousands of VMs and tens of thousands of GPUs. With ND MI300X v5 VMs, you can train and run the latest large language models (LLMs) faster and with fewer GPUs, saving you power, cost, and time.
ND MI300X v5 VMs are now available for preview and will soon become a standard offering in the Azure portfolio. Availability will be limited in the beginning, but we will support additional customers as more capacity comes online. Don't miss this opportunity to try out these amazing VMs. To be added to the queue, please sign up today to request access.