Back in 2019 when Azure launched the first GPU-partitioned (GPU-P) virtual machine (VM) offerings in the public cloud, our customers loved it and asked for a similar offering on NVIDIA GPUs. Our customers wanted the flexibility to choose the GPU that meets the workload requirements and get the benefits of GPU-P, which enables cost-effective configurations based on the requirements. While our existing NVsv3 VMs with NVIDIA M60 GPUs worked well to run graphics-heavy visualization workloads, our customers had few specific requirements to make the experience better.
- Flexible GPU sizes with partitioning on NVIDIA GPU.
- A high-frequency AMD CPU part to improve the performance of applications that are optimized for a single CPU thread.
- VMs with very high RAM to load large data sets for three-dimensional geological modeling applications like Schlumberger Petrel.
Announcing new NVads A10 v5 VM series based on AMD EPYCTM 74F3(V) processors and virtualized NVIDIA A10 Tensor Core GPU
Continuing with our promise to offer innovative solutions for our customers, we are very excited to announce that our latest NVads A10 v5 series is now available for preview. Azure was the first and the only public cloud provider to offer unprecedented GPU resourcing flexibility with GPU-partitioning and we are happy to now bring the same technology on NVIDIA A10 Tensor Core GPUs. Customers can select from VMs with one-sixth of an A10 GPU and scale all the way up to 2*A10 configuration. This offers cost-effective entry-level and low-intensity GPU workloads on NVIDIA GPUs, while still giving customers the option to scale up to powerful full-GPU and multi-GPU processing power.
|Size||vCPU||Memory (GiB)||GPU Memory (GiB)||Azure Network (GBps)|
With our hardware-based GPU virtualization solution built on top of NVIDIA virtual GPU, NVIDIA RTX Virtual Workstation, and industry-standard SR-IOV technology, customers can securely run workloads on virtual GPUs with dedicated GPU frame buffer. The third-generation AMD EPYC CPUs with a boost clock speed of 4 GHz and a base of 3.2 GHz can provide the power you need to run any application. While simultaneous multithreading (SMT) is enabled by default on NVads A10 v5 series, Azure provides the flexibility to turn SMT OFF for applications that cannot take advantage of multiple threads.
Customers can learn more about the NVadsA10 v5-series now and sign up for NVads A10 v5 access today. NVads A10 v5 VMs are initially available in the South Central US and West Europe Azure regions. NVads A10 v5 will be available in additional regions soon thereafter.