Microsoft announced today that it is strengthening its partnership with NVIDIA by upgrading its GPUs from the previous A100 to H100, launching a new ND H100 v5 virtual machine developed specifically for artificial intelligence.
Microsoft said the virtual machine is mainly for enterprises that need to integrate or develop artificial intelligence, attached is the virtual machine configuration as follows.
- 8 NVIDIA H100 Tensor Core GPUs, interconnected via next-generation NVSwitch and NVLink 4.0
- NVIDIA Quantum-2 CX7 InfiniBand performance can reach 400 Gb/s per GPU and 3.2Tb/s per virtual machine on a non-blocking Fat-Tree architecture network. With NVSwitch and NVLink 4.0, the pairwise bandwidth split between the 8 local GPUs in each virtual machine is 3.6TB/s.
- With Intel 4th generation Xeon Scalable processors
- GPUs connected using PCIe Gen5 with 64GB/s per GPU bandwidth
- 16 channels of 4800MHz DDR5 DIMMs
Microsoft previously announced ChatGPT-related services in the Azure OpenAI service, and this virtual machine complements that service by allowing third-party access to chatbot technology via Azure.
