A peek into Gigabyte's GPU Arm for AI, HPC shops
High-performance platform choices are going beyond the ubiquitous x86 standard
Arm-based servers continue to gain momentum with Gigabyte Technology introducing a system based on Ampere's Altra processors paired with Nvidia A100 GPUs, aimed at demanding workloads such as AI training and high-performance compute (HPC) applications.
It supports 16 DDR4 DIMM slots, which would be enough space for up to 4TB of memory if all slots were filled with 256GB memory modules. The chassis also has space for no fewer than eight Nvidia A100 GPUs, which would make for a costly but very powerful system for those workloads that benefit from GPU acceleration.
Gigabyte said it developed the G492-PD0 in response to demand for high-performance platform choices beyond the ubiquitous x86 server systems. It apparently joins the portfolio of Gigabyte G492 servers that support the Nvidia HGX A100 8-GPU "Delta" baseboard, which is how the platform can support up to eight GPUs, interconnected using Nvidia's NVSwitch chip.
In fact, the G492-PD0 chassis is a 4U rack-mount enclosure, but most of the space appears to be given over to housing the Nvidia baseboard and any GPUs that are installed, inside what Gigabyte calls a dedicated cooling chamber that ensures the highest airflow possible to cool high-performance components. The server mainboard is located in a 1U server module that sits on top of this cooling chamber and contains all the Ampere CPU, memory, storage, and front-facing expansion slots.
Provision for storage covers 6 x 2.5in bays for U.2 NVMe drives, plus 4 x M.2 slots internally for NVMe storage modules. For networking, there are two Gigabit Ethernet ports plus a dedicated port for management. Redundant power supplies are also standard.
- US fears China may have ten exascale systems by 2025
- Ready for testing: First-ever supercomputer powered by Intel's wildcard AI chips
- HPE building its 4th global 'supercomputer factory'
- AMD hasn't forgotten about that ambitious 2025 energy efficiency goal it promised
According to Gigabyte, the HGX baseboard supports Nvidia GPUDirect RDMA, which allows for direct data exchange between GPUs and third-party devices such as NICs or storage adapters. Also supported is GPUDirect Storage, which offloads the movement of data from storage to GPU memory from the CPU.
Gigabyte said it will continue to pursue a prolific line-up of servers in the market for all enterprise workloads, including Arm-based servers. Future platforms will be created based on market demand, hinting that this will include more liquid cooling solutions.
Gigabyte's move follows the recent announcement from Microsoft that its own Ampere Altra-based Azure servers are Arm SystemReady certified, which also applies to the Gigabyte G492-PD0.
More importantly, both Microsoft and Arm highlighted that such systems are ready for enterprise and datacenter deployment. ®