GIGABYTE Announces Support for New NVIDIA Technology
June 28, 2021 – GIGABYTE Technology, an industry leader in high-performance servers and workstations, today announced support for the newly unveiled NVIDIA A100 80GB PCIe, NVIDIA Quantum-2 NDR 400Gb/s InfiniBand switches, and NVIDIA Magnum IO GPUDirect Storage that is aimed at HPC and AI applications. Support for these products by GIGABYTE ensures compatibility for the HGX platform and high-performance computing clusters deployments.
Recognizing the demand for the NVIDIA HGX technology, GIGABYTE has prepared servers for both 8-GPU and 4-GPU HGX systems. The [https//www.gigabyte.com/Enterprise/GPU-Server/G492-ZD0-rev-A00]G492-ZD0 offers the ultimate in HPC performance in a 4U chassis with dual processors to supply the necessary lanes for 8 x A100 HGX. The G262-ZR0 is the 4-GPU HGX system that also uses dual processors. Both systems have expansion slots for networking and Gen 4 U.2 drive bays for the fastest throughput. These two systems have been submitted for certification by NVIDIA, to validate the best configuration for performance, security, and scalability.
The A100 40GB proved to be an excellent GPU for HPC and AI, and doubling the amount of GPU memory allows for 25 percent higher GPU memory bandwidth. The A100 40GB has a memory bandwidth of 1555 GB/s and the A100 80GB has 2039GB/s. This promotes faster results and a quicker return on investment as data and large neural networks can better communicate and do so more efficiently.
For those that need to share a single GPU with many users, the A100 80GB can use the NVIDIA Ampere architecture’s Multi-Instance GPU (MIG) technology and support up to seven instances. This allows for seven simultaneous users of GPU resources at a time, with up to 40GB that can be partitioned to a user.
The NVIDIA Quantum-2 modular switches provide scalable port configurations up to 2,048 ports of NDR 400Gb/s InfiniBand (or 4,096 ports of NDR200) with a total bi-directional throughput of 1.64 petabits per second - 5x over the previous-generation. Based on industry standards, the NVIDIA Quantum-2 switches, which are expected to sample by year end, are backward- and forward-compatible, enabling easy migration and expansion of existing systems and software. Being an NVIDIA Certified system partner, GIGABYTE will be happy to introduce new HGX NVIDIA 80GB PCIe GPUs with NVIDIA NDR 400G InfiniBand and NVIDIA Magnum IO system to worldwide industrial AI and HPC customers by end of this year.
More workloads want to avoid the CPU for faster communication and overall performance enhancements. GPUDirect Storage enables direct IO between GPU memory, network, and storage. Magnum IO GPUDirect Storage reduces the amount of CPU cores and CPU utilization. At the same time, it has very low IO latency.
With a deep portfolio of servers for high GPU compute and the integration of the latest technology by NVIDIA, GIGABYTE offers servers to excel in data analytics, AI training, and more. In the coming months, more servers will roll out including those with liquid cooling options for GPU centric and high-density servers.
“GIGABYTE servers are used in many top supercomputers in the world, helping researchers to discover and make great breakthroughs,” said Daniel Hou, VP of Network & Communication Business Unit at GIGABYTE. “As NVIDIA keeps pushing the limitations and finding novel solutions, we are always going to follow suit and offer our customers solutions that can withstand whatever our users throw at them.”
Source: GIGABYTE
