NVIDIA has now provided some figures comparing its existing Ampere A100 accelerator against AMD Instinct MI250 GPUs in a new technical blog. NVIDIA has previously announced the release of its next-generation H100 GPU, which is based on the Hopper graphics (GPU) architecture and will be available to consumers later this year.
The Hopper GPU will bring an estimated 26x boost in performance over the Pascal P100, which was released six years ago, and will be 3x quicker than the Moore’s Law trajectory.
In terms of performance, NVIDIA tested the Ampere A100 GPU in single and multi-GPU systems. The same configurations were used for AMD’s Instinct MI250. For performance testing, some of the most prominent Data Center workloads were tested, including LAMMPS, NAMD, openMM, GROMACS, and AMBER.
NVIDIA’s single Ampere A100 GPU outperformed AMD’s Instinct MI250 GPU accelerator by up to 1.9x, while the quad-GPU solution outperformed the Ampere system by up to 2.1x. The quad-GPU approach gave 2.8x higher perf/watt in terms of energy efficiency.
The NVIDIA A100 GPU’s high performance and power efficiency are the product of many years of constant software-hardware co-optimization to improve application performance and efficiency
Following are the notes from the testing:
Efficiency ratio of A100 to MI250 shown – higher is better for NVIDIA. Geomean over multiple datasets (varies) per application. Efficiency is Performance / Power consumption (Watts) as measured for the GPUs using measured using NVIDIA SMI and equivalent functionality in ROCml
AMD MI250 measured on a GIGABYTE M262-HD5-00 with (2) AMD EPYC 7763 with 4x AMD Instinct™ MI250 OAM (128 GB HBM2e) 500W GPUs with AMD Infinity Fabric™ technology. NVIDIA runs on ProLiant XL645d Gen10 Plus using dual EPYC 7713 CPUs and 4x A100 (80 GB) SXM4
LAMMPS develop_db00b49(AMD) develop_2a35ec2(NVIDIA) datasets ReaxFF/c, Tersoff, Leonard-Jones, SNAP | NAMD 3.0alpha9 dataset STMV_NVE | OpenMM 7.7.0 Ensemble runs for datasets: amber20-stmv, amber20-cellulose, apoa1pme, pme|
GROMACS 2021.1(AMD) 2022(NVIDIA) datasets ADH-Dodec (h-bond), STMV (h-bond) | AMBER 20.xx_rocm_mr_202108(AMD) and 20.12-AT_21.12 (NVIDIA) datasets Cellulose_NVE, STMV_NVE | 1x MI250 has 2x GCD
It should be noted that the AMD Instinct MI250 used here isn’t the full configuration because it’s built on the MI250X, but based on these findings, the A100 should be quite competitive versus AMD’s cDNA 2 offerings. With Hopper on the way, NVIDIA will push these figures even higher, which is where AMD’s Instinct MI300 with its brand new APU-like design comes in.