10.6 C
Sunday, February 11, 2024

Nvidia’s Grace Hopper GH200 superchip benchmarked, holds up properly towards AMD and Intel competitors

Why it issues: Nvidia’s Grace Hopper GH200 is Nvidia’s response to the superior AI and HPC processing calls for within the enterprise and information facilities. When saying Grace Hopper final 12 months, Nvidia’s CEO Jensen Huang mentioned that the tech business was hitting a tough wall with conventional architectures, which is why it has been more and more turning to GPUs and accelerated computing to unravel advanced computing duties.

Nvidia’s DGX GH200 supercomputing platform packs 256 Grace Hopper GH200 superchips, the place every chip combines a 72-core Arm-based Grace CPU and an H100 Tensor Core GPU. The chip additionally helps as much as 480GB of LPDDR5 reminiscence and 96GB of HBM3 or 144GB of HBM3e reminiscence.

These are spectacular specs any method you slice them, and now Linux-focused publication Phoronix acquired their palms on a $40K+ system based mostly on the GH200 for some benchmarking to see whether or not the GH200 can stand as much as Nvidia’s claims.

That is Nvidia’s strongest AI chip up to now, designed for giant-scale AI and high-performance computing functions. The corporate claims it delivers as much as ten occasions greater efficiency for functions operating terabytes of information and that this enhance in energy allows scientists and researchers to create unprecedented options for advanced issues.

From a company perspective, it serves as a substitute for x86 CPUs within the server phase, permitting Nvidia to supply a aggressive product towards AMD and Intel’s choices. Nvidia has mentioned that Grace CPU presents as much as twice the efficiency of the Intel Sapphire Rapids and AMD Genoa CPUs on the identical energy and as much as 3.5 occasions the effectivity of AMD’s last-gen Epyc Milan CPUs.

Phoronix performed the benchmarks in Linux, and the superchip proved aggressive. For the HPCG benchmark, the GH200 efficiency got here out simply forward of the Xeon Platinum 8380 2P and simply shy of the Epyc 9654 Genoa 2P efficiency, which Phoronix declared was “not unhealthy in any respect for an preliminary displaying.” It additionally discovered that the GH200 was the quickest towards the 2P Genoa(X) and Intel Emerald Rapids single processor configurations. The GH200 had almost twice the efficiency of the Ampere Altra Max 128-core Arm processor.

The GH200 tied with the AMD Epyc 9684X Genoa-X 1P processor for the Rodinia HPC benchmark with the LavaMD take a look at case. For AMG, the superchip system almost matched the Xeon Platinum 8380 2P for CPU efficiency.

Phoronix known as the processor’s efficiency “very spectacular” with the NWChem computational chemistry software program. The one GH200 almost tied AMD EPYC Genoa within the main 2P configuration.

Latest news
Related news


Please enter your comment!
Please enter your name here