AMD and Intel will have to come up with something better than Sapphire Rapids and Genoa.
Nvidia has unveiled its new DGX GH200 supercomputer platform, equipped with 256 Grace Hopper GH200 superchips . Each chip combines a 72-core ARM Grace processor and an H100 Tensor Core GPU. The chip also supports up to 480GB of LPDDR5 RAM and 96GB of HBM3 or 144GB of HBM3e memory.
These are really impressive technical specifications. The Phoronix edition received for testing a system based on the GH200 processor worth more than 40 thousand dollars. The purpose of testing is to verify Nvidia's claims about the high performance of the platform.
Currently, the Grace Hopper GH200 is the most powerful AI chip from Nvidia, designed for large-scale AI and supercomputing tasks. The company says it delivers 10x faster performance for applications running on terabytes of data, enabling scientists and researchers to create unprecedented solutions for complex tasks.
The development opens the way for Nvidia to enter the server processor market, which is now dominated by solutions based on the x86 architecture from AMD and Intel. The company claims that the Grace processor delivers twice as much efficiency as Intel's Sapphire Rapids and AMD's Genoa at the same power consumption, and is also 3.5 times faster than AMD's previous-generation Milan processors.
Phoronix was tested on the Linux operating system, and the superchip showed promising results. The GH200's performance was slightly higher than that of the Xeon Platinum 8380 2P, and slightly lower than that of the Epyc 9654 Genoa 2P, which Phoronix said was "not bad for the first show." The GH200 also proved to be the fastest compared to Genoa (X) and Emerald Rapids in single-processor configurations. Also, the GH200 is almost twice as efficient as the 128-core Arm Ampere Altra Max.
In testing Rodinia HPC, where the LavaMD load was used, the GH200 chip showed the same speed as the system with a single AMD Epyc 9684X Genoa-X processor.
And in another test-AMG, a single GH200 chip demonstrated performance close to the dual-processor configuration based on the Xeon Platinum 8380.
Phoronix experts called the chip's power "extremely impressive", especially when working with the NWChem computational chemistry software. The single GH200 is almost on par with AMD's EPYC Genoa dual-processor configuration.
Nvidia has unveiled its new DGX GH200 supercomputer platform, equipped with 256 Grace Hopper GH200 superchips . Each chip combines a 72-core ARM Grace processor and an H100 Tensor Core GPU. The chip also supports up to 480GB of LPDDR5 RAM and 96GB of HBM3 or 144GB of HBM3e memory.
These are really impressive technical specifications. The Phoronix edition received for testing a system based on the GH200 processor worth more than 40 thousand dollars. The purpose of testing is to verify Nvidia's claims about the high performance of the platform.
Currently, the Grace Hopper GH200 is the most powerful AI chip from Nvidia, designed for large-scale AI and supercomputing tasks. The company says it delivers 10x faster performance for applications running on terabytes of data, enabling scientists and researchers to create unprecedented solutions for complex tasks.
The development opens the way for Nvidia to enter the server processor market, which is now dominated by solutions based on the x86 architecture from AMD and Intel. The company claims that the Grace processor delivers twice as much efficiency as Intel's Sapphire Rapids and AMD's Genoa at the same power consumption, and is also 3.5 times faster than AMD's previous-generation Milan processors.
Phoronix was tested on the Linux operating system, and the superchip showed promising results. The GH200's performance was slightly higher than that of the Xeon Platinum 8380 2P, and slightly lower than that of the Epyc 9654 Genoa 2P, which Phoronix said was "not bad for the first show." The GH200 also proved to be the fastest compared to Genoa (X) and Emerald Rapids in single-processor configurations. Also, the GH200 is almost twice as efficient as the 128-core Arm Ampere Altra Max.
In testing Rodinia HPC, where the LavaMD load was used, the GH200 chip showed the same speed as the system with a single AMD Epyc 9684X Genoa-X processor.
And in another test-AMG, a single GH200 chip demonstrated performance close to the dual-processor configuration based on the Xeon Platinum 8380.
Phoronix experts called the chip's power "extremely impressive", especially when working with the NWChem computational chemistry software. The single GH200 is almost on par with AMD's EPYC Genoa dual-processor configuration.
