Introducing the Hopper architecture for high-performance computing systems at the GPU Technology Conference (GTC) held in March, Nvidia attracted great attention with the Hopper H100 with 80 GB VRAM. The company, which has recently confirmed the availability date of the new graphics card, has updated the performance characteristics of the graphics card.
Processing power goes up
Looking at the shared details, NVIDIA now performs more accurate single precision and double precision performance for its Hopper H100 GPU equipped with 16896 CUDA cores. The FP64 performance of the card in the previous details increased from 30 TFLOPS to 34, while the FP32 was updated from 60 to 67 TFLOPs. Thus, we can say that the clock speeds of the new graphics card have increased from the previously reported 1775 MHz to at least 1982 MHz.
Tested on Intel XeSS, Nvidia and AMD graphics cards
In addition, the NT8, FP8, FP16, BFLOT16 accelerated computing performances of the H100 have also been updated, with slightly lower estimates. While the new graphics card is expected to have a total TDP of 700 W, it will also serve over PCI Express with a TDP of half. Nvidia reports that the Hopper H100 will begin to appear in workstation systems in the first quarter of 2023.