NVIDIA officially released the specification information of the Hopper H100 GPU, and the performance is more powerful than expected. Earlier this year, NVIDIA announced the Hopper H100 GPU for AI data centers, up to 60 TFLOPs and 30 TFLOPs FP64.
However, as the release date approaches, NVIDIA has updated its graphics card specifications again, indicating that it has improved performance again. One reason for the improvement in performance metric numbers is that when the chip is put into production, GPU manufacturers can finalize the numbers based on the actual clock speed. NVIDIA is likely to use the conservative clock frequency to provide preliminary performance data, and as production is fully unfolding, the company found that the chip could provide better clock frequency.
Last month on GTC, NVIDIA confirmed that their Hopper H100 GPU is in full production and that partners will launch their first wave of products in October this year. It is also confirmed that Hopper's global promotion will include three phases, the first phase will be pre-orders for the NVIDIA DGX H100 system, and free labs to customers directly from NVIDIA, including systems such as Dell 's Power Edge server, now available on the NVIDIA launchpad.
NVIDIA Hopper GH100 GPU consists of a huge 144 SM (streaming multiprocessor) chip layout with a total of 8 GPCs. These GPCs have a total of 9 TPCs, each TPC further consists of 2 SM units. This gives us 18 SMs per GPC, and we have 144 out of the full 8 GPC configurations. Each SM consists of up to 128 FP32 units, which should give us a total of 18,432 CUDA cores.
