NVIDIA Hopper H100 GPU-Powered PCIe Graphics Card With 120 GB HBM2e Memory Capacity Spotted
NVIDIA has so far officially announced two versions of the Hopper H100 GPU, an SXM5 board and a PCIe variant. Both feature differently configured Hopper H100 GPUs and while their VRAM capacity is the same at 80 GB, the former utilizes the brand new HBM3 standard while the latter utilizes the HBM2e standard. Now based on information by s-ss.cc (via MEGAsizeGPU), NVIDIA might be working on a brand new PCIe version of the Hopper H100 GPU. The new graphics card won’t feature 80 GB HBM2e but will go all out with 120 GB of HBM2e memory. As per the information available, the Hopper H100 PCIe graphics card not only comes with all six HBM2e stacks enabled for 120 GB memory across a 6144-bit bus interface, but it also comes with the same GH100 GPU configuration as the SXM5 variant. This is a total of 16,896 CUDA cores and memory bandwidth that exceeds 3 TB/s. The single-precision compute performance has been rated at 30 TFLOPs which is the same as the SXM5 variant. So coming to the specifications, the NVIDIA Hopper GH100 GPU is composed of a massive 144 SM (Streaming Multiprocessor) chip layout which is featured in a total of 8 GPCs. These GPCs rock total of 9 TPCs which are further composed of 2 SM units each. This gives us 18 SMs per GPC and 144 on the complete 8 GPC configuration. Each SM is composed of up to 128 FP32 units which should give us a total of 18,432 CUDA cores. Following are some of the configurations you can expect from the H100 chip: The full implementation of the GH100 GPU includes the following units:
8 GPCs, 72 TPCs (9 TPCs/GPC), 2 SMs/TPC, 144 SMs per full GPU 128 FP32 CUDA Cores per SM, 18432 FP32 CUDA Cores per full GPU 4 Fourth-Generation Tensor Cores per SM, 576 per full GPU 6 HBM3 or HBM2e stacks, 12 512-bit Memory Controllers 60 MB L2 Cache
The NVIDIA H100 GPU with SXM5 board form-factor includes the following units:
8 GPCs, 66 TPCs, 2 SMs/TPC, 132 SMs per GPU 128 FP32 CUDA Cores per SM, 16896 FP32 CUDA Cores per GPU 4 Fourth-generation Tensor Cores per SM, 528 per GPU 80 GB HBM3, 5 HBM3 stacks, 10 512-bit Memory Controllers 50 MB L2 Cache Fourth-Generation NVLink and PCIe Gen 5
Now it is unknown if this is a test board or a future iteration of the Hopper H100 GPU that is being tested out. NVIDIA recently stated at GTC 22 that their Hopper GPU was now in full production and the first wave of products are rolling out next month. As yields get better, we may definitely see the 120 GB Hopper H100 PCIe graphics card and SXM5 variants in the market but for now, the 80 GB is what most customers are going to get.
NVIDIA HPC / AI GPUs
