You can buy NVIDIA's next-gen Hopper GPU + 80GB HBM2e for $36,550

NVIDIA first announced its next-gen H100 Hopper GPU at GTC 2022 (GPU Technology Conference) earlier this year, its next-gen GPU is built on the new 4nm process node at TSMC and rocking a huge 80 billion transistors. Well, that next-gen GPU is now for sale... and it'll cost ya.

2

Japanese HPC retailer GDEP Advance has the NVIDIA H100 GPU listed for sale, which costs a whopping 4,745,950 yen (or around $36,550) which includes taxes and shipping: the card on its own costs 4,313,000 yen (around $33,200). This is also for the PCIe-based model with passive cooling for server enclosures, so the price of other SKUs is unknown at this point.

As for the NVIDIA H100 GPU, it's based on the new Hopper GPU architecture with NVIDIA using TSMC's 4nm process technology and packing 80 billion transistor into 814mm2 of silicon. The PCIe-based version of the NVIDIA H100 rocks 80GB of HBM2e memory with 2TB/sec of memory bandwidth, rocking on the PCIe 5.0 x16 connector.

• Read more: NVIDIA H100 GPU full details: TSMC N4, HBM3, PCIe 5.0, 700W TDP, more
• Read more: NVIDIA reveals next-gen Hopper GPU architecture, H100 GPU announced
• Read more: NVIDIA can sustain the world's internet traffic with 20 x H100 GPUs

2

NVIDIA has more CUDA cores inside of the SXM variant of its new Hopper-based H100 GPU, with 16896 CUDA cores on the SXM-based H100, while the PCIe-based H100 has 14592 CUDA cores.

• Read more: NVIDIA Hopper GPU is up to 40x faster with new DPX instructions
• Read more: NVIDIA is turning data centers into 'AI factories' with Hopper GPU
• Read more: NVIDIA Eos: the world's fastest AI supercomputer, 4608 x DGX H100 GPUs

Another big difference is that the SXM-based H100 uses HBM3 memory, while the PCIe-based H100 uses HBM2e memory... meanwhile, NVIDIA is keeping the TDP in check on the passively-cooled PCIe-based H100 GPU: at 350W, while the SXM-based H100 scales considerably further to 700W TDP.