NVIDIA H20 AI GPUs
The nation of China Plans to Begin Mass Producing GPU manufacturer NVIDIA H20 AI GPUs in Q2 2024, Absolutely In accordance With US Laws
According to reports, NVIDIA intends to start mass producing its H20 AI GPU in the second quarter of 2024, specifically for the Chinese market.
NVIDIA Will Start Shipping AI GPUs to China in Q2 2024, With NVIDIA H20 AI GPUs Mass Production To Begin.
November of last year saw reports that NVIDIA was getting ready to release a new line of NVIDIA H20 AI GPUs for the Chinese market in place of the H100, L40, and L4 chips, which were prohibited by US policy. The most recent chips were now to be produced in complete compliance with US export and trade regulations.
Technically speaking, the NVIDIA H20 AI GPUs is a simplified version of the H100 GPU, with 96 GB of RAM that can operate at up to 4.0 Tb/s, 296 TFLOPs of compute capability, and the GH100 die, which has a performance density of 2.9 TFLOPs/die as opposed to 19.4 for the H100.
According to the chart, the H100 SXM is 6.68 times quicker than the H20 SXM; however, those FLOPs are FP16 Tensor Core FLOPs (with Sparsity), not INT8 or FP8 FLOPs. The GPU is an HGX solution with 8-way configurations and a 400W TDP. It still has a 900 GB/s NVLINK connection and supports 7-Way Multi-Instance GPU technology.
The US-China chip rivalry intensified, delaying the debut of these chips from late 2023 to early 2024. The US government has now softened its stance on commercial sales inside China, and NVIDIA has promised that its chips would fully comply with the rules and regulations put out by US trade and commerce authorities. This has already resulted in the GeForce RTX 4090D being released in the area.
The business is currently anticipating starting mass production of its H20 and the other AI GPUs in Q2 2024. Wistron, a significant substrate supplier for the green team, will manage this process. As some Chinese AI businesses look exploring alternatives like Huawei, who provide highly appealing price/performance value in the AI area, others have started repurposing the final available supply of GeForce RTX 4090 GPUs for their own purposes.
Due to constraints on US-based tech businesses, Huawei is likely to earn a significant market share in the nation. It seems that the first chips will be sent to Chinese AI clients around the middle or end of Q2 2024.
[…] NVIDIA AI inference platform, according to Park, a competitive cyclist and computer engineer based in Zurich […]