Mass production is just starting in the second quarter
Nvidia is experiencing difficulties in supplying AI accelerators to China. And not only because of sanctions. It is reported that the H20 model, created specifically for the Middle Kingdom, will only go into mass production in the second quarter of this year.
Initially, back in November, Nvidia introduced the HGX H20, L20 PCIe and L2 PCIe accelerators. H20, as the most productive, was planned to begin shipping to China on November 16, that is, a week after the announcement. Then the deadlines were moved to the first quarter of 2024, and now, as we see, production is just starting in the second. The reasons were not given, but it was initially said that the delay was due to problems faced by server manufacturers in integrating the chip.
HGX H20 is based on the same GPU as the top-end H100, but has very different parameters. In particular, the performance in INT8 calculations of the H20 is lower than even that of the RTX 4090, and compared to the H100 the difference is almost sevenfold.