The stock market has been rising recently, particularly semiconductor and technology companies, benefiting from news of cooling inflation and early interest rate cuts. In addition to factors in the macro environment, the main logic of the rise in semiconductors is that the cycle is gradually coming out of the bottom. The previous article also analyzed Intel's TSMC earnings report to the recovery of the HPC market. This week, Nvidia also announced an explosion of news, announcing the launch of the H200 data center GPU. So what is the performance of this H200 GPU, what performance improvements have been made compared to the H100, and does this mean that AMD's MI300 is not working?
H200:
Compared to the H100, the main difference of the H200 is that it is equipped with a 141 GB HBM3e, while the H100 is equipped with an 80 GB HBM. Such an upgrade provides higher memory frequency and bandwidth. In some environments, NVDA's performance is clearly limited by memory. The H200 helps solve this problem (at least until Nvidia's B100, which is expected to launch by the end of 2024). Although the H200's 141GB is higher than the AMD MI300's 128GB, it's still less than the 192 GB of AMD's upcoming Mi300x. Overall, however, Nvidia still has quite a few advantages in the software ecosystem, so this announcement will help Nvidia consolidate its leading position in the AI accelerator field.
So does the H200 have a big impact on the AMD MI300...