Chinese semiconductor thread II
- Thread starter vincent
- Start date
Baidu's Kunlunxin won the vast majority of China Mobile's AI inference chip bid for CUDA like ecosystem chip options. Again, Nvidia H20 is just not competitive right now
It is a mistake to aim for CUDA compatibility. Chinese AI vendors should just create their own standard. Otherwise they will perpetually be behind.
The Soviets stepped on that rake when they cloned the IBM S/360 and DEC PDP-11 to use their existing software.
The Soviets stepped on that rake when they cloned the IBM S/360 and DEC PDP-11 to use their existing software.
China is a saturated type of research and development, with teams following up on all technological paths.
Also, let's popularize the science.
1. In the field of high-performance computing, CUDA was initially assisted by China in implementing applications for Nvidia, and the first large-scale heterogeneous computing cluster built using CUDA was developed by China. In fact, using the Windows NT architecture to help Microsoft implement a supercomputer cluster is mainly done by Chinese people, but once it's done, it's cut into Linux versions.
2. Since the 1990s, China has made a large number of compatible chips for the most critical information systems in the United States, at least hundreds of alternative chips. After stepping through the pit, it is far beyond your imagination, and the compatibility level has been roast for a long time, and there are a lot of solutions. After 2015, China's military chip solutions gradually shifted from imitation substitution to comprehensive self-developed.
So now a core problem with NASA's SLS rocket (according to the GAO report) is that many specialized chips (such as the matching of RS25/SSME engines) are no longer produced in the United States. Currently, only East Asia has matching chips in the world. Guess who it is? Do you think Japan, South Korea, or Taiwan have special technology chips for the aerospace industry?