Zhaoxin introduces two CPU series The KH-4000 series will offer up to 32 cores. Today the Chinese company announces KH-40000 server CPU series and KX-6000G series for embedded systems. The first series will be available in 12, 16, and 32 core configurations. Zhaoxin KH-40000 The KH-4000 series are based on Yongfeng architecture with support for […]
There are two parts to M1/M2 performance. One is the SoC architecture that allows sharing memory without needing a bus. The other is the fact that instructions can be easily processed in batches and then independent instructions in each batch can be computed independently in parallel. The second part is the benefit that all RISC chips share. Being able to scale horizontally by adding more cores is a really nice property.
With regards to tech sanctions and stuff, I kind of suspect that China just doesn’t care. If US is already set on banning tech exports to China, they have very little leverage to enforce any infringement claims at that point. My guess would be that there are existing applications that still require x86 architecture, and that’s the reason for building such chips domestically.
It’s worth noting that there are a lot of different companies developing chips in China right now, and it looks like RISC is getting a lot of love there as well. It’s going to be really interesting to see where this all goes in the next decade. We are likely going to start seeing some interesting domestic designs coming out of China in the near future.
I’m also pretty excited to see if the experiments with other substrates actually end up being put into production at some point. I saw that China is experimenting with carbon based chips and some other alternatives to silicon.
If one of these approaches can be put into production at scale that could be the equivalent of going from vacuum tubes to transistors. We’re at the point where we’re hitting the limits of what’s possible to do with silicon, but even a crude chip on a different substrate could blow silicon performance away with lots of room for future improvements.
There are two parts to M1/M2 performance. One is the SoC architecture that allows sharing memory without needing a bus. The other is the fact that instructions can be easily processed in batches and then independent instructions in each batch can be computed independently in parallel. The second part is the benefit that all RISC chips share. Being able to scale horizontally by adding more cores is a really nice property.
With regards to tech sanctions and stuff, I kind of suspect that China just doesn’t care. If US is already set on banning tech exports to China, they have very little leverage to enforce any infringement claims at that point. My guess would be that there are existing applications that still require x86 architecture, and that’s the reason for building such chips domestically.
It’s worth noting that there are a lot of different companies developing chips in China right now, and it looks like RISC is getting a lot of love there as well. It’s going to be really interesting to see where this all goes in the next decade. We are likely going to start seeing some interesting domestic designs coming out of China in the near future.
I’m also pretty excited to see if the experiments with other substrates actually end up being put into production at some point. I saw that China is experimenting with carbon based chips and some other alternatives to silicon.
If one of these approaches can be put into production at scale that could be the equivalent of going from vacuum tubes to transistors. We’re at the point where we’re hitting the limits of what’s possible to do with silicon, but even a crude chip on a different substrate could blow silicon performance away with lots of room for future improvements.