I don’t see how Huawei could have made that jump. There’s a lot of generational knowledge and software development behind Nvidia’s chips. No company can become a competitor first time.
Last I heard, they expect it to be about half as powerful as what Nvidia can make.
But even that is very impressive considering the production obstacles Huawei has to overcome.
Remember the H100 is 3 years old, so Huawei isn’t catching up quite yet.
I think with a lot of these chips geared at AI there’s so much variation in how performance is measured you can say things like it beats an H100 inferencing on a specific INT4 model and that might be true. However, it might only hold for that narrow case.
So press releases aren’t worth much to me. Show me benchmarks.
Absolutely, I am sure Huawei has amazing resources for development, and they can do a lot.
But I’m also pretty sure they can’t beat Nvidia, with a process only allowing half the transistors at lower clock.
In fact if they manage to match Nvidia even in 5 years time, when they may have better production, that too would be an astounding feat.
But there is no way they can surpass Nvidia now.
They have made staggering progress, especially after the chip ban.
They costantly baffle the world with achievents not deemed possible.
You could be proven wrong
Yes they have, but SMIC can still only make the equivalent of TSMC 7nm process. And SMIC is the leading chip manufacturer in China. They could of course have better design, but it needs to be more than twice as efficient to match Nvidia performance on the currently available processes.
Yea no. By the time China can duplicate what NVIDIA has, they won’t need that kind of power anymore since the models will be better and more efficient. China is acutely aware of this so they aren’t actually aiming to beat NVIDIA on raw performance. Just cost. And if they do that they will win.
NVIDIA doesn’t have a strategy for when this AI bubble bursts but is quite clear China does.
threaded - newest
I don’t see how Huawei could have made that jump. There’s a lot of generational knowledge and software development behind Nvidia’s chips. No company can become a competitor first time.
.
Last I heard, they expect it to be about half as powerful as what Nvidia can make.
But even that is very impressive considering the production obstacles Huawei has to overcome.
Remember the H100 is 3 years old, so Huawei isn’t catching up quite yet.
I think with a lot of these chips geared at AI there’s so much variation in how performance is measured you can say things like it beats an H100 inferencing on a specific INT4 model and that might be true. However, it might only hold for that narrow case.
So press releases aren’t worth much to me. Show me benchmarks.
Absolutely, I am sure Huawei has amazing resources for development, and they can do a lot.
But I’m also pretty sure they can’t beat Nvidia, with a process only allowing half the transistors at lower clock.
In fact if they manage to match Nvidia even in 5 years time, when they may have better production, that too would be an astounding feat.
But there is no way they can surpass Nvidia now.
They have made staggering progress, especially after the chip ban.
They costantly baffle the world with achievents not deemed possible.
You could be proven wrong
Yes they have, but SMIC can still only make the equivalent of TSMC 7nm process. And SMIC is the leading chip manufacturer in China. They could of course have better design, but it needs to be more than twice as efficient to match Nvidia performance on the currently available processes.
Yea no. By the time China can duplicate what NVIDIA has, they won’t need that kind of power anymore since the models will be better and more efficient. China is acutely aware of this so they aren’t actually aiming to beat NVIDIA on raw performance. Just cost. And if they do that they will win.
NVIDIA doesn’t have a strategy for when this AI bubble bursts but is quite clear China does.