Enjoy 60 will most likely be an entry-level phone, so a suitable chip for it.
Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
Keeping costs down and profits up without affordable access to any foreign technology. That's the likely explanation. And that should give you a better understanding of just how tough and serious this tech war and trade war is.
it is suitable for domestic 14-10 nm process this year and early next year
Their 12nm process is using plenty of foreign tech. They really don't have a choice here for the next 2 years. Their 12nm process should be a lot better than what they had when they produced the original Kirin 710A. That's when I find these social media claims that they are bringing back Kirin 710A to be nonsensical.
No, I don't really accept that with stacked chiplets, you cannot get performance to at least Kirin 820/830. That's nonsense. You are likely to have twice the die surface area.
Guaranteed they're getting lower production costs and higher profits this way.
no, I mean 10nm finfet technology, not stacked
Has anyone heard of these guys? Shenzhen kaifa. Looks like they are also producing DRAMs. And they apparently will scale up to 120k wpm in a year and started production over a year ago. And they do their own production, testing & packaging according to this. Looks like they are using 8-layer stacking technology.
Is stacking common in DRAM? I don't recall reading about CXMT doing 8 layer stacking.
This is a strange news. Says here that Baidu will launch 文心一言 tomorrow as the Chinese version of ChatGPT I guess and that it will be use MLU590 (思元590) AI GPU by Cambricon (寒武纪).
Actually, I think this one is better. A more thorough document on what GPUs they will use. Apparently, they want to get to over 50% domestic chips in their data center over the next 3 years. They expect 10 to 20% of their GPUs to by MLU590. They consider it the top performing domestic GPU at the moment (better than Huawei's Ascend 910). A little strange here since their have Kunlun-2, but I guess they consider Cambricon product to be better. They also talked about other GPUs from Vestai and Tianshu, but those are not really in mass production. Didn't see them mention Biren & Moore Thread here. Not sure why. Maybe they are still validating them?
Looks like they are still buying A100 & A800. They specifically stated that they cannot be stuck with 1 supplier of GPUs.
Actually, I think this one is better. A more thorough document on what GPUs they will use. Apparently, they want to get to over 50% domestic chips in their data center over the next 3 years. They expect 10 to 20% of their GPUs to by MLU590. They consider it the top performing domestic GPU at the moment (better than Huawei's Ascend 910). A little strange here since their have Kunlun-2, but I guess they consider Cambricon product to be better. They also talked about other GPUs from Vestai and Tianshu, but those are not really in mass production. Didn't see them mention Biren & Moore Thread here. Not sure why. Maybe they are still validating them?
Looks like they are still buying A100 & A800. They specifically stated that they cannot be stuck with 1 supplier of GPUs.
Last edited:
well, looks like this ErnieBot will be unveiled at 2AM EST (so in 4 hours)
Looks like Baidu upgraded its data center in Yangquan, Shanxi recently finished upgrade here to support ErnieBot. And that future data centers will be built to support its large scale implementation.
This datacenter was built with 4 EFLOPS of calculation, the largest in Asia. Looks like Kunlun chips is widely used here.
- Status