Artificial Intelligence thread

meedicx

Junior Member
Registered Member
If deepseek is only relying on huawei now for its model inference and it took so long because of adapting to huawei is so hard, then why did chinese companies buy huawei cards in 100k+ to million in 2025 and 2024? Did they buy it just for decoration? What about cambricorn, Biren, moore threads and all these other cards?

The whole point of all these cards is that they can do deep learning. If training is so hard, then companies are buying them for atleast to do inference.

So your whole argument falls apart.

Again, the only logical reason for taking this long for deepseek is to be able to do training on these cards. Because inference has already been done since 2 years ago.

DeepSeek V3 series last year have always supported Ascend for inference. Huawei's CloudMatrix 384 announcement last year heavily marketed how many TPS it could run R1. If you look at the github of major Chinese LLMs like qwen or GLM, they support multiple Chinese ai cards including Ascend.

DeepSeek V4 used TileLang for all its kernels so in theory can support any chinese AI chip with TileLang support. The Ascend CANN team worked closely with DeepSeek so their hardware runs V4 faster and more efficiently for both training and inference.

From their public announcement, Atlas A3 (aka. CloudMatrix, uses last gen Ascend chips) supports DeepSeek V4-flash continuous pretraining (CPT) with high throughput, so in theory it could also be used for the full pretraining since the code is the same. Some rumors say flash was trained using Ascend. No info on pro, but will probably need the next generation 950DT supernode to support training at that size.
 
Last edited:

Michael90

Senior Member
Registered Member
Please, Log in or Register to view URLs content!



I'm not sure in what ways China could force a separation of a company that has almost completed its merger.
I think they know they can’t do anything anymore . It’s too little too late. I think it’s more of a symbolic gesture to dissuade other Chinese AI startups/companies from doing the same .
im actually surprised China had no laws in place to review any takeover of her high tech companies in critical emerging fields like AI . It’s actually shocking that their regulators had been sleeping on the wheel. This will be unthinkable in the US or west today. They have long set laws to review any takeover bid for their tech companies since 2017s upwards to discourage Chinese takeovers of their tech companies actually.
I don’t understand why the CCP is now blocking the founders from travelling . They did nothing wrong. There was no law in China that said they couldn’t do that. You can’t just wake up and retroactively apply rules to others (like India infamous tax rules . lol ).
 

SanWenYu

Major
Registered Member
Please, Log in or Register to view URLs content!



I'm not sure in what ways China could force a separation of a company that has almost completed its merger.
The transaction has become illegal. Specifically, Meta is not allowed to use the assets, IP included, obtained from the failed merger. If Meta and Manus want to ignore the ban and carry on still, there will be lawsuits waiting for the companies and their C-suites. It's now up to them. They can surely try to dodge the long arm of Chinese laws.
 
Last edited:

sunnymaxi

Colonel
Registered Member
I think they know they can’t do anything anymore . It’s too little too late. I think it’s more of a symbolic gesture to dissuade other Chinese AI startups/companies from doing the same .
im actually surprised China had no laws in place to review any takeover of her high tech companies in critical emerging fields like AI . It’s actually shocking that their regulators had been sleeping on the wheel. This will be unthinkable in the US or west today. They have long set laws to review any takeover bid for their tech companies since 2017s upwards to discourage Chinese takeovers of their tech companies actually.
I don’t understand why the CCP is now blocking the founders from travelling . They did nothing wrong. There was no law in China that said they couldn’t do that. You can’t just wake up and retroactively apply rules to others (like India infamous tax rules . lol ).
nope. China have many cards to play.

Meta earned a lot from mainland China despite zero presence in the country.

 

siegecrossbow

Field Marshall
Staff member
Super Moderator

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
my piece on the above chart. Key is to not just look at compute. for the longest time, Nvidia at same TFLOPS actually gave greater throughput than Chinese competitors. Now, with this hybrid MXFP8/MXFP4 + deep integration, Ascend may have finally caught up or surpassed Nvidia in this throughput/TFLOPS metric

 
Top