[Bigiron]

Nvidia expected to abide by US export curbs and Singapore laws, MTI says on queries about DeepSeek's chips

Singapore has come under the spotlight in a US investigation into whether DeepSeek, whose AI model’s performance has made headlines worldwide, had circumvented US restrictions on advanced Nvidia chips by buying them from third parties in other countries, including Singapore.

READ: https://bit.ly/3WLRdoz

Follow @Asiaonecom for all the latest updates

[goodboy]

"Senator, I am Singaporean."
🤣🤣🤣

[moonrab]

So it is Nvidia selling from Singapore 🤣

[Tangsen]

Typical of a loser mentality … lose … point finger at everyone, blame everyone instead of looking at the mirror

[goodboy]

So it is Nvidia selling from Singapore 🤣

You can Google for it. Nvidia Asia Hq in Suntec city

[WhatDoYouThink!]

For small gujin guraj companies it's easy to handle. But nvidia woh, how ah?

[Omnipresent88]

Hahaha amdk arse lickers stooge’s you don't know meh your daddies chut patern more than badminton LMAO 
https://media.tenor.com/Vt4Au-tHQoQAAAAM/lee-hsien.gif https://media.tenor.com/TIohhhv3TDoAAAAM...r-bean.gif

[Niubee]

The one used is A100 GPU. Very old ones. Even if u use the newest one and u won't get the maximum benefits because all usage of Nvidia GPU goes thru the CUDA application layer.

DeepSeek bypass the CUDA layer and wrote their own algorithm. That one one way to 弯道超车

[WhatDoYouThink!]

Bypassing cuda and rewrite the code in ptx (or mix with c/c++) can give much better math calculations and hardware specific SIMD operations. And to do that within such a short time frame is simply amazing

[watchfirst9]

Bypassing cuda and rewrite the code in ptx (or mix with c/c++) can give much better math calculations and hardware specific SIMD operations. And to do that within such a short time frame is simply amazing

tomshardware.com › tech-industry › artificial-intelligence › deepseeks-ai-breakthrough-bypasses-industry-standard-cuda-uses-assembly-like-ptx-programming-instead

DeepSeek's AI breakthrough bypasses industry-standard CUDA for some ...
5 days ago PTX sits between higher-level GPU programming languages (like CUDA C/C++ or other language frontends) and the low-level machine code (streaming assembly, or SASS).

[WhatDoYouThink!]

I think they wrote the whole program in cuda, and only need to rewrite the compute-intensive, iterative training loops in ptx which is only a small part.