Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Didn't China stop disclosing the performance data of its supercomputers?
How significant is it for Chinese supercomputers winning this green graph 500 race?
Since the US gov. put the Chinese supercomputer makers on its entity lists, it is said that China's new supercomputers including the exascale ones can only use chips made by SMIC at 14nm at the best. And these chips are said to be more power hungry and slower per core than the latest and greatest from nvidia and intel.
How significant is it for Chinese supercomputers winning this green graph 500 race?
Since the US gov. put the Chinese supercomputer makers on its entity lists, it is said that China's new supercomputers including the exascale ones can only use chips made by SMIC at 14nm at the best. And these chips are said to be more power hungry and slower per core than the latest and greatest from nvidia and intel.
Kling continue to lead in # of customers and videos generated.
Considering that hyper scaler compute infrastructure for AI training far surpasses supercomputer, it does not matter. AWS, Google and meta don't even bother with these benchmarks.
SMIC can manufacture chips at 7nm. And like previous Chinese supercomputers showed it is possible to have high peak performance with a dedicated design for supercomputers, instead of doing like the US which uses off the shelf parts to build their machines.
Aws, Azure, GCP, Meta, Alibaba, Deepseek have not benchmarked their clusters. Doesn't mean they are not the leaders in compute power.
Last edited:
I know SMIC can do 7nm. Now given this green graph 500 ranking, I am wondering whether this increases the likelihood that these "Tianhe exa-node prototypes" are at 7nm nodes.