随手摸下B200 (148SM)。Vector性能相比H100的提升比较常规(+25%)。全局原子CAS延迟是H100的2x，达到200ns，这大概就是MCM的代价。有意思的点在于wmma发生明显倒退，每SM性能回到了图形卡水平。本代大概只有tcgen05.mma可以跑出纸面tensor吞吐，这标志着计算卡重心逐渐转向类似NPU的TMEM？

David's random thoughts

随手摸下B200 (148SM)。Vector性能相比H100的提升比较常规(+25%)。全局原子CAS延迟是H100的2x，达到200ns，这大概就是MCM的代价。

有意思的点在于wmma发生明显倒退，每SM性能回到了图形卡水平。本代大概只有tcgen05.mma可以跑出纸面tensor吞吐，这标志着计算卡重心逐渐转向类似NPU的TMEM？

www.tg-me.com/cn/David& 39;s random thoughts/com.david_random/567

2.2K viewsApr 24 at 10:42

tg-me.com/david_random/567

Create: 2025-04-24
Last Update: 2025-06-01 20:08:46

BY David's random thoughts

Share with your friend now:
tg-me.com/david_random/567

David& 39;s random thoughts Telegram | DID YOU KNOW?

Find Channels On Telegram?

Telegram is an aspiring new messaging app that’s taking the world by storm. The app is free, fast, and claims to be one of the safest messengers around. It allows people to connect easily, without any boundaries.You can use channels on Telegram, which are similar to Facebook pages. If you’re wondering how to find channels on Telegram, you’re in the right place. Keep reading and you’ll find out how. Also, you’ll learn more about channels, creating channels yourself, and the difference between private and public Telegram channels.