and basically, what the #DeepSeek team did was, since the H800 chips didn’t have the power that the H100s, they had to optimize everything down to the assembled components of the chip.
back in the 1990s that’s how folks first using languages like Java had to code things ―the software language was far more powerful than the retail hardware. so coders like my ex- (who used Java to create abstract expressionist painting on the screen) was doing assembly level coding to achieve that in 1998.
🧵
=> More informations about this toot | View the thread | More toots from blogdiva@mastodon.social
=> View deepseek tag This content has been proxied by September (3851b).Proxy Information
text/gemini