factual

DeepSeek scheduled GPU communications below standard libraries

Because the H800 GPUs shipped to China legally had cut interconnect bandwidth, DeepSeek bypassed NVIDIA's standard NCCL communications library and manually scheduled communications by assigning specific SMs (GPU cores) to tasks using PTX, an assembly-level approach most companies avoid, in order to fully utilize their constrained hardware.

factualpending

Speaker

Dylan Patel

Evidence Quote

“they had to figure out how to get efficiencies. And one of those things is that instead of just calling the NVIDIA library NCCL, they scheduled their own communications”

Source

Deepseek Dylan Patel Nathan Lambert— Lex Fridman Podcast
Created: 6/13/2026, 3:31:25 AM

Your notes

Loading notes...