deepseek-v4.1IMAGE
763B DeepSeek V4.1 Flash runs at 8 tok/s on an RTX 3090 with 64GB DDR4
A user runs the 763B DeepSeek V4.1 Flash 1017 on a single RTX 3090 24GB plus 64GB DDR4, getting 8 tok/s decode and 6 tok/s prefill — poor performance, but remarkable that a 3090 can run it at all.
米柒说
@meseven_journeyArchived 10 days agoOriginally posted 2026.10.22
米柒说 @meseven_journeyOriginally posted 2026.10.22
@TheAhmadOsman Yep, it's crazy! DeepSeek v4.1 Flash 1017 - a 763B model on one RTX 3090 24g + 64g DDR4 the performance sucks: 8 tok/s decode / 6 tok/s prefill but it's still crazy how capable an RTX 3090 is
x.com/meseven_journey/status/2113075217316937951Archived 2026.10.22 11:59