DeepSeek V4.1 Flash hits 2,100+ tokens/sec per user on 8× AMD MI355X via megakernels
JiaZhihao · x · 2026-10-28
AMD announced a milestone with inference startup lithosai: 2,100+ tokens/sec per user on DeepSeek V4.1 Flash at full model quality on 8× AMD Instinct MI355X, enabled by persistent megakernels and instruction-level control on open hardware. AMD says this is just the beginning.
Related event: AMD and Lithos Hit 2,100+ Tokens/sec per User with DeepSeek on MI355X(3 posts)→
More from Infra
- Google TPUs coming to Oracle Cloud, first availability outside Google's platform — SumitGup · 2026-10-28
- Owning AI infrastructure lets Meta ship model variants faster, argues Ben Bajarin — BenBajarin · 2026-10-28
- Google's anti-scraping defense backfires: SEO tools now scrape nearly 3x more — gaganghotra_ · 2026-10-28
- Intel's Ryan Shrout polls AI agents on their CPUs: EPYC for Meta Muse and OpenAI dots, Xeon for Grok Bot and Claude Duet — ryanshrout · 2026-10-28
- CoreWeave builds first AI-purpose global backbone: 51 data centers, APAC/Middle East expansion — OnlineInference · 2026-10-28
- Swarms Cloud revamps Models page aggregating 2,000+ LLMs with pricing — KyeGomezB · 2026-10-28