Running DeepSeek Flash V4.1 IQ2_XS on Four R9700 GPUs: Real Numbers
poofph · reddit · 2026-10-26
A redditor benchmarked DeepSeek Flash-V4.1-1017 at IQ2XS quantization on a 4x R9700 server (Epyc 7402P, Supermicro H12SSL-NT, 256GB 8-channel DDR4 ECC 3200) using the Strata engine, sharing performance screenshots. Prefill is the main pain point due to the R9700's modest memory bandwidth; system RAM usage stayed surprisingly low. Notably, the Radiance engine's optimization that delivers 3-4x prefill and token speed for Flash Next on R9700s is reportedly coming to Strata soon, though coverage of DeepSeek's engine is uncertain.
More from Infra
- GPU rental prices up ~80% this year as token demand climbs, argues analyst — JOBhakdi · 2026-10-26
- Musk warns of stranded compute; trader counters inference demand is near-infinite — markjeffrey · 2026-10-26
- Open-source Kinoloom turns an 8 GB GPU into a local AI film studio with seamless latent-space long takes — Dry-Acanthisitta-652 · 2026-10-26
- $550 Dell R820 with 768 GB RAM: is an old monolithic server viable for LLM inference? — BobtheGodGamer · 2026-10-26
- Kubernetes works for vLLM inference, but agent sandboxes needed Modal's custom stack at 1M+ concurrency — NielsRogge · 2026-10-26
- Anthropic reportedly offers Blackstone 10%+ yield on $9B junior debt for 1.5GW TPU deal — Beth_Kindig · 2026-10-26