On RTX 5090, Qwen 27B hits 200 TPS but Flash next only 50: what model sits between for coding?
MasterNomie · reddit · 2026-09-28
A local-LLM user on RTX 5090 + 96GB DDR5 reports Qwen 3.8 27B decoding at 200+ TPS versus 50 TPS for Flash next, and asks for a coding model in between that sustains 75-100 TPS. Failing that, they propose splitting the workflow: Flash next for planning, 27B for implementation, with a RAM upgrade to 128GB planned.
More from Infra
- Sarvam AI Launches Token Factory: India-Hosted Inference for Open Models — itsOmSarraf_ · 2026-11-16
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Jailbroken PS4 runs a 9B open-source model at up to 43 tok/s by talking directly to its 2013-era GPU — AIFlow_ML · 2026-11-16
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16
- Stratechery's Fictional Obituary: Imagining Jensen Huang's 2026 Exit From Nvidia — Gregory_C_Allen · 2026-11-16
- vLLM serves DeepSeek V4.1-Flash at 954K tokens/s per GPU, $38.96/GPU-hour on GB300 NVL72 — bookwormengr · 2026-11-16