llamaperf

Qwen3.8 125B (6B active) Flash-Next

on NVIDIA RTX 5090 · NInfer · 250,000 ctx

Tone: mixedNVIDIA hardware
Nov 12, 2026
Throughput
150-200 t/s gen
VRAM reported
32 GB

Use cases

agenticlong-context

Summary

User reports Qwen3.8 27B with MTP on an RTX 5090 at 150-200 t/s at 250k context in NInfer, and Qwen3.8 Flash-Next on a DGX Spark at 50-100 t/s. Setup is NInfer with MTP speculation on the 5090; the DGX Spark run holds over 500k tokens of context. The user sees no quality gain from the larger Flash-Next model, since only 6B parameters activate per token, and asks how to split work between the two machines or whether to buy a second DGX Spark for DeepSeek or GLM.

Compare this run

Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.

Qwen3.8 125B · 6B active at 4-bit by hardware

Bars are medians on one scale, 0 to 232.0 tokens/s. Each dot is one run and opens its report. Press ▶ to watch text arrive at a row's speed.

Models most run on the NVIDIA RTX 5090

Bars are medians on one scale, 0 to 563.1 tokens/s. Each dot is one run and opens its report. Press ▶ to watch text arrive at a row's speed.

Everything reported on the NVIDIA RTX 5090 →Every Qwen3.8 report →Estimate Qwen3.8 on this card →