llamaperf

Qwen3.8 27B

on 2× NVIDIA RTX 5060 Ti 16GB · NInfer · 210,000 ctx

Tone: positiveNVIDIA hardware
Nov 9, 2026
Throughput
75.0 t/s gen · 2100.0 tokens/s prompt processing (prefill)Prompt processing is how fast the input is read, not how long you wait for the first word. How to measure time to first token.
Quant
NVFP4 (NVFP4)
KV cache
int8
System RAM
32 GB
VRAM reported
32 GB

Use cases

codingagentic

Summary

User reports Qwen3.8 27B at ~75 t/s decode and ~2100 t/s prefill on two RTX 5060 Ti 16GB cards (32GB total VRAM). Setup is NInfer with an NVFP4 model, int8 KV cache, 210000-token context, tensor parallel 2, and MTP speculative decoding with 3 draft tokens. At 190K-200K context decode holds around 70 t/s and prefill around 1700 t/s; with MTP on coding tasks decode reaches 80-100 t/s. Cards are capped near 160W each and draw 290-300W total; the user notes WSL Docker overhead and that native Linux might allow the full 260K context.

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 27B at 4-bit by hardware

Bars are medians on one scale, 0 to 94.0 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 5060 Ti 16GB

Bars are medians on one scale, 0 to 95.2 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 5060 Ti 16GB →Every Qwen3.8 report →Estimate Qwen3.8 on this card →