Skip to content
llm-speed

Measurement wanted · request DR-022 · opened 2026-11-06

Qwen3.8-27B on RTX 5090: runs are posted

This pair has been measured. Open them below, or see every posted run on the Qwen3.8-27B model page. More runs from other machines still help.

What's being requested

Model
Qwen3.8-27B · unsloth/Qwen3.8-27B-GGUF
Hardware
RTX 5090

Runs we already have

  • r_zhzepwtpurt · RTX 5090 (32GB) · 68.0 tok/s decode (long-context-decay) · Measured by llm-speed (@llmspeed)
  • r_exg-o5xt410 · RTX 5090 (32GB) · 71.3 tok/s decode (chat-short) · Measured by llm-speed (@llmspeed)
  • r_7nl1vae_gek · RTX 5090 (32GB) · 18.6 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_dzvdfwvwcbr · RTX 5090 (32GB) · 35.8 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_vizpaqn9m8c · RTX 5090 (32GB) · 65.0 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_hwqgx8_47vr · RTX 5090 (32GB) · 17.5 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_-93wbtb83mc · RTX 5090 (32GB) · 25.8 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_3ogoh9fqwyn · RTX 5090 (32GB) · 44.2 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_a9_1mrekhnd · RTX 5090 (32GB) · 67.8 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_i1r0ngxz4hh · RTX 5090 (32GB) · 17.3 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_4nge7d__8-x · RTX 5090 (32GB) · 30.3 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_gihyz5jfgc8 · RTX 5090 (32GB) · 46.1 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_1xnjfp4smsq · RTX 5090 (32GB) · 45.3 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_w1ghhizti-d · RTX 5090 (32GB) · 58.2 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_iqcsaltiygj · RTX 5090 (32GB) · 71.0 tok/s decode (concurrent-decode) · Measured by llm-speed (@llmspeed)
  • r_63ox3vvpwej · RTX 5090 (32GB) · 70.3 tok/s decode (chat-short) · Community-submitted
  • r_91l46y770n8 · RTX 5090 (32GB) · 68.8 tok/s decode (chat-short) · Community-submitted
  • r_9vytm8nv49s · RTX 5090 (32GB) · 71.5 tok/s decode (long-context-decay) · Measured by llm-speed (@llmspeed)
  • r_oifzo7k_bs2 · RTX 5090 (32GB) · 76.9 tok/s decode (long-context-decay) · Measured by llm-speed (@llmspeed)
  • r_pfun72hu7dz · RTX 5090 (32GB) · 79.5 tok/s decode (chat-short) · Measured by llm-speed (@llmspeed)

Run it yourself

Start llama-server with the model (flags below are a starting point; adjust offload to your machine), then point llm-speed at it. Add --llama-args='...' with the flags you used so the run records them.

pipx install llm-speed   # or: pipx upgrade llm-speed
# any recent llama.cpp release
llama-server -m /path/to/MODEL.gguf --port 8080 -ngl 999
# then, in a second terminal:
LLAMA_CPP_SERVER_URL=http://127.0.0.1:8080 \
  llm-speed bench --backend llama.cpp \
  --model /path/to/MODEL.gguf \
  --workload chat-short

The upload is signed and public. How contributing works · what the CLI sends.

Get an alert

Email alerts for this page are switching on soon. Until then, follow @llmspeed on X, where new runs are posted.

All requested runs