Arao AI
Back to search results

Qwen3.8-27B · UD-Q8_K_XL

2x NVIDIA RTX 309048GBllama.cpp b10236L0 Self-reported

This page aggregates 1 real-world runs of Qwen3.8-27B (UD-Q8_K_XL) on 2x NVIDIA RTX 3090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

43.65 tok/s

Decode

Decode speed

1,600.18 tok/s

Prefill

Prefill speed

0.42 s

TTFT

Time to first token

44.12 GB

VRAM

VRAM usage

L0 Self-reported

1

Measured runs

1

Independent sources

GitHub

Source platforms

Today

Last verified

Performance

  1. llama.cpp · UD-Q8_K_XL (current)Decode 43.65 · Prefill 1600.18 ·

Core figures

Decode (avg)
43.65 tok/s
Prefill (avg)
1,600.18 tok/s
TTFT (avg)
0.42 s
VRAM (avg)
44.12 GB
MTP acceptance rate
0.82
TTFB
— GB
Power draw
— W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-27B
Quantization
UD-Q8_K_XL
Framework
llama.cpp
Version
b10236
Context length
262144 tokens
Batch size
512
GPU layers
99

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
2x NVIDIA RTX 3090
Nominal VRAM
48 GB
Measured VRAM (avg)
44.12 GB
OS
Ubuntu 26.04 LTS
Driver
610.57.04
CUDA
13.3

Get started

Startup command

Command from the latest benchmark run
model=/models/qwen3.8-27b/Qwen3.8-27B-UD-Q8_K_XL.gguf ubatch=512 kv=q8_0 ngl=99 split=0.55,0.45

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L0 Self-reportedGitHubOriginal link Verified on 2026-09-20

    b10236 · Ubuntu 26.04 LTS · CUDA 13.3 · 262144 ctx

    43.65 tok/s

    Decode

    1,600.18 tok/s

    Prefill

    0.42 s

    TTFT

    44.12 GB

    VRAM

    0.82

    MTP

    — W

    Power draw

    [bench] narrative run 1/5: 1000 tok in 23.5s (43.22 decode TPS, ttft 394ms) [bench] narrative run 2/5: 1000 tok in 23.1s (44.20 decode TPS, ttft 516ms) [bench] narrative run 3/5: 1000 tok in 23.2s (43.81 decode TPS, ttft 395ms) [bench] narrative run 4/5: 991 tok in 23.3s (43.24 decode TPS, ttft 395ms) [bench] narrative run 5/5: 964 tok in 22.4s (43.77 decode TPS, ttft 392ms) [bench] prompt-processing run 1/1: 10004 prompt tok, 1600.2 PP tok/s, ttft 6252ms