Arao AI
Back to search results

Qwen3.5-27B · Q4_K_M

Start a discussion
NVIDIA RTX 309024GBStrata v0.5.xL2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-27B (Q4_K_M) on NVIDIA RTX 3090 with Strata, contributed by 1 independent source platforms; metrics are averages of published measurements.

123 tok/s

Decode

Decode speed

1,415 tok/s

Prefill

Prefill speed

1.61 s

TTFT

Time to first token

15.5 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ4_XS Decode 91 · Prefill 1085 ·
  2. llama.cpp · Q4_K_M Decode 81 · Prefill 971 ·
  3. llama.cpp · Q5_K_M Decode 92 · Prefill 957 ·
  4. llama.cpp · Q8_0 Decode 68 · Prefill 642 ·
  5. vLLM · IQ4_XS Decode 88 · Prefill 969 ·
  6. vLLM · Q4_K_M Decode 81 · Prefill 868 ·
  7. vLLM · Q5_K_M Decode 79 · Prefill 714 ·
  8. vLLM · Q8_0 Decode 50 · Prefill 591 ·
  9. Ollama · IQ4_XS Decode 103 · Prefill 847 ·
  10. Ollama · Q4_K_M Decode 93 · Prefill 803 ·
  11. Ollama · Q5_K_M Decode 78 · Prefill 912 ·
  12. Ollama · Q8_0 Decode 62 · Prefill 699 ·
  13. Strata · IQ4_XS Decode 138 · Prefill 1557 ·
  14. Strata · Q4_K_M (current)Decode 123 · Prefill 1415 ·
  15. Strata · Q5_K_M Decode 122 · Prefill 1402 ·
  16. Strata · Q8_0 Decode 97 · Prefill 795 ·

Core figures

Decode (avg)
123 tok/s
Prefill (avg)
1,415 tok/s
TTFT (avg)
1.61 s
VRAM (avg)
15.5 GB
MTP acceptance rate
—
TTFB
— GB
Power draw
424 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-27B
Quantization
Q4_K_M
Framework
Strata
Version
v0.5.x
Context length
8192 tokens
Batch size
512
GPU layers
99
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 3090
Nominal VRAM
24 GB
Measured VRAM (avg)
15.5 GB
OS
Linux 6.8.0
Driver
570.86.15
CUDA
12.6
Power draw
424 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-09

    v0.5.x · Linux 6.8.0 · CUDA 12.6 · 8192 ctx

    123 tok/s

    Decode

    1,415 tok/s

    Prefill

    1.61 s

    TTFT

    15.5 GB

    VRAM

    —

    MTP

    424 W

    Power draw

    Auto-generated LLM-sourced record 2026-10-09. Model: Qwen3.5-27B, Hardware: RTX 3090, Framework: 9, Quant: Q4_K_M.