Back to search results

gemma-4-12B-it · Q4_K_M

Start a discussion
NVIDIA RTX 309024GBvLLM v0.41L2 Cross-framework verified

This page aggregates 2 real-world runs of gemma-4-12B-it (Q4_K_M) on NVIDIA RTX 3090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

100.66 tok/s

Decode

Decode speed

612.88 tok/s

Prefill

Prefill speed

1.05 s

TTFT

Time to first token

15.28 GB

VRAM

VRAM usage

L2 Cross-framework verified

2

Measured runs

1

Independent sources

Other

Source platforms

2026-10-10

Last verified

Performance

  1. llama.cpp · Q4_K_M Decode 79.26 · Prefill 473.41 ·
  2. Ollama · Q3_K_L Decode 64.09 · Prefill 302.97 ·
  3. vLLM · Q6_K Decode 50.6 · Prefill 208.67 ·
  4. Ollama · IQ3_XS Decode 67.34 · Prefill 278.97 ·
  5. vLLM · Q4_K_M (current)Decode 100.66 · Prefill 612.88 ·
  6. Ollama · Q2_K Decode 115.28 · Prefill 513.9 ·
  7. Ollama · Q4_K_S Decode 84.3 · Prefill 357.49 ·
  8. Strata · IQ2_M Decode 174.12 · Prefill 755.26 ·
  9. vLLM · Q8_0 Decode 61.2 · Prefill 459 ·
  10. vLLM · Q5_K_M Decode 89.4 · Prefill 670.5 ·
  11. llama.cpp · Q8_0 Decode 56.2 · Prefill 421.5 ·
  12. llama.cpp · Q5_K_M Decode 82 · Prefill 615 ·

Core figures

Decode (avg)
100.66 tok/s
Prefill (avg)
612.88 tok/s
TTFT (avg)
1.05 s
VRAM (avg)
15.28 GB
MTP acceptance rate
—
TTFB
0.74 GB
Power draw
430.5 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-12B-it
Quantization
Q4_K_M
Framework
vLLM
Version
v0.41
Context length
4096 tokens
Batch size
32
GPU layers
999
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)
15.28 GB
OS
Windows 11 23H2
Driver
565.57
CUDA
12.4
Power draw
430.5 W

Sources & evidence

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

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

    v0.41 · Windows 11 23H2 · CUDA 12.4 · 4096 ctx

    92.31 tok/s

    Decode

    408.25 tok/s

    Prefill

    1.98 s

    TTFT

    22.76 GB

    VRAM

    —

    MTP

    430.5 W

    Power draw

    L2 batch seed 2026-10-10: vLLM on hw_id=2 running model_id=65 at Q4_K_M
  2. L0 Self-reportedOtherOriginal link Verified on 2026-10-06

    0.7.3 · Linux · CUDA 12.6 · 4096 ctx

    109 tok/s

    Decode

    817.5 tok/s

    Prefill

    0.12 s

    TTFT

    7.8 GB

    VRAM

    —

    MTP

    — W

    Power draw

    L0 理论估算(add-1000-records-batch3)