Arao AI
Back to search results

gemma-4-12B-it · IQ3_S

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.29L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-12B-it (IQ3_S) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

214.27 tok/s

Decode

Decode speed

831.85 tok/s

Prefill

Prefill speed

1.2 s

TTFT

Time to first token

18.73 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ2_XS Decode 253.43 · Prefill 894.02 ·
  2. llama.cpp · IQ3_S (current)Decode 214.27 · Prefill 831.85 ·
  3. vLLM · IQ1_S Decode 298.32 · Prefill 1152.98 ·
  4. vLLM · IQ2_M Decode 288.51 · Prefill 940.03 ·
  5. Ollama · IQ2_XS Decode 246.6 · Prefill 1083.78 ·
  6. vLLM · Q8_0 Decode 117.2 · Prefill 879 ·
  7. vLLM · Q5_K_M Decode 171.1 · Prefill 1283.3 ·
  8. vLLM · Q4_K_M Decode 208.7 · Prefill 1565.3 ·
  9. llama.cpp · Q8_0 Decode 107.5 · Prefill 806.3 ·
  10. llama.cpp · Q5_K_M Decode 157.1 · Prefill 1178.3 ·
  11. llama.cpp · Q4_K_M Decode 191.5 · Prefill 1436.3 ·

Core figures

Decode (avg)
214.27 tok/s
Prefill (avg)
831.85 tok/s
TTFT (avg)
1.2 s
VRAM (avg)
18.73 GB
MTP acceptance rate
—
TTFB
1.71 GB
Power draw
438.1 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-12B-it
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.29
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 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
18.73 GB
OS
Ubuntu 22.04
Driver
570.86
CUDA
12.8
Power draw
438.1 W

Sources & evidence

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

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

    v0.29 · Ubuntu 22.04 · CUDA 12.8 · 8192 ctx

    214.27 tok/s

    Decode

    831.85 tok/s

    Prefill

    1.2 s

    TTFT

    18.73 GB

    VRAM

    —

    MTP

    438.1 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=13 running model_id=65 at IQ3_S