Arao AI
Back to search results

gemma-4-12B-it · IQ2_XS

Start a discussion
NVIDIA RTX 409024GBllama.cpp v0.20L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-12B-it (IQ2_XS) on NVIDIA RTX 4090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

152.32 tok/s

Decode

Decode speed

683.44 tok/s

Prefill

Prefill speed

0.32 s

TTFT

Time to first token

14.2 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 (current)Decode 152.32 · Prefill 683.44 ·
  2. Strata · Q5_K_S Decode 117.55 · Prefill 384.33 ·
  3. llama.cpp · IQ3_XS Decode 114.16 · Prefill 511.31 ·
  4. llama.cpp · Q4_K_S Decode 100.24 · Prefill 360.86 ·
  5. vLLM · Q2_K Decode 139.41 · Prefill 531.29 ·
  6. Ollama · IQ4_XS Decode 95.26 · Prefill 447.14 ·
  7. Ollama · Q6_K Decode 71.51 · Prefill 296.42 ·
  8. vLLM · Q8_0 Decode 65.9 · Prefill 494.3 ·
  9. vLLM · Q5_K_M Decode 96.3 · Prefill 722.3 ·
  10. vLLM · Q4_K_M Decode 117.4 · Prefill 880.5 ·
  11. llama.cpp · Q8_0 Decode 60.5 · Prefill 453.8 ·
  12. llama.cpp · Q5_K_M Decode 88.3 · Prefill 662.3 ·
  13. llama.cpp · Q4_K_M Decode 107.7 · Prefill 807.8 ·

Core figures

Decode (avg)
152.32 tok/s
Prefill (avg)
683.44 tok/s
TTFT (avg)
0.32 s
VRAM (avg)
14.2 GB
MTP acceptance rate
—
TTFB
1.22 GB
Power draw
310.9 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-12B-it
Quantization
IQ2_XS
Framework
llama.cpp
Version
v0.20
Context length
16384 tokens
Batch size
256
GPU layers
33
Flash Attention
Off

Hardware

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

GPU
NVIDIA RTX 4090
Nominal VRAM
24 GB
Measured VRAM (avg)
14.2 GB
OS
Ubuntu 22.04
Driver
560.94
CUDA
12.8
Power draw
310.9 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-18

    v0.20 · Ubuntu 22.04 · CUDA 12.8 · 16384 ctx

    152.32 tok/s

    Decode

    683.44 tok/s

    Prefill

    0.32 s

    TTFT

    14.2 GB

    VRAM

    —

    MTP

    310.9 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=1 running model_id=65 at IQ2_XS