Arao AI
Back to search results

gemma-4-26B-A4B-it · IQ2_XS

Start a discussion
AMD Ryzen AI Max+ 395128GBllama.cpp v0.21L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-26B-A4B-it (IQ2_XS) on AMD Ryzen AI Max+ 395 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

92.57 tok/s

Decode

Decode speed

496.15 tok/s

Prefill

Prefill speed

1.77 s

TTFT

Time to first token

28.31 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

5 days ago

Last verified

Performance

  1. llama.cpp · IQ2_XS (current)Decode 92.57 · Prefill 496.15 ·
  2. llama.cpp · IQ3_XS Decode 81.12 · Prefill 341.45 ·
  3. Ollama · IQ3_XS Decode 79.36 · Prefill 408.93 ·

Core figures

Decode (avg)
92.57 tok/s
Prefill (avg)
496.15 tok/s
TTFT (avg)
1.77 s
VRAM (avg)
28.31 GB
MTP acceptance rate
—
TTFB
0.74 GB
Power draw
125 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-26B-A4B-it
Quantization
IQ2_XS
Framework
llama.cpp
Version
v0.21
Context length
16384 tokens
Batch size
256
GPU layers
20
Flash Attention
On

Hardware

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

GPU
AMD Ryzen AI Max+ 395
Nominal VRAM
128 GB
Measured VRAM (avg)
28.31 GB
OS
Windows 11 23H2
Driver
AMD Adrenalin 25.10.1
CUDA
N/A
Power draw
125 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-13

    v0.21 · Windows 11 23H2 · CUDA N/A · 16384 ctx

    92.57 tok/s

    Decode

    496.15 tok/s

    Prefill

    1.77 s

    TTFT

    28.31 GB

    VRAM

    —

    MTP

    125 W

    Power draw

    L2 batch seed 2026-10-13: llama.cpp on hw_id=24 running model_id=64 at IQ2_XS