Arao AI
Back to search results

Llama-3.2-3B-Instruct · IQ3_XS

Start a discussion
NVIDIA RTX PRO 6000 Blackwell 96GB96GBllama.cpp v0.40L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3.2-3B-Instruct (IQ3_XS) on NVIDIA RTX PRO 6000 Blackwell 96GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

338.03 tok/s

Decode

Decode speed

1,168.59 tok/s

Prefill

Prefill speed

0.68 s

TTFT

Time to first token

6.01 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. llama.cpp · IQ3_XS (current)Decode 338.03 · Prefill 1168.59 ·
  2. llama.cpp · Q8_0 Decode 331.08 · Prefill 1764.65 ·
  3. Ollama · IQ3_XS Decode 283.4 · Prefill 1452.77 ·
  4. Ollama · Q3_K_S Decode 364.08 · Prefill 1984.53 ·

Core figures

Decode (avg)
338.03 tok/s
Prefill (avg)
1,168.59 tok/s
TTFT (avg)
0.68 s
VRAM (avg)
6.01 GB
MTP acceptance rate
—
TTFB
2.7 GB
Power draw
301.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3.2-3B-Instruct
Quantization
IQ3_XS
Framework
llama.cpp
Version
v0.40
Context length
16384 tokens
Batch size
256
GPU layers
99
Flash Attention
Off

Hardware

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

GPU
NVIDIA RTX PRO 6000 Blackwell 96GB
Nominal VRAM
96 GB
Measured VRAM (avg)
6.01 GB
OS
Ubuntu 22.04
Driver
560.94
CUDA
12.8
Power draw
301.7 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.40 · Ubuntu 22.04 · CUDA 12.8 · 16384 ctx

    338.03 tok/s

    Decode

    1,168.59 tok/s

    Prefill

    0.68 s

    TTFT

    6.01 GB

    VRAM

    —

    MTP

    301.7 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=68 running model_id=70 at IQ3_XS