Arao AI
Back to search results

Llama-3.2-3B-Instruct · Q5_K_S

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.34L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3.2-3B-Instruct (Q5_K_S) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

373.63 tok/s

Decode

Decode speed

1,720.71 tok/s

Prefill

Prefill speed

1.98 s

TTFT

Time to first token

7.77 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 · Q4_K_S Decode 375.58 · Prefill 1439.62 ·
  2. llama.cpp · Q5_K_S (current)Decode 373.63 · Prefill 1720.71 ·
  3. vLLM · IQ4_XS Decode 321.22 · Prefill 1224.88 ·
  4. vLLM · Q6_K Decode 292.37 · Prefill 1232.31 ·
  5. llama.cpp · Q2_K Decode 325.57 · Prefill 1657.68 ·
  6. Strata · Q4_K_M Decode 367.38 · Prefill 1255.45 ·
  7. Strata · Q5_K_S Decode 289.95 · Prefill 1022.85 ·
  8. llama.cpp · Q8_0 Decode 436.8 · Prefill 3276 ·
  9. llama.cpp · Q5_K_M Decode 635.3 · Prefill 4764.8 ·
  10. llama.cpp · Q4_K_M Decode 776.5 · Prefill 5823.8 ·

Core figures

Decode (avg)
373.63 tok/s
Prefill (avg)
1,720.71 tok/s
TTFT (avg)
1.98 s
VRAM (avg)
7.77 GB
MTP acceptance rate
—
TTFB
2.99 GB
Power draw
366.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3.2-3B-Instruct
Quantization
Q5_K_S
Framework
llama.cpp
Version
v0.34
Context length
2048 tokens
Batch size
512
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
NVIDIA RTX 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
7.77 GB
OS
Windows 11 24H2
Driver
570.86
CUDA
12.8
Power draw
366.9 W

Sources & evidence

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

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

    v0.34 · Windows 11 24H2 · CUDA 12.8 · 2048 ctx

    373.63 tok/s

    Decode

    1,720.71 tok/s

    Prefill

    1.98 s

    TTFT

    7.77 GB

    VRAM

    —

    MTP

    366.9 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=13 running model_id=70 at Q5_K_S