Arao AI
Back to search results

Llama-3.2-3B-Instruct · FP16

Start a discussion
NVIDIA RTX 4080 Super16GBStrata v0.28L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3.2-3B-Instruct (FP16) on NVIDIA RTX 4080 Super with Strata, contributed by 1 independent source platforms; metrics are averages of published measurements.

277.08 tok/s

Decode

Decode speed

1,141.1 tok/s

Prefill

Prefill speed

0.67 s

TTFT

Time to first token

13.76 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 · IQ2_XS Decode 372.98 · Prefill 1597.99 ·
  2. llama.cpp · IQ4_XS Decode 289.96 · Prefill 1071.96 ·
  3. Strata · FP16 (current)Decode 277.08 · Prefill 1141.1 ·
  4. vLLM · IQ3_S Decode 334.5 · Prefill 1159.32 ·
  5. Strata · Q3_K_L Decode 358.87 · Prefill 1331.79 ·

Core figures

Decode (avg)
277.08 tok/s
Prefill (avg)
1,141.1 tok/s
TTFT (avg)
0.67 s
VRAM (avg)
13.76 GB
MTP acceptance rate
0.61
TTFB
2.22 GB
Power draw
308 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3.2-3B-Instruct
Quantization
FP16
Framework
Strata
Version
v0.28
Context length
16384 tokens
Batch size
1024
GPU layers
12
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 4080 Super
Nominal VRAM
16 GB
Measured VRAM (avg)
13.76 GB
OS
Windows 11 24H2
Driver
560.94
CUDA
12.8
Power draw
308 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

    v0.28 · Windows 11 24H2 · CUDA 12.8 · 16384 ctx

    277.08 tok/s

    Decode

    1,141.1 tok/s

    Prefill

    0.67 s

    TTFT

    13.76 GB

    VRAM

    0.61

    MTP

    308 W

    Power draw

    L2 batch seed 2026-10-19: Strata on hw_id=20 running model_id=70 at FP16