Arao AI
Back to search results

Ministral-3-8B-Instruct-2512 · Q4_K_M

Start a discussion
NVIDIA H2097GBStrata v0.29L2 Cross-framework verified

This page aggregates 1 real-world runs of Ministral-3-8B-Instruct-2512 (Q4_K_M) on NVIDIA H20 with Strata, contributed by 1 independent source platforms; metrics are averages of published measurements.

73.54 tok/s

Decode

Decode speed

286.94 tok/s

Prefill

Prefill speed

1.72 s

TTFT

Time to first token

15.88 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

11 days ago

Last verified

Performance

  1. vLLM · Q5_K_S Decode 49.64 · Prefill 220.24 ·
  2. Strata · Q4_K_M (current)Decode 73.54 · Prefill 286.94 ·

Core figures

Decode (avg)
73.54 tok/s
Prefill (avg)
286.94 tok/s
TTFT (avg)
1.72 s
VRAM (avg)
15.88 GB
MTP acceptance rate
0.75
TTFB
0.59 GB
Power draw
359.1 W

Configuration

Member-level fields are taken from the most recent run

Model
Ministral-3-8B-Instruct-2512
Quantization
Q4_K_M
Framework
Strata
Version
v0.29
Context length
65536 tokens
Batch size
1024
GPU layers
5
Flash Attention
On

Hardware

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

GPU
NVIDIA H20
Nominal VRAM
97 GB
Measured VRAM (avg)
15.88 GB
OS
Ubuntu 22.04
Driver
572.83
CUDA
12.8
Power draw
359.1 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-24

    v0.29 · Ubuntu 22.04 · CUDA 12.8 · 65536 ctx

    73.54 tok/s

    Decode

    286.94 tok/s

    Prefill

    1.72 s

    TTFT

    15.88 GB

    VRAM

    0.75

    MTP

    359.1 W

    Power draw

    L2 batch seed 2026-10-24: Strata on hw_id=56 running model_id=74 at Q4_K_M