Arao AI
Back to search results

DeepSeek-R1-Distill-Qwen-14B · IQ4_XS

Start a discussion
NVIDIA H100 80GB HBM380GBvLLM v0.25L2 Cross-framework verified

This page aggregates 1 real-world runs of DeepSeek-R1-Distill-Qwen-14B (IQ4_XS) on NVIDIA H100 80GB HBM3 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

398.24 tok/s

Decode

Decode speed

1,414.76 tok/s

Prefill

Prefill speed

2.13 s

TTFT

Time to first token

23.86 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. vLLM · IQ4_XS (current)Decode 398.24 · Prefill 1414.76 ·
  2. vLLM · Q5_K_S Decode 364.79 · Prefill 1787.53 ·
  3. Ollama · IQ2_XS Decode 291.58 · Prefill 1457.4 ·
  4. Ollama · IQ3_XS Decode 340.5 · Prefill 1221.18 ·

Core figures

Decode (avg)
398.24 tok/s
Prefill (avg)
1,414.76 tok/s
TTFT (avg)
2.13 s
VRAM (avg)
23.86 GB
MTP acceptance rate
—
TTFB
3.19 GB
Power draw
557.8 W

Configuration

Member-level fields are taken from the most recent run

Model
DeepSeek-R1-Distill-Qwen-14B
Quantization
IQ4_XS
Framework
vLLM
Version
v0.25
Context length
16384 tokens
Batch size
32
GPU layers
999
Flash Attention
Off

Hardware

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

GPU
NVIDIA H100 80GB HBM3
Nominal VRAM
80 GB
Measured VRAM (avg)
23.86 GB
OS
Ubuntu 22.04
Driver
560.94
CUDA
12.8
Power draw
557.8 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

    v0.25 · Ubuntu 22.04 · CUDA 12.8 · 16384 ctx

    398.24 tok/s

    Decode

    1,414.76 tok/s

    Prefill

    2.13 s

    TTFT

    23.86 GB

    VRAM

    —

    MTP

    557.8 W

    Power draw

    L2 batch seed 2026-10-19: vLLM on hw_id=55 running model_id=71 at IQ4_XS