Arao AI
Back to search results

DeepSeek-R1-Distill-Qwen-14B · IQ3_S

Start a discussion
NVIDIA RTX PRO 6000 Blackwell 96GB96GBvLLM v0.53L2 Cross-framework verified

This page aggregates 1 real-world runs of DeepSeek-R1-Distill-Qwen-14B (IQ3_S) on NVIDIA RTX PRO 6000 Blackwell 96GB with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

236.15 tok/s

Decode

Decode speed

1,107.64 tok/s

Prefill

Prefill speed

1.28 s

TTFT

Time to first token

21.5 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 · Q5_K_S Decode 172.78 · Prefill 845.2 ·
  2. vLLM · IQ3_S (current)Decode 236.15 · Prefill 1107.64 ·

Core figures

Decode (avg)
236.15 tok/s
Prefill (avg)
1,107.64 tok/s
TTFT (avg)
1.28 s
VRAM (avg)
21.5 GB
MTP acceptance rate
—
TTFB
1.89 GB
Power draw
340 W

Configuration

Member-level fields are taken from the most recent run

Model
DeepSeek-R1-Distill-Qwen-14B
Quantization
IQ3_S
Framework
vLLM
Version
v0.53
Context length
8192 tokens
Batch size
8
GPU layers
999
Flash Attention
On

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)
21.5 GB
OS
Windows 11 24H2
Driver
560.94
CUDA
12.4
Power draw
340 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

    v0.53 · Windows 11 24H2 · CUDA 12.4 · 8192 ctx

    236.15 tok/s

    Decode

    1,107.64 tok/s

    Prefill

    1.28 s

    TTFT

    21.5 GB

    VRAM

    —

    MTP

    340 W

    Power draw

    L2 batch seed 2026-10-19: vLLM on hw_id=68 running model_id=71 at IQ3_S