Arao AI
Back to search results

Llama 5 70B · IQ4_XS

Start a discussion
NVIDIA RTX 509032GBvLLM 0.7.3L0 Self-reported

This page aggregates 1 real-world runs of Llama 5 70B (IQ4_XS) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

40.3 tok/s

Decode

Decode speed

302.3 tok/s

Prefill

Prefill speed

0.33 s

TTFT

Time to first token

32 GB

VRAM

VRAM usage

L0 Self-reported

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. vLLM · IQ3_S Decode 53.6 · Prefill 402 ·
  2. vLLM · IQ4_XS (current)Decode 40.3 · Prefill 302.3 ·
  3. vLLM · Q6_K Decode 24.8 · Prefill 186 ·
  4. vLLM · Q5_K_M Decode 29.3 · Prefill 219.8 ·
  5. llama.cpp · IQ3_S Decode 49.2 · Prefill 369 ·
  6. llama.cpp · IQ4_XS Decode 37 · Prefill 277.5 ·
  7. llama.cpp · Q6_K Decode 22.8 · Prefill 171 ·
  8. llama.cpp · Q5_K_M Decode 26.9 · Prefill 201.8 ·
  9. vLLM · Q4_K_M Decode 35.8 · Prefill 268.5 ·

Core figures

Decode (avg)
40.3 tok/s
Prefill (avg)
302.3 tok/s
TTFT (avg)
0.33 s
VRAM (avg)
32 GB
MTP acceptance rate
—
TTFB
— GB
Power draw
— W

Configuration

Member-level fields are taken from the most recent run

Model
Llama 5 70B
Quantization
IQ4_XS
Framework
vLLM
Version
0.7.3
Context length
4096 tokens
Batch size
512
GPU layers
99
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)
32 GB
OS
Linux
Driver
570.65
CUDA
12.6

Sources & evidence

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

  1. L0 Self-reportedOtherOriginal link Verified on 2026-10-06

    0.7.3 · Linux · CUDA 12.6 · 4096 ctx

    40.3 tok/s

    Decode

    302.3 tok/s

    Prefill

    0.33 s

    TTFT

    32 GB

    VRAM

    —

    MTP

    — W

    Power draw

    L0 理论估算（add-1000-records-batch3）