Arao AI
Back to search results

Hy-Memory-30B-A3B · Q6_K

Start a discussion
NVIDIA H2097GBllama.cpp v0.44L2 Cross-framework verified

This page aggregates 1 real-world runs of Hy-Memory-30B-A3B (Q6_K) on NVIDIA H20 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

57.59 tok/s

Decode

Decode speed

253.97 tok/s

Prefill

Prefill speed

1.63 s

TTFT

Time to first token

65.1 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 · Q6_K (current)Decode 57.59 · Prefill 253.97 ·
  2. Strata · IQ1_S Decode 178.17 · Prefill 571.19 ·
  3. Strata · IQ3_XS Decode 139.08 · Prefill 595.25 ·

Core figures

Decode (avg)
57.59 tok/s
Prefill (avg)
253.97 tok/s
TTFT (avg)
1.63 s
VRAM (avg)
65.1 GB
MTP acceptance rate
—
TTFB
0.46 GB
Power draw
326.4 W

Configuration

Member-level fields are taken from the most recent run

Model
Hy-Memory-30B-A3B
Quantization
Q6_K
Framework
llama.cpp
Version
v0.44
Context length
2048 tokens
Batch size
256
GPU layers
28
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)
65.1 GB
OS
Ubuntu 22.04
Driver
560.94
CUDA
12.4
Power draw
326.4 W

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.44 · Ubuntu 22.04 · CUDA 12.4 · 2048 ctx

    57.59 tok/s

    Decode

    253.97 tok/s

    Prefill

    1.63 s

    TTFT

    65.1 GB

    VRAM

    —

    MTP

    326.4 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=56 running model_id=87 at Q6_K