Hy-Memory-30B-A3B · IQ3_XS
This page aggregates 1 real-world runs of Hy-Memory-30B-A3B (IQ3_XS) on NVIDIA RTX PRO 6000 Blackwell 96GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
355.22 tok/s
Decode
Decode speed
1,526.75 tok/s
Prefill
Prefill speed
2.48 s
TTFT
Time to first token
41.16 GB
VRAM
VRAM usage
1
Measured runs
1
Independent sources
Other
Source platforms
16 days ago
Last verified
Performance
- llama.cpp · IQ2_M Decode 366.29 · Prefill 1896.32 ·
- llama.cpp · IQ3_XS (current)Decode 355.22 · Prefill 1526.75 ·
- Strata · IQ2_XS Decode 349.83 · Prefill 1267.72 ·
Core figures
- Decode (avg)
- 355.22 tok/s
- Prefill (avg)
- 1,526.75 tok/s
- TTFT (avg)
- 2.48 s
- VRAM (avg)
- 41.16 GB
- MTP acceptance rate
- —
- TTFB
- 2.84 GB
- Power draw
- 336.2 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Hy-Memory-30B-A3B
- Quantization
- IQ3_XS
- Framework
- llama.cpp
- Version
- v0.40
- Context length
- 8192 tokens
- Batch size
- 256
- 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 PRO 6000 Blackwell 96GB
- Nominal VRAM
- 96 GB
- Measured VRAM (avg)
- 41.16 GB
- OS
- Windows 11 24H2
- Driver
- 565.57
- CUDA
- 12.8
- Power draw
- 336.2 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
v0.40 · Windows 11 24H2 · CUDA 12.8 · 8192 ctx
355.22 tok/s
Decode
1,526.75 tok/s
Prefill
2.48 s
TTFT
41.16 GB
VRAM
—
MTP
336.2 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=68 running model_id=87 at IQ3_XS