GLM-5.3-Flash · IQ3_S
This page aggregates 1 real-world runs of GLM-5.3-Flash (IQ3_S) on NVIDIA RTX 5070 Ti with SGLang, contributed by 1 independent source platforms; metrics are averages of published measurements.
5.7 tok/s
Decode
Decode speed
42.8 tok/s
Prefill
Prefill speed
2.34 s
TTFT
Time to first token
16 GB
VRAM
VRAM usage
1
Measured runs
1
Independent sources
Other
Source platforms
Today
Last verified
Performance
- SGLang · Q4_K_M Decode 3.8 · Prefill 28.5 ·
- SGLang · IQ3_S (current)Decode 5.7 · Prefill 42.8 ·
- SGLang · IQ4_XS Decode 4.3 · Prefill 32.3 ·
- vLLM · Q4_K_M Decode 3.9 · Prefill 29.3 ·
- vLLM · IQ3_S Decode 5.9 · Prefill 44.3 ·
- vLLM · IQ4_XS Decode 4.4 · Prefill 33 ·
- Strata · IQ2_XS Decode 8.3 · Prefill 62.3 ·
Core figures
- Decode (avg)
- 5.7 tok/s
- Prefill (avg)
- 42.8 tok/s
- TTFT (avg)
- 2.34 s
- VRAM (avg)
- 16 GB
- MTP acceptance rate
- —
- TTFB
- — GB
- Power draw
- — W
Configuration
Member-level fields are taken from the most recent run
- Model
- GLM-5.3-Flash
- Quantization
- IQ3_S
- Framework
- SGLang
- Version
- 0.4.4
- 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 5070 Ti
- Nominal VRAM
- 16 GB
- Measured VRAM (avg)
- 16 GB
- OS
- Linux
- Driver
- 570.65
- CUDA
- 12.6
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
0.4.4 · Linux · CUDA 12.6 · 4096 ctx
5.7 tok/s
Decode
42.8 tok/s
Prefill
2.34 s
TTFT
16 GB
VRAM
—
MTP
— W
Power draw
L0 理论估算（add-1000-records-batch3）