Qwen-Image on H100 SXM: VRAM, speed and fit
Qwen-Image at 16-bit runs on the H100 SXM, about 36 s per image.
Image generation · 1328 × 1328
Runs comfortably · on H100 SXM · 16-bit (BF16)
- GPU memory
- 39.2 GB
- of 80.0 GB
- Time
- 36 s
- per image
Where the memory goes
- Model
- 38.1 GB
- Working
- 347 MB
- Overhead
- 819 MB
- OS
- 0.0 GB
- Free
- 40.8 GB
- Memory
- 39.2 GB of 80.0 GB
- Output
- 1328 × 1328 · 50 steps
- Time
- 36 s per image
- Running cost
- $16/mo at home · $72/mo on a cloud A40
Validation: Validated with conditions
| Check | Target | Result | Status |
|---|---|---|---|
| Fits in GPU memory | 80.0 GB usable | 39.2 GB needed | Pass |
| Headroom | 5% free | 40.8 GB free (51%) | Pass |
| Time per image | under 30 s | 36 s | Check |
Targets: an image in under 30 s, and 5% of memory left free. Calculated from published specs, not benchmarked.
The setup
- Model
- Qwen-Image
- Parameters
- 20.4B
- Precision
- 16-bit (BF16)
- Output
- 1328 × 1328
- Steps
- 50
- Hardware
- H100 SXM
- GPU can use
- 80.0 GB of 80.0 GB
Running cost
- At home
- $16/mo
- 710 W at the wall, $0.001 an image
- Renting instead
- $72/mo
- A40 at $0.59/h, $0.04 an image
4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 21, 2026.
Where to get it
Qwen-Image on similar machines
- A100 80 GB SXMRuns
All machines: Qwen-Image VRAM requirements. Everything that runs here: what can the H100 SXM run?
Other models that run on H100 SXM
Questions
Can Qwen-Image run on H100 SXM?
Qwen-Image at 16-bit runs on the H100 SXM, about 36 s per image.
Report VR-1G4PV74, calculated Nov 22, 2026 from published specs, not benchmarked.