FLUX.2 [klein] 9B on Mac mini M4 24 GB: VRAM, speed and fit
FLUX.2 [klein] 9B at 4-bit runs on the M4 24 GB, about 2.1 min per image.
Image generation · 1024 × 1024
Tight fit · on M4 24 GB · 4-bit (GGUF Q4_K_M · NVFP4)
99% used
- GPU memory
- 23.8 GB
- of 24.0 GB
- Time
- 2.1 min
- per image
Where the memory goes
- Model
- 12.7 GB
- Working
- 2.3 GB
- Overhead
- 819 MB
- OS
- 8.0 GB
- Free
- 159 MB
- Memory
- 23.8 GB of 24.0 GB
- Output
- 1024 × 1024 · 4 steps
- Time
- 2.1 min per image
- Running cost
- $1.28/mo at home · $29/mo on a cloud RTX 2000 Ada
Validation: Not validated
| Check | Target | Result | Status |
|---|---|---|---|
| Fits in GPU memory | 16.0 GB usable | 15.8 GB needed | Pass |
| Headroom | 5% free | 159 MB free (1%) | Check |
| Time per image | under 30 s | 2.1 min | Fails |
Targets: an image in under 30 s, and 5% of memory left free. Calculated from published specs, not benchmarked.
The setup
- Model
- FLUX.2 [klein] 9B
- Parameters
- 9.1B
- Precision
- 4-bit (GGUF Q4_K_M · NVFP4)
- Output
- 1024 × 1024
- Steps
- 4
- Hardware
- M4 24 GB
- GPU can use
- 16.0 GB of 24.0 GB
Worth trying
- 768² output: Fits, 73 s an image
- RTX PRO 2000 Blackwell: Runs it, 7.9 s an image
- M5 Max · 36 GB: Runs it, 9.9 s an image
- Ryzen AI Max+ 395 · 32 GB: Runs it, 15 s an image
Running cost
- At home
- $1.28/mo
- 59 W at the wall, under $0.001 an image
- Renting instead
- $29/mo
- RTX 2000 Ada at $0.24/h, under $0.001 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
FLUX.2 [klein] 9B on similar machines
- RTX PRO 6000 BlackwellRuns
- Mac mini M4 16 GBToo big
- Mac mini M4 32 GBRuns
- Mac mini M4 Pro 24 GBRuns
- Mac mini M4 Pro 48 GBRuns
- Mac mini M4 Pro 64 GBRuns
All machines: FLUX.2 [klein] 9B VRAM requirements. Everything that runs here: what can the Mac mini M4 24 GB run?
Other models that run on Mac mini M4 24 GB
Questions
Can FLUX.2 [klein] 9B run on Mac mini M4 24 GB?
FLUX.2 [klein] 9B at 4-bit runs on the M4 24 GB, about 2.1 min per image.
Report VR-0WD6OO6, calculated Nov 22, 2026 from published specs, not benchmarked.