Flux.1 from Black Forest Labs demands serious VRAM. Its 12 GB model footprint disqualifies most consumer GPUs outright. But the RTX 3090 and its 24 GB frame buffer? That is a different story entirely. We loaded Flux.1 onto a dedicated 3090 server via GigaGPU and measured exactly what you can expect in production.
Measured Throughput at 1024×1024
| Metric | Value |
|---|---|
| Iterations/sec | 0.82 it/s |
| Seconds per image | 24.39 sec (20 steps) |
| Images per minute | 2.46 |
| Resolution | 1024×1024 |
| Sampler | Euler a / DPM++ 2M Karras |
| Performance rating | Good |
Benchmark conditions: 20-step generation at 1024×1024, batch size 1, FP16 precision. Using A1111 WebUI or ComfyUI backend.
How VRAM Breaks Down
| Component | VRAM |
|---|---|
| Model weights | 12.0 GB |
| Sampling buffer | ~2.4 GB |
| Total RTX 3090 VRAM | 24 GB |
| Free headroom | ~12.0 GB |
After Flux.1 loads, you still have roughly 12 GB free. That is enough to run a ControlNet stack, bump resolution to 2048×2048 for select outputs, or even co-host a lightweight Whisper instance for multi-modal pipelines. Few other cards at this price point give you that kind of flexibility with a 12 GB model already resident.
What It Costs per Image
| Cost Metric | Value |
|---|---|
| Server cost | £0.75/hr (£149/mo) |
| Cost per 1K images | £5.08 |
| Images per £1 | 197 |
At £5.08 per thousand images, the RTX 3090 undercuts most commercial image APIs by a wide margin. For studios generating a few hundred images a day, this is the sweet spot between cost and capability. Compare other GPUs on our benchmark dashboard.
Who Should Use This Configuration
The 3090 is not the fastest Flux.1 card we have tested, but its 24 GB VRAM buffer makes it the most versatile in the mid-range. If your workflow involves ControlNet, inpainting, or batching multiple resolutions, the headroom matters more than raw speed. For pure throughput at higher volume, take a look at the best GPU for image generation comparison.
Quick deploy:
docker run --gpus all -p 8188:8188 ghcr.io/ai-dock/comfyui:latest
Further reading: Flux.1 hosting guide, all benchmark results, and Stable Diffusion hosting for comparing older architectures.
Deploy Flux.1 on RTX 3090
Order this exact configuration. UK datacenter, full root access.
Order RTX 3090 Server