An image every 10.8 seconds. That is what Flux.1 delivers on NVIDIA’s flagship RTX 5090 — more than double the throughput of last-generation cards running the same model. We benchmarked this combination on a GigaGPU dedicated server to find out whether the premium price tag translates into real production value.
Raw Numbers: 1024×1024 Generation
| Metric | Value |
|---|---|
| Iterations/sec | 1.85 it/s |
| Seconds per image | 10.81 sec (20 steps) |
| Images per minute | 5.55 |
| Resolution | 1024×1024 |
| Sampler | Euler a / DPM++ 2M Karras |
| Performance rating | Very Good |
Benchmark conditions: 20-step generation at 1024×1024, batch size 1, FP16 precision. Using A1111 WebUI or ComfyUI backend.
32 GB VRAM: Where It Gets Interesting
| Component | VRAM |
|---|---|
| Model weights | 12.0 GB |
| Sampling buffer | ~2.4 GB |
| Total RTX 5090 VRAM | 32 GB |
| Free headroom | ~20.0 GB |
Twenty gigabytes of headroom after the model loads. That is enough to run Flux.1 alongside a full Stable Diffusion pipeline, add multiple ControlNet models, or push resolution well beyond 2048×2048 for hero images. No other consumer-tier card comes close to this kind of multi-model flexibility with Flux.1 resident.
Production Economics
| Cost Metric | Value |
|---|---|
| Server cost | £1.50/hr (£299/mo) |
| Cost per 1K images | £4.50 |
| Images per £1 | 222 |
At £4.50 per thousand images, the 5090 sits between the RTX 4060 tier and professional A-series cards on cost. What you gain is throughput and headroom — critical for teams doing high-resolution work or running complex ComfyUI workflows. See the benchmark comparison for a direct cost-per-image table.
When the 5090 Makes Sense for Flux.1
If your daily volume exceeds a few hundred images, or you need the flexibility to run ControlNet, upscalers, and Flux.1 simultaneously, the 5090 is the card to pick. For lighter workloads, the RTX 3090 at £149/mo offers better value per pound. Read more in our Flux.1 hosting guide and GPU selection comparison.
Quick deploy:
docker run --gpus all -p 8188:8188 ghcr.io/ai-dock/comfyui:latestBrowse all benchmark results or explore PaddleOCR hosting for document AI pipelines.
Deploy Flux.1 on RTX 5090
Order this exact configuration. UK datacenter, full root access.
Order RTX 5090 Server
