Hi everyone. I'm actually new to AI image generation, but my sales team wanted a custom solution for creating marketing visuals. I managed to set up a local Stable Diffusion node for us.
The total hardware came to about $3,800. The big parts were an RTX 4090 (ouch, my budget 😅) and a solid CPU. I went with 64GB RAM because I read it helps with batch processing.
I'm a bit nervous about whether I over-spec'd it. For 5-10 users generating maybe 50-100 images a day, is this enough? Also, are there any obvious bottlenecks in my setup I should fix before we go live? Any advice is super appreciated!
Your hardware spec is more than sufficient, possibly even overkill for that daily image count. The RTX 4090 is the main workhorse, and its 24GB VRAM will let you run larger models or generate batches without issue. The 64GB RAM is a good future-proofing choice.
The bottleneck likely won't be hardware, but your software stack and how you manage concurrent requests. For 5-10 users, you'll need a proper queuing system in your web UI (like Automatic1111 or ComfyUI) to handle simultaneous jobs gracefully. Without that, you'll get memory conflicts even with that GPU.
You might want to run a simple benchmark to set expectations. Time how long it takes to generate a 512x512 image with a standard SD 1.5 model, then extrapolate for your team's typical workload. That will give you a concrete metric for throughput.
BenchMark