← Blog
26.09.17 - How Much Faster Are Two RTX 5090s? A 1-GPU vs 2-GPU AI Benchmark with GDEP Solutions
With GDEP Solutions we compared AI image and video generation on one vs two GPUs — advanced image editing ran about 6.9× faster.
On GDEP Solutions' liquid-cooled "AquSys" workstation (two GeForce RTX 5090s), we measured how AI generation speed changes with one vs two GPUs, comparing ten image and video workflows in ComfyUI with Raylight.
Key results (1 GPU → 2 GPUs)
- Advanced image editing: 38.11 s → 5.49 s (about 6.94×)
- Avatar video from image + audio: 163.46 s → 52.75 s (about 3.10×)
- Multi-angle image generation: 202.85 s → 120.65 s (about 1.68×)
- Removing people from video: 669.39 s → 410.45 s (about 1.63×)
The main gain comes from splitting large models across two GPUs, which avoids memory offloading. The data helps decide how many GPUs each type of work really needs.
▶ Read the article: GDEP Solutions | 1 GPU vs 2 GPU AI benchmark