Table of Contents
Quick Verdict
Training a machine learning classifier on synthetic images needs volume, not gallery-quality output. SD 1.5 at £1.41 per 1,000 images versus SDXL at £1.72 is an 18% cost saving — and SD 1.5’s smaller 4 GB VRAM footprint means you could run multiple instances on a single GPU for even higher throughput on a dedicated GPU server.
SDXL’s superior prompt adherence and visual detail matter for customer-facing outputs but add no value when images are training data or placeholder content. For batch generation at scale, SD 1.5 is the more economical engine.
Full data below. More at the GPU comparisons hub.
Specs Comparison
SD 1.5 generates at 512×512 by default while SDXL targets 1024×1024. If your batch pipeline upscales afterward, SD 1.5 combined with a super-resolution model can sometimes match SDXL quality at lower total cost.
| Specification | SD 1.5 | SDXL |
|---|---|---|
| Parameters | 860M (UNet) | 3.5B (UNet) |
| Architecture | Latent Diffusion | Latent Diffusion |
| Context Length | 512×512 | 1024×1024 |
| VRAM (FP16) | 4 GB | 7 GB |
| VRAM (INT4) | N/A | N/A |
| Licence | CreativeML Open RAIL-M | CreativeML Open RAIL++-M |
Guides: SD 1.5 VRAM requirements and SDXL VRAM requirements.
Batch Processing Benchmark
Tested on an NVIDIA RTX 3090 at default resolutions with max batch utilisation. See our benchmark page.
| Model (INT4) | Batch tok/s | Cost/M Tokens | GPU Utilisation | VRAM Used |
|---|---|---|---|---|
| SD 1.5 | 2.2 img/min | $0.0021/img | 92% | 4 GB |
| SDXL | 3.3 img/min | $0.0046/img | 95% | 7 GB |
SDXL generates more images per minute in batch mode (3.3 versus 2.2) despite its larger model size, likely due to better batch scheduling optimisations. However, each SDXL image costs 2.2x more. See our best GPU for LLM inference guide.
See also: SD 1.5 vs SDXL for API Serving (Throughput) for a related comparison.
See also: SDXL vs Flux.1 for Cost-Optimised Batch Processing for a related comparison.
Cost Analysis
At 100,000 images, SD 1.5 costs £141 versus SDXL’s £172. The saving is modest but consistent across volumes.
| Cost Factor | SD 1.5 | SDXL |
|---|---|---|
| GPU Required | RTX 3090 (24 GB) | RTX 3090 (24 GB) |
| VRAM Used | 4 GB | 7 GB |
| Images/min | 2.5 | 2.9 |
| Cost/1K Images | £1.41 | £1.72 |
See our cost calculator.
Recommendation
Choose SD 1.5 for synthetic data generation, texture libraries, dataset augmentation, and any batch pipeline where 512×512 resolution suffices and volume determines project cost.
Choose SDXL for batch generation of customer-facing assets: product catalogue images, marketing variants, and any output where 1024×1024 quality is the minimum acceptable standard.
Run overnight on dedicated GPU servers for maximum utilisation.
Deploy the Winner
Run SD 1.5 or SDXL on bare-metal GPU servers with full root access, no shared resources, and no token limits.
Browse GPU Servers
