Quick Answer
A local AI and Stable Diffusion rig in Klerksdorp is decided by GPU VRAM. The RTX 4060 Ti 16 GB (~R10,500) runs SDXL at roughly 2 iterations/sec and Flux.1 without out-of-memory errors; the RTX 4070 Ti Super 16 GB is faster still. Pair the GPU with a Ryzen 5 7600 and 32 GB DDR5. VRAM, not CPU speed, sets what models you can run.
VRAM tiers for Stable Diffusion
SD loads the whole model into GPU memory, so VRAM caps the workload:
- 8 GB: SD 1.5 only, SDXL struggles
- 12 GB: SDXL comfortable, Flux tight
- 16 GB (RTX 4060 Ti 16 GB): SDXL + Flux + ComfyUI batches - the value pick
- 24 GB (RTX 4090): fastest single-card option
On the RTX 4060 Ti 16 GB, SDXL runs about 2 it/s at 1024px (a 25-step image in roughly 12-15 seconds).
The supporting build and North West delivery
- CPU: Ryzen 5 7600 (~R4,200) - feeds the GPU
- RAM: 32 GB DDR5-6000 (~R2,000)
- Storage: 1 TB NVMe (~R1,200) - checkpoints are 2-7 GB each
- PSU: 650-750 W 80+ Gold for stable renders
Klerksdorp is in North West, so Evetech deliveries take 2-3 business days via the inland route; power-test before signing.
FAQ
What iterations/sec can I expect on a 16 GB card?
The RTX 4060 Ti 16 GB runs SDXL at roughly 2 it/s at 1024px, so a typical 25-step image completes in about 12-15 seconds. SD 1.5 is much faster, and enabling xFormers speeds both up.
Which models can a 16 GB card run?
SDXL, Flux.1, SD 1.5 and most ComfyUI workflows with reasonable batch sizes. 16 GB is the comfortable tier for current image models; only video models and very large batches push toward 24 GB.
Does Evetech deliver to Klerksdorp?
Yes, Evetech ships to Klerksdorp in North West in 2-3 business days. Power-test the GPU and confirm it's detected before signing.
and LoRAs on the fast NVMe and use ComfyUI's model caching - it cuts the load time when switching between SDXL, Flux and SD 1.5 models during a session.