Use case · computer vision

Computer vision at scale, from $0.015 per GPU-hour.

Tagging, detection, segmentation, and classification models are small and embarrassingly parallel. On SaladCloud they run on thousands of consumer GPUs at the edge for far less than managed vision APIs or high-end cloud GPUs.

Deploy YOLOv8 Talk to sales

309,000images tagged per dollar, RAM++ on RTX 2080

50,000images segmented per dollar, SAM

81%less than Azure for YOLOv8 object detection

$0.15to train a custom YOLOv8 model

Image tagging

Older GPUs, better economics.

For tagging at scale, the network's older cards deliver far more images per dollar than new high-end GPUs. The RTX 2080 tags 309,000 images per dollar with Recognize Anything Model++, a 60–300x cost reduction versus managed services like Azure. The GTX 1650 processes more images per dollar for ConvNextV2 Tagger V2 than any GPU in the market.

Object detection

The sweet spot between cheap CPUs and fast GPUs.

GPUs are far faster than CPUs for YOLOv8, and on SaladCloud they start at $0.015 per hour. Processing a live stream costs about $0.036 per hour on Salad against $0.19 per hour on Azure general compute, 81% less.

Models