The largest pool of idle GPUs on Earth.
Community Cloud is Salad's on-demand infrastructure pool: consumer GPUs shared by people worldwide. Salad Container Engine is the orchestration layer that deploys and keeps your containers running across that pool.
Built for high-volume, stateless, interruption-tolerant work.
Inference at scale, batch pipelines, simulation sweeps, rendering, data processing. If a job can be retried and runs on 8–32 GB of VRAM, Community Cloud is usually the most cost-efficient place on the internet to run it.
Priority pricing
Four tiers: High, Medium, Low, Lowest. Higher tiers are protected from preemption by other workloads; lower tiers cost less. Pick per container group.
Elastic to 10,000+ replicas
Scale up with one API call. No provisioning delays, no reserved quotas, no negotiation.
Pay for running time
Per-second billing applies only while instances are running. Allocation, cold start, and image download are not billed.
Vetted hardware
Every GPU is benchmarked before admission and continuously trust-rated. Only nodes with a solid rating receive workloads.
Geo-distributed
Nodes in 190+ countries. Constrain workloads by region when compliance or latency requires it.
Same engine as everywhere else
Move a container group from Community Cloud to Salad Dedicated or Private Cloud without changing the image or the API.
RTX-class GPUs, RTX 3060 through RTX 5090.
Consumer GPUs outperform data-center parts on price-performance for most inference. The RTX 5090's 32 GB of VRAM runs the majority of open models in production quantizations.
RTX 5090
32 GB · Blackwell · from $0.250/hr
RTX 4090
24 GB · Ada · from $0.160/hr
RTX 5080
16 GB · Blackwell · from $0.150/hr
RTX 3090
24 GB · Ampere · from $0.090/hr
450,000+ GPUs, one network.
Salad Chefs run our desktop app when their PC is idle and earn rewards for the compute they contribute. It is the supply side of a nine-year-old two-sided marketplace, and it is why Community Cloud's prices are what they are.
Ethically sourced
Every node is voluntarily shared. Owners can see when Salad is active, pause anytime, and are compensated for their contribution.
Continuously growing
New GPU generations arrive on the network within weeks of retail launch, years before they reach most clouds.
Sustainable by construction
Using hardware that already exists avoids the build-out, cooling, and grid demand of new data centers.
"If you want access to 1000s of GPUs, you can get them on SaladCloud for better cost-efficiency. Salad is also really customer friendly, something you cannot get from a larger cloud provider as a startup."
Zachary Lawrence · CEO, Klyne.aiRead the case study →Start on the lowest-cost GPUs in the market.
Volume discounts above 100 GPUs. Committed contracts go further.