What is Salad?
Distributed GPUs from $0.02 an hour for inference.
Who is Salad best for?
Salad is best for developers running inference jobs that fit on one consumer GPU and can cope with a node dropping out.
Image generation at scale
Many consumer RTX cards can run the same model in parallel; Civitai serves 10 million images a day on more than 600 Salad GPUs.
Speech-to-text and voice
Salad's own Transcription API runs on the same network, starting at $0.16 per audio hour.
Retryable batch jobs
The Lowest tier is the cheapest and can be interrupted, which suits queued work that simply restarts, such as image tagging or molecular dynamics runs.
Cutting a cloud GPU bill
Idle home and small-business PCs cost less per hour than data-center cards, as long as the job needs neither multi-GPU NVLink training nor sub-10 ms latency.
How much does Salad cost?
Salad is pay as you go with no free trial: older GPUs start at $0.015 an hour, an RTX 4090 costs $0.16 to $0.33 an hour depending on priority, and CPU time is $0.005 per vCPU-hour, all billed by the second.
- Plus $0.001 per GB of RAM per hour; 1 vCPU with 1 GB RAM for $0.006 an hour
- 8 vCPU with 16 GB RAM for $0.056 an hour
- Runs at the Lowest priority tier
- $0.140 an hour at High priority
- About $43.80 a month running nonstop at Lowest
- vCPU and RAM included
- $0.330 an hour at High priority
- About $116.80 a month running nonstop at Lowest
- vCPU and RAM included
- $0.500 an hour at High priority
- About $182.50 a month running nonstop at Lowest
- vCPU and RAM included
Every GPU class has four priority tiers: High is never preempted by other workloads, Medium and Low can be, and Lowest is the cheapest and the easiest to interrupt. More GPU classes, such as the RTX 3060 and RTX 3070, are available in the portal from $0.015 per GPU-hour at Lowest priority. Billing runs per second while an instance is running; allocation, image download and cold starts are free. Deployments above 100 GPUs, long-running jobs and committed spend get volume pricing from sales, and reserved RTX PRO 6000 Blackwell servers (Salad Dedicated) are quoted per contract. Full pricing on salad.com. We checked these prices on .
How do readers rate Salad?
No one has rated Salad yet. Be the first.
Sign in to rate. One rating per account.
What are the best Salad alternatives?
Two tools on Recaply do Salad's job.
Questions about Salad
What is Salad?
Distributed GPUs from $0.02 an hour for inference.
Who is Salad for?
Salad is best for developers running inference jobs that fit on one consumer GPU and can cope with a node dropping out.
Is Salad free?
No. Salad has no free plan or free trial. SaladCloud is pay as you go with no minimum commitment, billed by the second only while your containers run, so a small test costs little. GPUs start at $0.015 an hour for older cards.
How much does Salad cost?
Salad's paid plans are CPU instances at $0.005 per vCPU-hour, RTX 3080 (10 GB) at $0.060 per GPU per hour at Lowest priority, RTX 4090 (24 GB) at $0.160 per GPU per hour at Lowest priority and RTX 5090 (32 GB) at $0.250 per GPU per hour at Lowest priority.
What are the best Salad alternatives?
The closest alternatives on Recaply are CLORE.AI and Float16.cloud.
Where can I try Salad?
On its official website, salad.com.
Can Salad GPUs be interrupted mid-job?
Yes. Salad's GPUs sit in other people's PCs and can go offline. High priority is never preempted by other workloads, while lower tiers can be, and Salad's container engine moves the work to another node of the same GPU class and bills only running time.
Modal
Modular
Tensorfuse
Apache SkyWalking
TiDB
FuriosaAI