Salad Cloud

What's good about Salad Cloud

• Aggregating idle consumer hardware gives some of the lowest published per-GPU-hour rates available
• Per-second billing with no contracts or minimums; allocation, image download, and cold start are not billed
• vCPU and RAM are included with every GPU class at no additional charge
• Four priority tiers make interruptibility an explicit, priced choice per container group
• Live per-GPU-class availability counts are published alongside the rate card
• Managed container orchestration handles node loss and reallocation automatically

Salad Cloud GPU prices

37 GPU models across 148 configurations. Rank compares Salad Cloud's cheapest on-demand listing for each GPU with every provider's cheapest.

What a run costs on Salad Cloud

24 hours at today's cheapest listing for each GPU, at the size it is listed.

• 1× RTX 4090 for 24 hours

  Lowest Priority at $0.160/GPU-hr

  $3.84

  $6.24 under the market median rate

• 1× RTX 5090 for 24 hours

  Lowest Priority at $0.250/GPU-hr

  $6.00

  $8.16 under the market median rate

• 1× RTX 3090 for 24 hours

  Lowest Priority at $0.090/GPU-hr

  $2.16

  $3.12 under the market median rate

The market median is every provider's cheapest on-demand price for that GPU.

Billing & terms

Per-second on-demand billing

Published per-GPU-hour rates billed by the second for the time each replica spends running; allocation, image download, and cold start are not charged, and vCPU and RAM are included.

Priority tiers

High, Medium, Low, and Lowest tiers price each GPU class differently - lower tiers cost less and are more interruptible, with Lowest intended for retryable batch work.

CPU-only container groups

Billed per vCPU-hour plus per GB of RAM per hour for container groups that need no GPU.

Volume discounts

Large deployments, long-running jobs, and committed spend are priced below list by sales.

Salad Dedicated contracts

Reserved bare metal and virtual machines quoted separately per allocation and term.

Private Cloud licensing

Licensed per deployment to run the orchestration stack on hardware you already own.

Regions

Nodes in 190+ countries across the distributed network; capacity is chosen by GPU class and priority tier rather than by selectable region, and availability varies by class.

Support

Documentation at docs.salad.com, a developer hub and Recipes library, status page at cloud-status.salad.com, Trust Center, Discord and GitHub communities, and direct engineering and sales contact for volume and enterprise deployments.

Watch out for

• On-demand catalog is consumer RTX GPUs only — data center accelerators are not offered outside the contracted Dedicated product
• Published GPU classes top out at 32 GB of VRAM, limiting the size of a model that fits on a single GPU
• Nodes are individually owned machines that can disconnect at any time, so workloads must be stateless or retryable
• Medium, Low, and Lowest tiers can be preempted by higher-priority container groups
• Reserved capacity, volume pricing, and Private Cloud licensing are quoted by sales rather than published

Features

Community Cloud

On-demand pool of idle consumer GPUs with nodes in 190+ countries, deployed onto through Salad Container Engine.

Salad Container Engine

Managed orchestration that schedules Docker containers onto nodes it does not control, isolates them behind a hypervisor, streams logs, and reallocates work when a node disappears.

Priority Tiers

High, Medium, Low, and Lowest tiers price the same GPU class differently so a workload can trade preemption risk for cost; High is never preempted by other workloads.

Salad Dedicated

Reserved bare metal and virtual machines on contract, starting with RTX PRO 6000 Blackwell, quoted per allocation and term.

Private Cloud

Salad's orchestration stack licensed per deployment on hardware you already own, pooling idle workstation and server GPUs into a private OpenAI-compatible token service.

Managed APIs and Tooling

AI Gateway (an OpenAI-compatible endpoint for open-weight models, in closed beta), Transcription API, Recipes deployment templates, S4 temporary storage, Gateway Service residential IPs, and a Virtual Kubelet for Kubernetes.

Services

• Salad Container Engine on Community Cloud

  On-demand container orchestration across the distributed pool of idle consumer GPUs, with more than 40 GPU classes selectable by priority tier.

• Salad Dedicated

  Reserved bare metal and virtual machines on contract, separate from the on-demand community pool.

• CPU-only container groups

  Container groups with no GPU attached, billed per vCPU-hour and per GB of RAM per hour, running at the Lowest priority tier.

• Private Cloud

  The same orchestration stack licensed to run inside your organization, pooling idle GPUs in workstations and servers you already own so company data never leaves company devices.

Getting started

1. Step 1

   Create a SaladCloud account

   Sign up at portal.salad.com to reach the dashboard, API keys, and deployment tools.

2. Step 2

   Pick a GPU class and priority tier

   Compare published per-GPU-hour rates and live availability on the pricing page, then choose the tier that matches how interruptible the workload is.

3. Step 3

   Deploy a container group

   Supply a Docker image, CPU/memory/GPU resources, and a replica count through the portal or the public API; Recipes provide ready-made templates for popular open models.

4. Step 4

   Monitor and scale

   Track replicas in the portal and adjust GPU class, priority, or replica count; the container engine reallocates work as nodes join and leave the network.

5. Step 5

   Talk to sales for reserved or volume capacity

   Large deployments, committed spend, and Salad Dedicated allocations are quoted directly rather than bought self-service.