Evaluating decentralized compute providers requires analyzing trade-offs between low hourly rates and variable host environments. In this detailed salad cloud review, we analyze how Salad Cloud leverages idle consumer gaming hardware via the Salad Container Engine to offer GPU compute at significantly lower prices than traditional datacenter platforms[source].
Key takeaways
- Unbeatable batch pricing: Container instances start at $0.204 per hour for RTX 4090 cards and $0.294 per hour for RTX 5090 GPUs at batch priority levels[source].
- No charge for cold boot: Billing commences only after your container finishes pulling images and transitions to a running state on host nodes[source].
- Decentralized node pool: Salad claims over 2 million registered nodes and more than 60,000 active GPUs supplied by individual PC owners worldwide[source].
- Stateless execution model: Containers run with ephemeral storage and no direct SSH access, requiring external object storage for state persistence[source].
- Compliance limitations: Lack of SOC 2, HIPAA, or PCI-DSS certifications restricts enterprise deployment for sensitive regulated data[source].
Salad Cloud review: Quick verdict
Salad Cloud delivers exceptional cost savings for fault-tolerant, stateless GPU workloads like asynchronous batch inference, image generation, and video processing. By utilizing idle consumer gaming PCs running isolated OCI containers, Salad achieves hourly rates up to 70% lower than traditional cloud platforms[source]. However, variable host internet bandwidth, ephemeral container storage, and the absence of enterprise compliance frameworks mean it is not designed for low-latency live APIs or sensitive financial data[source].
How Salad Cloud works and service model
Salad Cloud operates on a peer-to-peer compute model that fundamentally differs from hyperscale datacenters or managed bare-metal providers[source].
Salad Cloud organizes customer container workloads across three operational tiers[source]:
- Salad Portal & API: The entry point where developers define container groups, specify environment variables, configure hardware requirements (VRAM, vCPUs, RAM), and select target priority levels[source].
- Salad Container Engine (SCE): The orchestration layer that monitors available compute capacity across the peer-to-peer network, evaluates node stability using Trust Ratings, and deploys container images to verified hosts[source].
- Distributed Consumer Nodes: Individual host machines owned by retail PC users worldwide. Each node runs containers inside isolated Linux virtual machines via Hyper-V, executing workloads on GPUs such as the RTX 3090, RTX 4090, or RTX 5090[source][source].
The decentralized node network
Rather than owning data center racks, Salad aggregates compute capacity from retail PC owners who share their idle hardware in exchange for rewards[source]. Salad reports a global pool exceeding 2 million total registered nodes and over 60,000 available GPUs[source].
Workloads are scheduled through the Salad Container Engine (SCE), which packages Docker containers and deploys them to compatible host nodes across the network[source].
Container execution environment
Salad does not provide full virtual machines, root access, or direct SSH connections to host nodes[source]. Compute instances run exclusively as standard OCI-compliant container groups[source].
Because nodes run on Windows host systems using WSL2 and Hyper-V isolation, a small portion of GPU memory may be reserved by the host OS, resulting in slightly less than 100% available VRAM[source]. Container filesystems are strictly ephemeral; any data written locally is permanently destroyed when a container stops or gets reassigned to a different node[source].
Salad Cloud pricing and node priority tiers
Salad Cloud uses a dynamic, priority-based pricing structure where cost corresponds directly to workload allocation priority[source].
| GPU Model | VRAM | Batch Hourly Rate | Standard Hourly Rate | Priority Hourly Rate |
|---|---|---|---|---|
| NVIDIA RTX 3080 | 10GB | $0.080[source] | $0.110[source] | $0.140[source] |
| NVIDIA RTX 3090 | 24GB | $0.150[source] | $0.220[source] | $0.310[source] |
| NVIDIA RTX 4090 | 24GB | $0.204[source] | $0.340[source] | $0.480[source] |
| NVIDIA RTX 5090 | 32GB | $0.294[source] | $0.450[source] | $0.620[source] |
Priority level mechanics
- Batch priority: The lowest hourly rate[source]. Containers can be interrupted or preempted when higher-priority workloads request capacity. Ideal for stateless queues.
- Standard priority: Provides balanced node reservation for general processing tasks[source].
- Priority tier: Gives containers maximum retention priority on active nodes[source].
Zero-cost cold boot policy
A major financial advantage of Salad Cloud is its billing model for container setup. Traditional cloud providers charge per minute from the moment an instance begins booting, including slow container image pulls. Salad Cloud charges $0.00 during the initial image download and initialization phase[source]. Metered billing starts only when your container passes health checks and reaches a running status[source].
Bandwidth and egress pricing
Salad Cloud does not assess separate network egress fees for standard outbound data transfers[source]. This makes it highly economical for data-intensive workloads such as video rendering or bulk image output.
Important features and platform security
Deploying containerized applications on consumer hardware introduces unique security and reliability requirements.
Container security and host isolation
Salad enforces bidirectional isolation to safeguard both customer workloads and host PCs[source]:
- Hyper-V Virtualization: Containers execute inside isolated Linux virtual machines managed by Hyper-V[source]. Host users cannot view memory contents or inspect container storage.
- Encrypted credentials: Registry authentication tokens and environment secrets are encrypted in transit and at rest[source]. Private registry keys are discarded immediately after pulling container layers[source].
- No inbound network access by default: Containers do not expose open ports to the host local area network[source]. Outbound network connections route through secure proxy gateways[source].
Trust Rating system
To manage node quality, Salad uses an automated Trust Rating algorithm that continuously measures host uptime, hardware stability, and network bandwidth[source]. Nodes with high trust scores are prioritized for standard and priority container allocations, while unproven or unstable nodes are assigned low-priority batch workloads[source].
Regulatory compliance limitations
Salad explicitly documents that its network is not certified for enterprise compliance standards such as SOC 2 Type II, HIPAA, or PCI-DSS[source]. Organizations handling protected health information, financial transactions, or classified data must not deploy those workloads on Salad Cloud[source].
Who should use Salad Cloud
Salad Cloud is an optimal compute platform for specific architecture patterns:
- Asynchronous batch inference: Processing queue-backed jobs where individual node restarts do not impact overall system completion[source].
- AI image and video generation: Running Stable Diffusion or ComfyUI pipelines that write output directly to AWS S3 or Google Cloud Storage[source].
- Cost-sensitive AI startups: Teams looking for the best cheap GPU cloud to maximize compute hours per dollar[source].
- 3D rendering and transcoding: Rendering Blender scenes or encoding video files where free egress reduces operational overhead[source].
Who should skip Salad Cloud
You should choose alternative provider architectures if:
- You require low-latency, real-time HTTP API endpoints with guaranteed sub-100ms cold start times[source].
- Your workload requires root VM access, custom kernel modules, or persistent block storage attached to instances[source].
- You operate under mandatory SOC 2, HIPAA, or ISO 27001 regulatory compliance requirements[source].
- You require high-speed multi-GPU interconnects like NVLink for large-scale distributed model training[source].
Alternatives to Salad Cloud
Comparing Salad against alternative cloud hosts helps determine whether decentralized compute fits your requirements.
Salad Cloud vs RunPod
RunPod offers dedicated data center instances with persistent disk storage, template pods, and root terminal access[source]. While RunPod's on-demand rates ($0.69 to $0.99 per hour for consumer GPUs) exceed Salad's batch rates, RunPod provides predictable hardware performance and persistent storage[source]. Read our detailed RunPod review for complete technical specifications.
Salad Cloud vs Vast.ai
Vast.ai operates a marketplace connecting renters to individual host servers[source]. Unlike Salad's container-only abstraction, Vast.ai grants direct SSH access, configurable persistent storage, and full VM management[source]. However, pricing on Vast.ai varies by individual host operator. Review our Vast.ai review to compare marketplace mechanics.
Salad Cloud vs TensorDock
TensorDock provides low-cost virtual machines across independent regional datacenters. TensorDock includes root SSH access and persistent storage, catering to users who need full server control rather than containerized batch processing. See our TensorDock review for platform details.
Pros and cons
Pros
- **Exceptional value**: Batch pricing starting at $0.204/hr for RTX 4090 and $0.294/hr for RTX 5090[source].
- **Zero-cost cold boots**: No billing while pulling container images or running startup scripts[source].
- **No egress fees**: Outbound data transfers are included without extra bandwidth charges[source].
- **Vast capacity**: Access to over 60,000 consumer GPUs globally[source].
Cons
- **Stateless storage**: Ephemeral container filesystems require external object storage[source].
- **Variable cold starts**: Image pull times depend on consumer host internet bandwidth[source].
- **No SSH or root access**: Execution restricted to managed container groups[source].
- **No enterprise compliance**: Lacks SOC 2, HIPAA, and PCI-DSS security certifications[source].
Research methodology and sources
GPU Picks evaluates cloud compute platforms using official documentation, published pricing tables, and architectural specifications[source]. We do not conduct hands-on performance benchmarks or publish subjective testing metrics. For details on our evidence principles, visit our methodology page or browse GPU models in our GPU lookup tool.
Is Salad Cloud cheaper than traditional cloud providers?
Yes, Salad Cloud's batch pricing model offers hourly rates up to 70% lower than traditional cloud providers by aggregating idle consumer gaming hardware[source].
Does Salad Cloud charge for container startup time?
Can I attach persistent disk storage to Salad containers?
How does Salad Cloud secure containers on consumer PCs?
Does Salad Cloud support root SSH access?
No, Salad Container Engine runs managed container groups without direct SSH access or root virtual machine terminals[source].
What GPUs are available on Salad Cloud?
Salad offers consumer GPUs including the NVIDIA RTX 3080, RTX 3090, RTX 4090, and RTX 5090, depending on active host node availability[source].