What Does Security Look Like on SaladCloud Secure?
SaladCloud Secure nodes are hosted in professionally managed datacenters operated by partners with strong security postures and third-party compliance audits. These facilities are designed for high-trust workloads and meet the expectations of enterprise AI teams. Key security attributes:- SOC 2 Type 2 attested and ISO27001 certified datacenter partners
- Restricted physical access, including 24/7 surveillance and badge-based entry controls
- Hardware-level isolation — no shared cores or memory across tenants
- Local storage is isolated per job and wiped after use
- All data in transit is encrypted
- Salad workloads run in isolated, containerized environments with no cross-job memory access
Datacenter-Grade Infrastructure
SaladCloud Secure deployments are built on modern infrastructure that matches best practices from leading enterprise GPU clouds.Network & Performance
SaladCloud Secure nodes are provisioned with a minimum of 1 Gbps bandwidth per server, ensuring fast data transfers and low-latency performance for AI/ML workloads.- External bandwidth: ≥ 1 Gbps per node (up/down)
- Redundant internet uplinks for automatic failover
- DDoS mitigation systems in place
- Internal switching designed for multi-100 Gbps aggregate throughput
- Datacenter targets:
- Packet loss < 0.1%
- P95 RTT < 4ms
- P95 jitter < 3ms
Storage Expectations
SaladCloud Secure servers are equipped with high-speed local storage designed for performance during active compute jobs.- Each node provides fast, isolated local storage for use during job execution — ideal for temporary datasets, model layers, and intermediate outputs
- Local storage is not persistent across interruptions; data is cleared if a node is reclaimed or rebooted
- For workloads requiring persistence, you can optionally integrate external object storage (e.g., S3-compatible) or configure periodic syncing from your container
Note: SaladCloud Secure does not currently provide persistent volumes across job interruptions. If your workload requires long-term data retention, we recommend syncing with remote storage or using checkpointing.
Performance, Uptime & Cost
- GPU Types Available:
- 8x NVIDIA A100 (80 GB) Perfect for large-scale training and production inference
- 8x NVIDIA A100 (40 GB) Great for models fine-tuning, and production inference workloads
- 8x NVIDIA L40S (48 GB) Ready for generative AI, rendering, and production inference
- Provisioned in full 8-GPU sets for maximum performance and bandwidth across multi-GPU jobs
- Uptime: Machines remain interruptible but are hosted in stable datacenters with redundant power and secured environments — ensuring significantly higher uptime than residential nodes.
- Cost: Despite datacenter-grade hardware and security, SaladCloud Secure is priced significantly lower than other enterprise GPU offerings. Datacenters often have idle capacity between large reservations — SaladCloud takes advantage of these gaps to access premium GPUs at the lowest possible cost and pass those savings directly to our customers.
How Do I Access or Monitor GPU Availability?
You’ll soon be able to check and reserve SaladCloud Secure inventory directly from the Salad Portal or via API. For early access or bulk reservations, contact our team.SaladCloud Community vs. Secure
| Feature | SaladCloud Community | SaladCloud Secure |
|---|---|---|
| Environment | Idle gaming PCs (residential) | Idle GPUs in secure datacenters |
| Security | SOC 2 | SOC 2 Type 2, ISO27001, 24/7 datacenter security |
| GPU Types | Consumer-grade | A100 (80GB), A100 (40GB), L40S (48GB) |
| Resource Unit | Single GPU | Full sets of 8 GPUs per Instance |
| Uptime | Best-effort | Higher stability, still interruptible |
| Pricing | Lowest on the market | Still lowest on the market |