NVIDIA A2
STO2
GPU memory
16 GB GDDR6
- Architecture
- Ampere
- Configurations
- 4 (b2 and l2)
- Price from
- €0.34/hour
- Power limit
- 60 W
- Status
- Enabled through support
- Use cases
- Inference, light training, video encoding
GPU instances in Safespring Compute
Safespring Compute provides GPU flavors for inference, model training, video encoding, and running large language models.
GPU resources are provided as flavors, which are predefined resource profiles. Each GPU flavor has a fixed combination of GPU, vCPU, and RAM. Flavors in the l2 series also include local storage. Current configurations and prices are listed on the price page.
A2 is suitable for inference, lighter training, and video encoding. A2 flavors are available in STO2 and are enabled for the project through support. H100 NVL is available in STO1 on request and is optimized for inference with large language models. The card has 94 GB of HBM3 memory, PCIe 5.0 x16, and passive cooling.
NVIDIA H100 NVL
STO1
GPU memory
94 GB HBM3
- Architecture
- Hopper
- Configurations
- 1 (b2)
- Price from
- €2.93/hour
- Power limit
- Up to 400 W
- Status
- Available on request
- Use cases
- Inference with large language models
Choose between local and central storage
l2 flavors include local NVMe storage on the compute node. The root disk follows the instance lifecycle and is deleted with the instance. Safespring cannot restore the data if the local disk fails. Use l2 for stateless or short-lived workloads, or make sure that the data is backed up.
b2 flavors do not include a local disk and boot from a persistent volume in central block storage. The volume remains independently of the instance and can be either fast or large.
GPU nodes for Kubernetes
GPU support in Safespring Kubernetes Engine is available in STO2 for worker nodes using A2 flavors whose names end in gA2. The documented SKE support does not include H100 NVL. The managed control plane and the Compute instances used as worker nodes are billed separately.
GPU and machine learning documentation and examples
Explore technical documentation for GPU instances, a practical guide to running a local language model, and an example of machine learning on Safespring’s infrastructure.
Configurations and prices
The GPU instance price list contains current configurations with vCPU, RAM, local storage, GPU model, and prices per hour and per 30 days.
If you are unsure which flavor or storage option fits your workload, contact Safespring to discuss your requirements.
