Serverless GPUs for AI

Serverless GPUs for AI

More GPUs for Your AI Projects?

More GPUs for Your AI Projects?

On-demand access to remote GPUs on a massive scale, at lower prices in the market.

Early Access

Unlimited
Scale for AI

Overcome GPUs scarcity and access thousands of GPUs from your Kubernetes cluster.

Flexible Cost
Structure

Challenge the traditional GPUs renting model by accessing resources on demand wherever they are and at lowest prices.

Seamless Kubernetes
Integration

Access GPUs from your local or cloud-based Kubernetes cluster with all the tools you already use and love.

GPUs
Autoscaling

Fast spin-up times and responsive auto-scaling on demand without wasting time in renting wait queues.

Zero
capital

Avoid significant investments in on-prem
hardware or long-term commitments with Cloud GPUs providers.

How It Works

Install the Agent

Download and install the k8sGPU Agent into your Kubernetes clusters. This lightweight agent creates a virtual node within your cluster, acting as a seamless bridge to remote GPUs. Developed based on Microsoft's Virtual Kubelet project, it ensures compatibility with your existing Kubernetes tools.

Launch Your Workloads

Once a virtual node is in place, schedule your ML/AI pods on it as you would on regular worker nodes. The k8sGPU Agent handles the dynamic allocation of remote GPUs and runs your pods on them as if they were in your local cluster.

Scale as Needed

Adapting to your project's changing demands is hassle-free with our solution. Our responsive K8sGPU auto-scaling dynamically allocates and deallocates remote GPUs in real-time, ensuring both optimal performance and cost-efficiency.

Reduce Budget Clutter

Most rented GPUs stay idle for over 12 hours a day, on average. With k8sGPU, you only pay for the actual GPU usage time. This eliminates the need for long-term rental commitments and reduces the entire GPUs rental costs.

Architecture

Cloud Native & Kubernetes ready Solution

FAQ

This page provides answers to frequently asked questions and addresses common issues related to our service.

How do I access the service?

You can sign up for the service here

Please contact our team here to tailor the service to your specific needs.

Yes, the agent is open source and free of charge.

Install the k8sGPU agent on your Kubernetes cluster using Helm. For detailed instructions, please refer to our documentation.

Join our Slack channel for support or report bugs by opening an issue on our GitHub account.

We support most public Managed Kubernetes Services, such as AKS, EKS, and GKE, as well as on-premise environments like Kubernetes vanilla and common distributions such as OpenShift, Tanzu, and Rancher. Local environments like kind are supported; however, K3s support is in development.

Schedule your pods on the virtual node by setting specific affinity rules and tolerations for virtual nodes. For more information, see our documentation.

You can run any container image that is available from any accessible container registry. If you're new to containers, we recommend learning about container technology first.

Yes, you can run multiple pods simultaneously on the same GPU, but be aware that GPU memory and processing time are shared among all your workloads.

GPUs are dynamically reserved on demand when pods are assigned to virtual nodes and released upon their termination. During periods of high demand, some GPU types may temporarily be unavailable. If this occurs, try again later or contact us if you continue to experience issues.

We offer multiple GPU types. To request a specific GPU, such as the Nvidia A100, refer to the corresponding Kubernetes RuntimeClass resource. For further details, see our documentation.

No, each GPU is exclusively assigned to a user and not shared.

Currently, pods on the virtual node are not interruptible as each gets a dedicated GPU. We are developing an interruptible service option that will be more cost-effective.

MIG support is currently in development and will be available soon.

Yes, your pods are isolated in a multi-tenant environment with strict network policies and are only exposed to the Internet as required.

Pods on the virtual node run on remote infrastructure and are not accessible within your local cluster. If your pod exposes an API, you can publish it on the Internet by setting a specific annotation on the pod. Note that you are responsible for securing access to your API.

Currently, pod APIs are only exposed to the public Internet as needed. We plan to offer private access via VPN in future updates.

Your workloads will continue to run remotely. Once connectivity is restored, your pods will synchronize with their local counterparts. Billing is based solely on actual GPU usage.

Pods can store and retrieve data from any accessible S3 bucket globally. You can use our optimized S3 endpoint or your own object storage solution.

Yes, pods on the virtual node can mount Secrets and ConfigMaps, but local and persistent volumes are not supported.

Yes, our service complies with the General Data Protection Regulation (GDPR) for data processed within European data centers.

The provisioning of GPUs typically takes a few minutes, depending on the current load and availability. We strive to minimize wait times.

At the first launch of the service, you'll receive regular reports detailing your GPU usage. Later, we will introduce a dashboard for real-time monitoring and access to historical data

We offer multiple support tiers, including 24/7 support for critical issues. More details are available on our support page.

Discover the high-efficiency distributed
GPUs network for AI/ML

Early Access

About Seeweb

Seeweb is a leader in cloud computing infrastructure in Italy, offering a wide range of services, from web hosting to cloud solutions. Committed to reliability and performance, Seeweb enables businesses to leverage the power of the cloud for growth and innovation.

About Clastix

Clastix is a pioneering technology firm specializing in Kubernetes solutions that drive efficiency and innovation in cloud computing. Focusing on simplifying complex environments, Clastix empowers businesses to harness Kubernetes' full potential, ensuring scalable, secure, and efficient operations.

About Seeweb

Seeweb is a leader in cloud computing infrastructure in Italy, offering a wide range of services, from web hosting to cloud solutions. Committed to reliability and performance, Seeweb enables businesses to leverage the power of the cloud for growth and innovation.

About Clastix

Clastix is a pioneering technology firm specializing in Kubernetes solutions that drive efficiency and innovation in cloud computing. Focusing on simplifying complex environments, Clastix empowers businesses to harness Kubernetes' full potential, ensuring scalable, secure, and efficient operations.

Become an Early Adopter

Email: aicloud@seeweb.ai

Seeweb © srl 1998-2022 – c.s. 103.000 EURO - CCIAA 17793/98 - REA 126233 - p.iva e c/f 02043220603

Privacy & Cookie | Condizioni generali