Seeweb Simplifies GPU Usage with Clastix Capsule
Delivering a Serverless GPU-as-a-Service for Many Customers from a Shared Pool of Hardware, with Project Capsule and Virtual Kubelet.
Download Full Case Study PDF
REDUCE THE COST OF RUNNING KUBERNETES AT SCALE
- Company
- Seeweb
- Website
- seeweb.it/en
- Headquarters
- Frosinone, Italy
- Industry
- Cloud Infrastructure & Hosting Provider
- Employees
- ~50
- IT environment
- Serverless GPU service built on Capsule, integrated into Seeweb's Cloud ecosystem
- Clastix offering
- Project Capsule
- Use case
- Serverless GPU-as-a-Service · Multi-Tenant Kubernetes
AT A GLANCE
Challenge
Renting GPUs traditionally meant a slow, manual process of selecting, configuring, and provisioning remote machines before a model could even run.
Solution
A serverless GPU service that lets users access remote GPUs straight from their own Kubernetes cluster, with Capsule enforcing multi-tenant isolation.
Outcome
GPU workloads launched in under three minutes, with hourly pay-per-use billing and no dedicated hardware to buy or manage.
SUMMARY
Success story overview
Seeweb, a leading Italian cloud infrastructure provider, set out to offer a public, serverless GPU service that its customers could consume on demand, without buying or managing dedicated hardware. The goal was to serve many customers from a shared pool of hardware while giving each one strict isolation, predictable performance, and a simple, standard Kubernetes experience. Working with Clastix, Seeweb built Serverless GPU: by installing a lightweight agent, users create a virtual node in their local Kubernetes cluster that presents Seeweb's remote GPUs, immediately available for scheduling AI workloads. Project Capsule groups namespaces per customer and automatically enforces network policies, resource quotas, and fair-use constraints, while Virtual Kubelet pools GPU capacity elastically. The result: teams can launch a GPU workload in under three minutes, paying only for actual hourly usage, delivered from Italy- and Europe-based infrastructure with a 99.99% uptime SLA.
KEY REQUIREMENTS
- Serve many customers from a shared pool of GPU hardware
- Strict isolation and predictable performance for each customer
- Control-plane isolation and elastic GPU scheduling at scale •
- A simple, standard Kubernetes experience
- No changes to customers' existing pipelines
KEY OUTCOMES
- GPU workloads launched in under three minutes
- Hourly pay-per-use billing, no dedicated hardware to buy or manage
- Automatic per-tenant control-plane isolation and policy enforcement via Capsule
- Works with AKS, EKS, GKE, vanilla Kubernetes, OpenShift, Tanzu, and Rancher
- RedCarbon cut its model training time by 40% using Serverless GPU
Building a multi-tenant Kubernetes platform for your organisation?
We've built this architecture with organisations across the globe. We can help you build it too.
No commitment required. Start with a conversation.
MORE CUSTOMER STORIES
