Vulcan
Turn your on-premises GPU servers into your company's own GPU cloud
An in-house GPUaaS platform that transforms your bare-metal GPU servers into a multi-tenant self-service cloud with a single Helm command. No public cloud required — it layers directly on top of your existing DGX/HGX. Accessible to everyone · full visibility into who used what · your data stays on premises.
Introduction Video
Does Any of This Sound Familiar?
GPU Requests That Take Days
Getting a GPU means another Slack message or email — then waiting for days
Utilization Below 30%
In-house GPU utilization doesn't even reach 30%
Half a Day per Setup
DevOps spends half a day setting up each new user's environment
Black-Box Operations
Nobody knows who is using which GPUs, or how much
Painful Cloud Bills
The monthly cloud GPU invoice keeps growing
Data Cannot Leave
Your business cannot upload models or data to an external cloud
Zombie Resources
Resources from finished PoCs are never reclaimed and just sit there
K8s Skills Shortage
You lack the specialists to run Kubernetes yourselves
If you have GPUs in house yet these symptoms keep recurring, the problem is your operating structure, not the hardware. If three or more of the eight apply to you, it is time to seriously consider adopting Vulcan.
One Platform. Three Promises.
Central Control
A single admin governs GPU nodes, users, quotas, and billing from one console — scattered resources become one management screen
Self-Service
Users provision their own workspaces — no admin approval needed · instant 3-way access via VSCode, Terminal, and SSH
Multi-Tenant Isolation
Complete isolation between users and projects via namespace + RBAC + NetworkPolicy — data and traffic mutually blocked
Same GPU Servers, Completely Different Operations
| Before · Without Vulcan | After · With Vulcan |
|---|---|
| Slack request for a GPU → days of waiting | Instant workspace provisioning with a console click |
| Half a day of environment setup per user | SSH, firewall, and drivers all fully automated |
| No usage tracking — black-box operations | Usage history automatically recorded per user and project |
| No process for reclaiming abandoned workspaces | Stop · Restart · Delete in one action from the console |
| Extra spend on external cloud GPUs in the end | Recovered in-house GPU utilization → lower cloud costs |
Key Outcomes
Workspace provisioning lead time
DevOps environment setup
Usage history visibility
Cloud GPU costs
Not Just GPU Rental — A GPU Platform
User Self-Service — Zero Admin Approvals
Pick GPU count, storage, and image in the console, then one click. Namespaces, containers, PVCs, quotas, and policies are created automatically
Operations Automation — 2-Click Node Registration
Two clicks — Probe → Provision — auto-configure drivers, k3s, and VPN on a GPU node. No DevOps intervention
Visibility & Chargeback — 1-Second Metrics
DCGM-based usage history aggregated automatically. Internal chargeback across departments and affiliates, quota and budget controls, and audit logs
100% On-Premises · Air-Gapped — Zero External Calls
Every component runs on your internal network. The same UX even in closed networks — meets finance, healthcare, public-sector, and defense requirements
Full ML Lifecycle — From Provisioning to Serving and Billing
Workspace
Self-provision with your choice of GPU count and image (VSCode/SSH) — start developing instantly in a browser terminal or VSCode, no installation
Models
Store training runs and weights — manage models trained in your workspace as-is
Endpoints
Serve models as vLLM API endpoints — with API Key issuance
Monitor
Automatic usage and billing aggregation — history tracked per user and project
Operations, All in One Console (Product Tour)
Infrastructure Control
Oversee GPUs, VPN, and workspaces from the Admin dashboard · automatic node registration (Probe → Provision) · real-time 8-GPU utilization, temperature, and power monitoring
Governance
Wizards from Tenant → Project → user — assign GPU quotas and roles in a few clicks, with automatic SSO account (Keycloak) issuance
Self-Service · Development · Serving
Self-provision a workspace → develop in a browser terminal or VSCode → serve that same model as a vLLM endpoint
Advanced Operations
DGX node details, VPN and SSH gateway (sshpiper, port forwarding), and workspace status tracking — all from the console with no separate SSH sessions
Four People Who Need Vulcan
AI Researchers · ML Engineers
"I need to run training today, but I have to borrow a GPU over Slack again" → provision your own workspace with one click · keep using the tools you know, VSCode and SSH
DevOps · Infrastructure Owners
"Setting up each new researcher's environment takes half a day" → SSH keys, firewall, and drivers fully automated · operate without a K8s expert
CTOs · R&D Executives
"GPU utilization is at 30% but the cloud bill grows every month" → visibility into in-house GPU utilization · phase down cloud dependency
CISOs · Security & Compliance
"We can't upload models or data to an external cloud" → 100% on-premises · closed-network operation · meets finance, healthcare, public-sector, and defense requirements
Five Scenarios That Deliver Immediate Value
Research Institutes · Government Programs
Self-service GPU allocation for students and researchers · project-level quotas make budget execution transparent
AI Startups
Turn in-house GPUs into an internal SaaS · cut spend on external clouds like RunPod and Lambda
Enterprise R&D
Affiliate Domains · departmental Projects · team-level usage history recorded automatically · internal chargeback and audit readiness
Education · MOOC
1-click workspace provisioning per course · bulk reclamation at semester end
Public Sector · Defense
The same UX in externally isolated environments via an internal Harbor mirror (air-gapped)
Start with a 1-Week PoC
Install Vulcan on your own GPUs and provision five workspaces — in a 1-week PoC you can directly compare cost and wait times against cloud GPUs. Technical documentation is available for advance review, and for air-gapped environments we separately discuss an internal Harbor mirror configuration. vulcan.apps.azwell.ai · +82-2-550-4888