Skip to main content

Radix Core

Up to 21% Faster GPU Training

Closed-loop GPU scheduling that balances memory, compute, and power across all training runs. No stack changes. Results in minutes.

Join WaitlistHyperscale Solutions

The Training Problem

GPU schedulers treat hardware as monolithic units. They account for GPU count but ignore memory, compute, and power constraints across training runs.

Stranded Capacity

Memory-bound jobs block compute-bound jobs. Power limits throttle entire nodes. Your fleet has capacity you can't reach.

Throughput Degradation

Up to 21% training throughput is lost to naive scheduling (VaultScaler internal benchmarks). At fleet scale, that's millions in wasted compute every year.

CFO Pressure

GPU spend is a scrutinized line item. You need utilization metrics that justify the investment, not excuses about scheduler limitations.

Up to 21.4% throughput improvement

(VaultScaler internal benchmarks)

over industry-standard FIFO scheduling

How Radix Core Works

Deploy in minutes. See results the same day. No changes to your existing training code or workflow.

1

Install

Install the Helm chart on your existing Kubernetes cluster

2

Configure

Choose zero egress or API mode based on your security requirements

3

Connect

Port-forward the service and connect to the dashboard

4

Optimize

Get GPU insights and throughput gains in 60 seconds

Works with Kubernetes. Your team's workflow stays the same.

# Add Helm repo
$ helm repo add radix \
    https://charts.radix.vaultscaler.com

# Install Radix Core
$ helm install radix-core radix/radix-core \
    --namespace radix-system \
    --create-namespace

# Verify installation
$ kubectl get pods -n radix-system
NAME                          READY   STATUS    RESTARTS   AGE
radix-agent-7f8b4d6c5-x2k9l   1/1    Running   0          47s
radix-dashboard-5c9a8e3f1-m4n  1/1    Running   0          47s
radix-observer-6d4b2a9c8-q7w3  1/1    Running   0          46s

What Radix Core Delivers

Light

Deploy to Kubernetes in minutes. No existing stack changes. No migration. No retraining your team.

Demonstrable

Leadership-ready dashboards comparing Radix vs. your current scheduler. CFO-defensible ROI from day one.

Adaptive

Closed-loop control with continual scheduling refinement that improves over time as it learns your workload patterns.

Secure

Originally built for air-gapped deployment with zero egress. SLSA Level 3 attestations. Your data never leaves your infrastructure.

Security That Satisfies Your Infosec Team

Your GPU fleet processes sensitive training data. Radix Core is built for the most demanding security environments.

Air-Gapped Ready

Deploy in fully disconnected environments. No internet access required after installation.

SLSA Level 3

Cryptographic attestations for build provenance. Verified supply chain security.

Zero Egress

Your training data never leaves your infrastructure. Telemetry is opt-in only.

Radix Core Pricing

Predictable per-GPU pricing. Start with a free 14-day trial.

Trial

Free

14-day trial, up to 400 GPUs

  • Full Radix Core access
  • No stack changes
  • Zero-risk validation
Join Waitlist

Radix Core

$45/GPU/mo

GPU scheduling optimization

  • Full deployment
  • Production support
  • Up to 21% throughput gains
Join Waitlist

Hyperscale

Custom

1,000+ GPU deployments

  • Dedicated support
  • Architecture review
  • Custom SLAs
Get in Touch

Ready to Maximize Your GPU Fleet?

Join the waitlist for early access, or book an architecture call to discuss your fleet.

Join WaitlistGet in Touch