Skip to main content

Pricing Guide

Galadriel uses marketplace-driven pricing where supply and demand determine GPU costs through a transparent orderbook system. This guide explains how pricing works and strategies to save money.

Core Pricing Model

Market-Driven vs Fixed Pricing

Result: Lower average prices, better utilization, more supply.

Platform Fee

All transactions include a 10% platform fee:
What the 10% covers:
  • ✅ Orderbook matching engine
  • ✅ Payment processing (Stripe fees ~3%)
  • ✅ Kubernetes orchestration
  • ✅ Network setup (InfiniBand config)
  • ✅ Monitoring and support
  • ✅ Platform development and operations

Typical Pricing Patterns

By Time of Day

Price History Example

View historical prices:
Output:

Cost Optimization Strategies

1. Time-Shift Workloads (Save 30-40%)

Run workloads during off-peak hours:
How to do it:

2. Use Limit Orders (Save 10-25%)

Set your maximum price and wait:
How to do it:

3. Monitor the Orderbook

Check prices before buying:
If the spread is wide or prices are high, consider:
  • Placing a limit order below best ask
  • Waiting for off-peak hours
  • Using flexible timing

4. Resell Unused Time

Finished early? Recover costs by reselling:
How to do it:
See the Reselling Guide for details.

Comparing Order Types

Cost Comparison Example

Job: 16 GPUs × 24 hours in us-west-1
Platform fee (10%) is included in all costs shown above.

When to Use Each Order Type

Market Orders:
  • ✅ Deadline-driven workloads
  • ✅ Production inference serving
  • ✅ Time-sensitive experiments
  • ❌ Cost-optimized training
  • ❌ Batch processing jobs
Limit Orders:
  • ✅ Non-urgent training runs
  • ✅ Cost-sensitive workloads
  • ✅ Predictable schedules
  • ❌ Immediate needs
  • ❌ Guaranteed execution time
Flexible Orders:
  • ✅ Maximum cost savings
  • ✅ Batch jobs
  • ✅ Dataset preprocessing
  • ✅ Model evaluation
  • ❌ Interactive development
  • ❌ Fixed deadlines

Billing & Invoicing

How Billing Works

1. Order Placement
2. Order Matching
3. Early Termination

Monthly Invoice Example

View billing:

What’s Included (No Extra Fees)

All instances include:
  • Compute: Full node exclusivity (8 GPUs minimum)
  • InfiniBand: 400 Gbps RDMA, GPUDirect enabled
  • Network Egress: Unlimited data transfer out
  • Storage: At least 2TB of local NVMe storage per node
  • Support: Email/Slack support, documentation
  • API Access: Unlimited API calls
  • SSH Access: Direct SSH to all your nodes
  • Kubernetes Access: Full kubectl access
No hidden fees. Just the 10% platform fee on compute.

Refund & Cancellation Policy

Early Termination by Customer

Cancellation Before Start

Rationale: Protects providers who blocked capacity

Hardware Failure (Platform Responsibility)

Price Guarantees

For Customers

1. No Surprise Charges
  • Limit orders guarantee max price
  • Market orders show price before confirmation
  • No hidden fees beyond 10% platform fee
2. Price Lock
  • Once matched, price is fixed for entire duration
  • No surge pricing mid-job
  • Early termination gets prorated refund at matched price
3. Price History
  • All historical prices publicly visible
  • Can analyze trends before ordering
  • Algorithm recommendations based on patterns

For Providers

1. Minimum Price Guarantee
  • Your listing price is the minimum you’ll accept
  • Platform never matches below your min price
  • Can adjust prices anytime (for future orders)
2. Payment Guarantee
  • Platform holds customer payment upfront
  • You get paid even if customer cancels
  • Automatic payout after job completion

Advanced: Price Estimation

Get accurate price estimates before ordering:
Output:

Next Steps

How the Marketplace Works

Understand orderbooks and order types

Cluster Specs

GPU types, system specs, and access model

Reselling Guide

Recover costs by reselling unused time

Examples

Real-world cost optimization examples