Intelligent routing for the compute era.
Dyson dynamically routes HPC and AI workloads to optimal hardware based on cost, latency, and hardware affinity. No manual orchestration.
What is Dyson?
Modern compute workloads run across heterogeneous hardware: cloud GPUs, on-prem clusters, decentralized nodes, and edge devices. Dyson sits between your application and all available compute, making real-time routing decisions based on three dimensions:
Cost
Lowest price for the job
Latency
Fastest time to first token
Affinity
Hardware-specific optimization
Dynamic routing visualization
Dyson evaluates all compute targets and picks the optimal path.
Incoming job
Batch inference
Dyson router
Scores every target against cost, latency, queue depth, and hardware affinity before reserving credits.
GPU L4 Pool
London edge
Score
96
Price
0.08 cr/min
Latency
41ms
GPU A100 Pool
Frankfurt cloud
Score
78
Price
0.21 cr/min
Latency
72ms
CPU Batch
Spot cluster
Score
61
Price
0.02 cr/min
Latency
128ms
Decision time
184ms
Credits reserved
3.2 cr
Policy match
GPU affinity
Key features (planned)
Dynamic Routing
Automatically route workloads to the optimal compute target based on real-time availability and constraints.
Cost Optimization
Minimize infrastructure spend by selecting the cheapest viable hardware for every job, without manual bidding.
Hardware Affinity
Pin latency-sensitive workloads to specific GPU architectures or regions for predictable performance.
Load Balancing
Distribute workloads evenly across available hosts. Prevent hotspots and maximize cluster utilization.
Plans for routed compute workloads
Dyson has a monthly platform plan plus usage credits. Usage is billed by runtime, hardware type, and routing complexity.
Developer
For developers and small teams running routed workloads through Dyson.
- Includes monthly compute credits
- Run Python, container, and model inference jobs
- CPU and GPU routing support
- Basic cost and latency tracking
- App and API access
- Best-effort job scheduling
Growth
For startups and engineering teams with recurring workloads.
- Larger monthly credit allocation
- Lower effective usage rates
- Priority job scheduling
- Multi-profile CPU and GPU execution
- Job history, logs, and analytics
- Cost-performance recommendations
Enterprise
For companies with high-volume workloads, private infrastructure, or multi-provider compute needs.
- Committed credit pools
- Private routing policies
- Custom CPU, GPU, TPU, or cloud-provider routing
- Cost, latency, and energy analytics
- Capacity planning and deployment review
- VPC/private deployment options
Usage credit meter
Preview meters show how credits can settle by profile while routing complexity and reserved capacity are priced into the plan.
cpu-small
0.02 cr/min
1 vCPU, 512Mi
cpu-balanced
0.05 cr/min
2 vCPU, 2Gi
cpu-highmem
0.12 cr/min
4 vCPU, 8Gi
gpu-l4
0.08 cr/min
4 vCPU, 16Gi, NVIDIA L4
Start with hosted checkout
Developer and Growth plans open Stripe Checkout. Usage still settles through shared CrossGL credits after jobs run.
The future of compute routing
Dyson is under active development as part of the CrossGL compute platform. Developer starts at $29/month plus usage credits, with Growth and Enterprise options for recurring workloads.