Resource Tiers
A run’s compute size determines how much CPU, RAM, and GPU its worker gets, and therefore how fast it completes and how many credits it burns. You choose the size with sliders (CPU, RAM, GPU count) each time you run, bounded by your plan.
Choosing a size
In the editor’s Cloud compute panel:
- CPU slider, 1 to 64 vCPU (Free plan caps at 4).
- RAM slider, 4 GB per vCPU minimum, up to 512 GB (256 GB on GPU workers).
- GPU slider, 0 (CPU-only) to 4 NVIDIA A10s. Any GPU routes the run to a GPU worker; the dispatch optimizer auto-matches the cheapest covering instance (OCI A10 first, AWS fallback).
- Auto-size, on by default: the panel sizes CPU/RAM from the workflow itself (the busiest concurrent stage’s thread demand, the engine runs independent nodes in parallel, each claiming its
threadssetting) and picks a GPU automatically when any node requires one. Any manual slider edit turns auto off; one click re-enables it.
The billed rate is the larger of 0.00127 credits/sec per vCPU and 0.000195 credits/sec per GB of RAM:
| Size | CPU | RAM | Credits/sec | Credits/hour | Good for |
|---|---|---|---|---|---|
| 1 vCPU / 4 GB | 1 | 4 GB | 0.0013 | 5 | Tiny tests, scripting nodes |
| 4 vCPU / 16 GB | 4 | 16 GB | 0.0051 | 18 | QC, bacterial genomes |
| 8 vCPU / 64 GB | 8 | 64 GB | 0.0125 | 45 | Human WES, RNA-seq |
| 16 vCPU / 128 GB | 16 | 128 GB | 0.0250 | 90 | Human WGS, STAR, Cell Ranger |
| 32 vCPU / 256 GB | 32 | 256 GB | 0.0499 | 180 | Large assemblies |
| 64 vCPU / 512 GB | 64 | 512 GB | 0.0998 | 359 | Huge eukaryotic assemblies |
GPU workers add 0.025 credits/sec per A10 (≈ $0.90/hr) on top of the CPU/RAM formula, so the 4 vCPU / 16 GB GPU base runs at ~108 credits/hour, and each additional A10 adds ~90 credits/hour. GPU workers are available on every plan (Free tops out at the 4/16 base).
Subscription caps
Your plan caps the largest size you can run and how long a single job may run:
| Tier | Max size | Max runtime/job | Concurrent jobs | Storage |
|---|---|---|---|---|
| Free | 4 vCPU / 16 GB | 1 hour | 1 | 5 GB included, then metered |
| Starter | Unlimited | Unlimited | 2 | Pay-as-you-go |
| Professional | Unlimited | Unlimited | 5 | Pay-as-you-go |
| Team | Unlimited | Unlimited | 10 | Pay-as-you-go |
| Enterprise | Custom | Unlimited | Custom | Custom |
Storage
Storage auto-scales with no cap. Every plan gets an included allowance (5 GB on Free; paid plans are metered from the first GB) and anything beyond it is billed at 3 credits per GB-month, metered daily in 1/30th slices. If a team runs out of credits while past its allowance, new uploads and runs pause until a top-up, stored data and downloads are never blocked, and deleting files immediately frees the allowance.
Sizing tips
- Let auto-size pick. It reads each node’s threads demand and sizes for the busiest concurrent stage, usually the right call.
- Match RAM to the bottleneck node. Alignment and assembly nodes are usually the memory ceiling; pick a size that comfortably fits the largest step.
- Bigger isn’t always pricier per job. A larger instance finishes faster; since billing is per-second, a 2× faster run on a 2× costlier instance is roughly cost-neutral, but with less wall-clock time.
- Test small, run big. Validate a graph on a subset at 4 vCPU / 16 GB, then run the full dataset at a larger size.
- Preview the cost. The Compute panel shows a live credits/hour quote using the same formula the server bills at, and
POST /api/billing/estimatereturns the exact per-second rate for a spec.