Simple GPU pricing
Instances for persistent GPU machines, billed per minute. Sandboxes for short, isolated jobs with or without a GPU, billed per second. Snapshots to keep your disk between runs.
GPU instances
Persistent GPU machines, ready to SSH into in minutes.
| GPU | VRAM | vCPUs | RAM | Disk | Action | |
|---|---|---|---|---|---|---|
| RTX A6000 | 48 GB | 6 included · 8 max | 48–64 GB | 100–500 GB | $0.35/hr | Launch A6000 → |
| L40 | 48 GB | 6 included · 12 max | 48–96 GB | 100–500 GB | $0.79/hr | Launch L40 → |
| A100 80GB | 80 GB | 8 included · 16 max | 64–128 GB | 100–500 GB | $1.09/hr | Launch A100 → |
| H100 SXM | 80 GB | 4 included · 16 max | 32–128 GB | 100–500 GB | $3.20/hr | Launch H100 → |
| Extra vCPU | Above the included vCPUs | RAM scales with vCPUs | $0.04/vCPU-hr |
Storage and data
Instance disks and snapshots, with no fees to move data out.
| Storage | Details | PriceMetered by the hour; monthly rates assume 730 hours. |
|---|---|---|
| Instance disk | First 100 GB per GPU included; billed only while running | $0.219/GB/mo |
| Snapshots | A saved instance disk, kept until you delete it | $0.05/GB/mo |
| Data egress | No egress fees | $0.00/GB |
Sandboxes
Isolated machines for agents and short, automated jobs, with or without a GPU. CPU-only sandboxes pay just the vCPU, memory and storage rates. Request access and our sales team will follow up to get you set up.
Sandboxes scale up and down to 0, so you pay exactly when workloads are running.
| Resource | Price |
|---|---|
| H100 SXM | $0.000819/GPU-sec |
| A100 80GB | $0.000542/GPU-sec |
| vCPU | $0.0000142/vCPU-sec |
| Memory | $0.00000444/GiB-sec |
| Storage | $0.0000000333/GiB-sec |
| Data egress | $0.00/GB |
Estimate your bill
Pick a configuration and how long it runs, and see the monthly total.
- A100 80GB × 1176 hrs × $1.09$191.84
- 4 extra vCPUs176 hrs × $0.16$28.16
- 200 GB extra disk176 hrs × $0.06$10.56
- Snapshots, 100 GB$0.05 per GB/mo$5.00
$1.31/hr while running · 176 hrs
How does per-minute billing work?
Instances, including extra vCPUs and disk, are billed per minute only while running. Snapshot storage is billed separately until you delete the snapshot. Sandboxes are billed per second until they stop or their TTL runs out.
How does Thunder Compute price multi-GPU instances?
Pricing is listed by GPU type and GPU count. You choose the GPU, number of GPUs, CPU/RAM, and storage, then pay per minute for the resources you use.
Does Thunder Compute offer reserved or volume pricing?
Yes. Teams can reserve dedicated GPUs for a fixed term at a lower rate than on-demand, and we offer contract and volume pricing. Contact sales to get a quote.
Does Thunder Compute offer persistent storage?
Yes. We offer persistent storage in the form of snapshots. When you finish your work you can take a snapshot of the instance and restore from it another time. Snapshots are billed hourly according to size.
Does Thunder Compute charge data egress fees?
No, Thunder Compute does not bill for data egress.
How does Thunder Compute keep GPU costs low?
We keep GPU costs low by scheduling workloads efficiently and reducing idle time across our infrastructure. We pass those savings on to you through lower GPU prices.
How do I check GPU availability, and what if a GPU is sold out?
Check the current GPU options with tnr create or use the MCP get_availability tool for live stock by configuration. Availability can change before you launch. If creation fails because a type is sold out, check again and choose another available configuration or try later.
Can my instance or sandbox be preempted?
No. Instances and sandboxes are never preempted. They run until you stop them or, for sandboxes, their TTL ends.