Evidence on the listing
- published hardware benchmarks
- auditor-signed provider attributes
- attestation where confidential compute is required
- region and interconnect notes
Marketplace · GPU & AI
GPU & AI Compute on VirtEngine
Accelerator capacity as a catalogue product: GPU machines, dedicated nodes, training allocations and inference endpoints — with benchmarks, attestation and metered settlement.
- 01 GPU listing
- 02 Order / bid
- 03 Lease
- 04 Provider daemon
- 05 GPU backend
- 06 Node · endpoint
- 07 Metered usage
- 08 Settlement
Benchmarks describe the hardware. The provider's backend provides it; the chain records and settles the lease.
Product types
How GPU capacity is listed
Buyer view: what GPU capacity can you source? (Provider economics live on the provider solution page, not here.)
- Workload: Training / Inference / Batch
- GPU: A100 / H100 class
- Memory: ≥ X GB
- Topology: single / multi-GPU
- Region: AU / ap-southeast
- Attestation: required / optional
GPU & AI
Illustrative exampleGPU Virtual Machine
Example Provider
Example region
- 1 × A100 80GB
- 16 vCPU
- 64 GB RAM
$0.11 / vCPU-hour + GPU-hour
GPU & AI
Illustrative exampleDedicated GPU Node
Example Provider
Example region
- 1 × H100
- 24 vCPU
- 192 GB RAM
- NVMe scratch
$1.29 / GPU-hour
GPU & AI
Illustrative exampleMulti-GPU Training Node
Example Provider
Example region
- 4 × GPU
- 96 vCPU
- 768 GB RAM
- Interconnect-aware
Reserved plan
More examples
- Training Allocation — reserved cluster slice
- Inference Endpoint — managed autoscaling runtime
- HPC GPU Queue — scheduler-backed allocation
The same accelerator capacity can be sold as a machine, a node, a reservation, a queue allocation or a managed endpoint. What changes is who operates the software layer above it.
Trust
Benchmarks, attestation and metering
Accelerator listings are where evidence matters most: two nodes with the same GPU model can perform very differently.
GPU hours 86.2 h
Storage GB-months 1.4 GB-mo
Illustrative GPU-hour and storage meters.
Related guides
Where the depth lives
GPU compute providers
How GPU operators list, bid, benchmark and get paid — the provider side of accelerator capacity.
AI & ML workloads
Training, fine-tuning and inference patterns, and how they map onto marketplace leases.
HPC & batch
Scheduler-backed GPU queues, allocations and job templates.
Under the hood: x/benchmark for published hardware performance, x/audit for auditor-signed attributes, x/enclave for confidential compute attestation.
Continue