Skip to content

Add VoltageGPU (Intel TDX confidential GPU instances, public JSON price feed) #79

Description

@Jabsama

Hello, and thanks for keeping this running for so long.

VoltageGPU (https://voltagegpu.com) is a small French provider not currently in the database. We publish a machine-readable price feed, so adding us should not cost you any scraping work.

Feed

https://voltagegpu.com/api/pricing/snapshot — public, no key, refreshed every five minutes. It carries, per tier: resourceName, model, gpuCount, vram, pricePerHour (total for the instance), available (live count) and gpuAttestation.

Current on-demand instances

Tier GPUs VRAM Total $/h $/GPU/h
rtx6000b-small 1x RTX PRO 6000 Blackwell 96 GB 3.80 3.80
h100-small 1x H100 80 GB 6.95 6.95
h200-small 1x H200 141 GB 8.08 8.08
h100-xlarge 8x H100 640 GB 55.62 6.95

Billing is per second, no minimum and no commitment; one hour is prepaid at deploy and the unused part is refunded on release. There is also a separate standard tier without an enclave, at lower prices, which we never describe as confidential.

Why we might be worth a row beyond the price

Every instance above runs inside an Intel TDX trust domain, and the tenant generates the Intel TDX quote and the NVIDIA GPU attestation themselves, inside the VM, on a nonce they choose, then verifies both against Intel and NVIDIA rather than against us. Which SKUs have dated, published evidence and which do not is itself a public endpoint: https://voltagegpu.com/api/attestation/evidence

Since your advanced filters already include Virtualisation and Provider Accreditations, confidential computing may be a dimension worth modelling at some point. Nobody in this market seems to expose it in a comparable way today, and the honest version of it includes the limits: on our 8-GPU node the GPUs attest individually but the NVSwitch fabric is not attestable from inside a TDX guest, which we published at https://voltagegpu.com/blog/nvswitch-not-attestable-from-tdx-guest-8x-h100

Company

VOLTAGE EI, France, SIREN 943 808 824. Not to be confused with Voltage Park, a different US provider already in several comparison sites. GPU capacity is hosted by a US sub-processor, listed publicly; the contracting entity is French.

Happy to adjust the feed if a field would make ingestion easier.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions