What is the GPU Instance Specification (GIS)?

GIS is a vendor-neutral, machine-readable JSON format for describing any GPU cloud offering. One JSON document fully describes a GPU instance — hardware, compute, pricing, availability, and normalized comparison metrics.

The GIS specification is open and free forever under CC BY 4.0. The computespec.dev site is free to browse. Pro features like historical pricing data, alerts, and API access are paid.

How is computespec different from GPU comparison sites?

Comparison sites scrape and normalize data independently. computespec defines the open standard format itself — the layer underneath all comparison tools. Any tool, marketplace, or provider can adopt GIS format.

Is it true that gis = gpu instance specification — an open json format?

GIS = GPU Instance Specification — an open JSON format

Is it true that one gis document = one gpu cloud offering, fully described?

One GIS document = one GPU cloud offering, fully described

Is it true that 6 sections: gpu, compute, pricing, availability, normalized, meta?

6 sections: gpu, compute, pricing, availability, normalized, meta

Foundational4 min read

What is GIS?

The GPU Instance Specification is an open, vendor-neutral JSON format for describing any GPU cloud offering. One document per instance. Machine-readable. Comparable.

On this page

What it is

The GPU Instance Specification (GIS) is an open JSON format that describes a GPU cloud offering in a standardized way. One JSON document = one GPU instance from one provider, fully described.

Think of it like a nutrition label for GPU cloud. Every food product has the same label format — calories, fat, protein, carbs — so you can compare any two products instantly. GIS does the same for GPU instances — model, VRAM, TFLOPS, pricing, availability — in one consistent format.

GIS is published under CC BY 4.0 — free to use, share, and build on for any purpose, including commercial.

Why it exists

The GPU cloud market has 50+ providers and $50B+ in annual spend, but no common format for describing offerings. Every provider uses different names, different pricing structures, different spec sheets.

AWS:        p5.48xlarge
Lambda:     gpu_1x_h100_sxm5
RunPod:     a100-80gb-sxm
CoreWeave:  h100-sxm-80gb

Same hardware. No shared vocabulary.

Every comparison tool, procurement system, and AI agent independently scrapes, normalizes, and structures the same data — producing incompatible representations. GIS eliminates this duplication by defining the format once.

The six sections

A GIS document has 6 sections (4 required, 2 optional):

Section	Required	What it describes
`gpu`	✅	GPU model, VRAM, TFLOPS, bandwidth, interconnect, architecture
`compute`	✅	vCPUs, RAM, storage, network
`pricing`	✅	On-demand, spot, reserved prices, billing granularity
`meta`	✅	Last updated, source URL, verified status, spec version
`availability`	○	Regions, instance type, SLA uptime
`normalized`	○	Computed comparison metrics: cost/GPU/hr, cost/TFLOP/hr, VRAM/dollar

Extra fields are allowed at any level. Validators ignore unknown fields. Tools preserve them. This makes GIS forward-compatible — you can add provider-specific data without breaking anything.