vram.fyi A field guide to local AI hardware.
LOCAL AI HARDWARE

Every build has a place.

Add your hardware or try an example to find yours.
BUILD WORKSPACE

Your build 0 devices

Your builds

Device Qty Memory FP32
The compute landscapeLOG SCALE
Your buildYour componentsOther buildsReference systemDGX SparkClick a build to switch. Tap a reference for details.

A map, not a benchmark.

Multi-device points sum rated memory and FP32. That is a coordinate, not a promise of pooled VRAM, NVLink, or linear token throughput.

Unified RAM is shared with the OS/CPU; the plotted capacity is total system RAM, not guaranteed GPU-usable memory. A memory-only mod moves right, not up. Tensor / FP4 / sparsity multipliers are intentionally excluded.

Catalog details distinguish vendor FP32 ratings, calculated estimates, and imported figures awaiting verification. Estimates are not measured AI throughput.