A FLOP is a floating-point operation — one unit of the math your GPU does every second. FLOPS (FLOPs per second) is how we talk about GPU speed: more FLOPS means more work done, faster.
Floprate measures your card's FLOPS while stress-testing its stability, so you can catch a “flop” — a failing or unstable GPU — before it costs you.
Floprate delivers OCCT-level fault detection — and goes further by identifying which low-level component is faulting. CUDA / shader cores, Tensor cores, TMUs, ROPs, VRAM, and more are probed under load, so you see the exact failing unit on a timeline instead of a single pass/fail. Tap a component in the demo to learn what it does.
Floprate breaks raw compute into distinct metrics: FP32, FP16, and FP64 vector throughput; Tensor core throughput (FP16, BF16, TF32, FP8, INT8) for AI workloads; memory bandwidth; and health telemetry like power draw and thermals. Hover a spoke on the radar to see what each one means.
As users benchmark, Floprate collects telemetry across GPU models and calculates average scores for each model. Your result is then compared against that field average — and the best recorded scores for the same model — so you can see where your card stands, not just a number in isolation.
Run our native benchmark client on your machine. Floprate verifies hardware identity, stress-tests for faults while measuring throughput, and produces AI performance scores sealed to that GPU via TPM.
Every test is cryptographically signed using a Trusted Platform Module (TPM) quote, proving the exact GPU model and configuration. Telemetry stays tied to the hardware profile — forged logs and spoofed specs fail verification.
Awaiting Pipeline Run
Start the telemetry simulation on the left to see how raw measurements become verified benchmark scores.
Once the attested benchmarker has traction, Floprate will open a marketplace where buyers and sellers trade GPUs backed by these verified stability and performance reports — not unverifiable screenshots.