GPUTrust Logofloprate

What is a FLOP?

A FLOP is a floating-point operation: one unit of the maths 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.

Floating Point Operation
3.14×2.71=8.51
2.00 TFLOPS

floprate leads in GPU fault identification

All tests were run on the same faulty NVIDIA GeForce RTX 2080 Ti.

floprate

FAILED

4,985 faults · first at 0:08

  • Shader coresClean
  • Texture unitsClean
  • Special maths units4,985 faults
  • Tensor coresClean
  • RasterizerClean
  • Output unitsClean
  • Video memoryClean

Error timeline

TFU
Fatal

floprate identifies which 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.

The Metrics

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 or focus a metric pill to learn what it measures.

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.

AI Training

Training specific metrics and counters for AI workloads.

AI Inference

Inference performance and utilisation metrics.

Scientific Compute

High precision compute and HPC related metrics.

General

Memory bandwidth, power draw and thermal health.

From Telemetry to Trust

Prove your GPU is stable with cryptographically signed results. Give buyers confidence and your card a stronger case for a higher price.

Stability

STABLE

This GPU is stable and all tested components are healthy.

No faults detected, clocks on spec, temperatures in range.

No errors recorded

Tested device

NVIDIA GeForce RTX 4090

24 GB GDDR6X

Avg Temp
58.9°C
Max Temp
67.4°C
Avg Power
302W
Peak Power
422W
Max Clock
2790MHz
Max Mem Clock
10501MHz

Power & Temperature

Power Draw
Temperature

Utilisation & Clocks

GPU Util
Memory Util
Graphics Clock
Memory Clock
Coming Soon

A Marketplace Built on Verified Scores

We are building towards releasing a marketplace where buyers and sellers trade GPUs backed by these verified stability and performance reports, not unverifiable screenshots.