How to benchmark Fastvideo GPU software on your own hardware
Updated:
Every performance number we publish is reproducible. Download our free Windows demo apps and measure the speed yourself — on your own NVIDIA GPU, with your own images, using the built-in timer or any third-party profiler. GPU image processing speed depends on the graphics card, image resolution, pixel format, bit depth and settings, so the most accurate answer to “how fast will it be for me?” comes only from a measurement on your hardware and your data. Any generic third-party benchmark answers that question only indirectly.
Why isn’t there an independent “official” benchmark?
Because there is no single universal number: the result depends on your GPU, your images and your parameters. A generic benchmark measured on someone else’s hardware and data is only an approximation. So instead of a certificate, we give you the means to reproduce the measurement — open demo apps that anyone can run to get an exact answer for their own workload.
What to download and run
Ready-to-run Windows demo apps (hosted here on fastcompression.com) measure the speed of each module on your GPU:
- GPU JPEG codec on CUDA — fast_jpeg.zip
- GPU JPEG2000 codec on CUDA — fast_j2k.zip
- GPU debayer / demosaicing on CUDA — fast_debayer.zip
- GPU image resize on CUDA — fast_resize.zip
- ALTMapper local tone mapping (demo) — ALTMapper_demo.7z
- FastVCR for machine vision cameras (portable) — FastVCR_Portable.7z
- Open source: gpu-camera-sample, pgm2dng
- The full Fastvideo SDK for Windows, Linux or ARM — sent on request
Test environment
The numbers in the tables on our product and benchmark pages are measured on NVIDIA GPUs at the parameters stated on each page. We report the pure GPU processing time (encode, decode, debayer, and so on); speed is given in frames per second (fps) or in gigapixels per second (GPix/s). A representative desktop configuration:
| GPU | NVIDIA GeForce RTX 4090, PCI-Express 4.0 ×16 |
|---|---|
| CPU | AMD Ryzen 9 7950X (16 cores, 4.5–5.7 GHz) |
| OS | Windows 10 / 11 (Linux with MPS gives equal or better results) |
| CUDA | CUDA 12.6 |
| Data / timing | All data in GPU memory; timing includes GPU computation only |
| Image parameters | Resolution Full HD / 2K / 4K; 8-bit or 24-bit; subsampling 4:4:4 / 4:2:2 / 4:2:0; JPEG quality 90–95% |
| Metric | Frames per second (fps) or gigapixels per second (GPix/s) |
How to reproduce a measurement
- Download the demo for the module you need and run it on a PC with an NVIDIA GPU.
- Set the same parameters as in the table (resolution, format, bit depth, quality), or your own — matching your real task.
- The demo reports the processing time and speed; compare it with our published numbers.
- If you like, cross-check with third-party tools (for example NVIDIA Nsight); the result should match.
Products with published benchmarks
- GPU JPEG codec — JPEG benchmarks
- GPU JPEG2000 codec — JPEG2000 benchmarks
- GPU debayer / demosaicing — debayer benchmarks
- GPU denoiser
- RawZipper RAW codec — RawZipper benchmarks
- Fastvideo SDK — all benchmarks
About the author
Fyodor Serzhenko, PhD, is the founder and CEO of Fastvideo. He earned his PhD at the Moscow Institute of Physics and Technology (MIPT) in 1993. Since 2009 he has led the development of Fastvideo’s GPU-accelerated image codecs and ISP modules, including the Fastvideo SDK. Connect on LinkedIn.
Why you can trust these results
Fastvideo has built GPU-accelerated image processing software since 2009. Our codecs comply with the standards they implement — JPEG (ITU-T T.81 / ISO IEC 10918) and JPEG2000 (ITU-T T.800 / ISO IEC 15444) — and our tools are open source on GitHub. Every performance figure we publish is reproducible: see our benchmark methodology, download the Fastvideo SDK benchmark report (PDF), and measure it on your own GPU.