Fastvideo white papers

Updated: September 17, 2026

Fastvideo white papers are our own detailed reports on image and video processing — measurements, methods and results, written for engineers who have to choose a solution and defend that choice.

Each paper exists in two forms: a page on this site, which is always the current version, and a PDF of the same text with a contents page and page numbers — for reading offline, for passing on inside a company, and for citing.

Published

Fastvideo JPEG2000 versus nvJPEG2000 on an NVIDIA RTX 4090: a reproducible benchmark

Edition 2026-09 · 75 pages · PDF, 2.8 MB

Two GPU implementations of JPEG 2000 — the fvJPEG2000 codec of the Fastvideo SDK and the nvJPEG2000 library from NVIDIA — measured on one machine at the same compressed file size. Encoding and decoding speed in every combination of thread count and batch size, quality control at matched file size, the price of a fixed file size, GPU energy per frame, CPU load, and a profiler breakdown that accounts for the difference. The Fastvideo encoder is 3.9 to 6.6 times faster; the two decoders are at the same level on throughput, while the NVIDIA decoder returns a single frame 1.5 to 2.1 times sooner.

In progress

These are being written and measured now. We publish a paper when its measurements are closed, so there are no dates here.

JPEG 2000 on a GPU against JPEG 2000 on a CPU

The same images through three implementations — fvJPEG2000 and nvJPEG2000 on the GPU, OpenJPEG on the CPU — at the same compressed file size, with the CPU build and its options stated. It answers the question we are asked most often: what performance can actually be achieved with JPEG 2000 on a GPU, and what a CPU can still deliver on the same images. The text is written; the full run is being finished.

JPEG 2000 on NVIDIA Jetson Orin AGX and Orin NX

Speed and power consumption of JPEG 2000 on embedded boards, by power mode, including lossless encoding of 12-bit raw frames. For airborne and embedded work the limit is rarely the compute: it is watts, heat and channel, and those are what this paper measures.

Citing and rights

Cite a white paper by its name, edition and address, for example: Serzhenko, F. (2026). Fastvideo JPEG2000 versus nvJPEG2000 on an NVIDIA RTX 4090: a reproducible benchmark. Fastvideo white paper, edition 2026-09.

The code and the measurement data behind this white paper are archived separately: Serzhenko, F. (2026). jpeg2000-benchmark, version 2026.09. Zenodo. https://doi.org/10.5281/zenodo.22811905

The text of a white paper is published under CC BY-ND 4.0: it may be copied and passed on in full, with attribution, but not published in an altered form. The measurement results and the tables are under CC BY 4.0 — take the numbers, recompute them, argue with them.

If one of the papers in preparation is close to what you are building, tell us: we will send it when it is out, and we will know that the topic is worth the work.

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.