Fastvideo company has released ultrafast JPEG codec for NVIDIA GPU. Peak performance of the codec reaches 6 GBytes per second and higher for images loaded into RAM. For instance, a full-color 4K image with resolution 3840 x 2160 can be compressed by 10 times in merely 6 milliseconds on NVIDIA GeForce GTX Titan.
Fast compression and decompression of JPEG images has a wide range of applications in industries where handling of large volumes of data in real time is required, or for operation of professional equipment generating massive streams of images. This technology is extremely useful for high speed and high resolution cameras, as well as for many other applications including processing and visualization of 4K and 8K images in real time. Since JPEG format is still highly widespread across the web, fast compression, decompression and processing of such images remains very important for many web applications including heavy-loaded social networks and image databases.
Ultrafast JPEG codec developed by Fastvideo applies NVIDIA CUDA technology to boost image compression and decompression using parallel computing. The codec implements highly optimized approach that parallels absolutely all stages of Baseline JPEG algorithm including entropy encoding and decoding. And with powerful NVIDIA GPU such solution has allowed to achieve utterly high performance of the codec that leaves all competing products behind. Usage of GPU frees up CPU resources for other tasks and provides unequalled performance during JPEG compression and decompression.
In particular, on NVIDIA GeForce GTX Titan, Fastvideo JPEG codec delivers encoding throughput up to 6 GBytes per second for images stored in operating memory of the computer, and that significantly exceeds the peak performance of all existing multithreaded JPEG codecs based on standard processors including multi-core CPUs. Such compression and decompression speeds are neither available at FPGA-based hardware solutions. Fastvideo JPEG compressor can encode color 4K image with resolution 3840 x 2160 and compression ratio 10 with 4:4:4 sampling in merely six milliseconds.
Fig.1 - Performance of JPEG compression and decompression on GPU and CPU for the following codecs:
Aside from JPEG compression and decompression, JPEG codec finds applications in related fields as well. It can be used to solve the task of batch image resize for already compressed images. JPEG resize task requires fast image decompression, resize and compression. GPU allows to carry out JPEG resize really fast, therefore such solution is attractive to web services focused on processing of photos, which are usually stored in JPEG format. In this case, one don’t have to preliminary generate and store several copies of the same image with different resolutions, instead, the required size of the image is created from the source image in real time upon user’s request.
JPEG SDK from Fastvideo for ultra fast image compression and decompression is available for Windows 7/8 and NVIDIA GPU series 400, 500, 600 and 700. GPUs based on Fermi, Kepler and Titan architectures meet these criteria. Demo version of JPEG codec for NVIDIA GPUs is available from Fastvideo website and works under Windows 7/8. Linux version of the codec is to be released in the nearest future.
Fastvideo was founded in 2009 in Dubna, Russia. It is one of a few developers of high speed cameras in Russia also specializing in GPU image processing. Ultrafast JPEG codec created in Fastvideo is currently the fastest JPEG codec on the market. The company also develops custom solutions for real time image processing and compression on GPU.
Download page: http://www.fastcompression.com/download/download.htm
Here one can find the latest results for performance of CUDA JPEG Codec.