Fastvideo parallelizes JPEG pipeline to reach 3800 fps at 4K
Fastvideo has released specifications and performance benchmarks for its CUDA-based JPEG codec, which utilizes full GPU parallelization to achieve 3800 fps for 4K encoding on NVIDIA RTX 4090 GPUs. The implementation supports baseline JPEG standards and is integrated into the vendor's existing GPU SDK for imaging and video processing workflows.
Key Takeaways
- Achieves 3800 fps for 4K encoding and 5500 fps for 1080p on NVIDIA GeForce RTX 4090 hardware.
- Parallelizes entropy encoding and decoding on the GPU, stages traditionally restricted to serial CPU processing.
- Supports 100% Baseline JPEG compliance with optional 12-bit grayscale and color modes.
- Writes restart markers by default to enable significantly faster parallel decoding of standard images.
- Integrates with FFmpeg for MJPEG streams and supports HD-SDI input from Blackmagic and Bluefish cards.
Why It Matters
Fastvideo’s shift to a 100% GPU-resident pipeline solves the specific Latency and PCIe bus congestion issues that plague high-resolution machine vision and broadcast workflows. By eliminating the 'ping-pong' effect between CPU and GPU memory, this codec allows 4K data to be processed at nearly five times the real-world bandwidth of a PCIe 4.0 x16 interface. For the broader ecosystem, it sets a high bar for software-as-a-service encoding density, potentially reducing the need for specialized hardware encoders in live production. Watch for whether high-speed camera manufacturers adopt this SDK for real-time RAW-to-compressed pipelines on edge devices like NVIDIA Jetson.
Additional Context
The release of Fastvideo's CUDA-based JPEG codec coincides with a significant refresh of the underlying compute platform. Per NVIDIA documentation from May 2026, the launch of CUDA 13.3 introduced C++ Tile programming, a model designed to automate memory movement and parallelism. This technical shift supports the industry's broader move toward keeping complex image processing entirely within GPU memory. While traditional codecs like libjpeg-turbo rely on serial CPU execution, the streaming and broadcast sectors are increasingly leveraging these architectural updates to handle the massive data overhead of 8K and high-framerate 4K production. Market dynamics in mid-2026 also reflect a shift in hardware utility. Per Tom's Hardware in June 2026, the NVIDIA RTX 4090 remains a staple for professional creators and small studios due to its 24GB VRAM and performance in non-AI rasterization tasks, even as the newer RTX 50-series targets gaming-centric features like Multi Frame Generation (DLSS 4.5). This longevity makes the 4090 a predictable target for high-throughput software codecs. Additionally, related emerging standards like JPEG AI are beginning to show promise; per IEEE data cited in late 2025, neural-based compression can achieve nearly 2000x faster encoding than VVC on GPUs, signaling a future where parallelized, hardware-accelerated imaging is the default for both legacy and next-generation formats.
Read full article at fastcompression.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source