21/08/2026
The most important media-tech release this month may not have “AI” in the product name.
It’s FFmpeg 9.0.
FFmpeg sits underneath an enormous amount of video infrastructure people use every day without ever seeing it. Encoding, transcoding, streaming, filtering, conversion, packaging — and often somewhere inside much larger commercial products.
Version 9.0 is interesting because hardware acceleration continues to become much more than simply sending the final encode to the GPU. There is additional Vulkan work, CUDA filtering, expanded AMD AMF capabilities, AVX-512 optimisations and more.
For us at RTCnow, this is the part that matters.
When you operate real streaming infrastructure, performance is not measured by whether a feature looks impressive in a changelog. It is measured in streams per GPU, GPU memory, CPU load, latency, stability and what happens after the system has been running continuously rather than during a five-minute benchmark.
Every additional operation that can stay efficiently on the GPU potentially matters.
But a major FFmpeg release also means something else: we do not upgrade production simply because the changelog looks interesting.
The process is much less exciting.
Build it. Reproduce existing pipelines. Compare output. Measure CPU and GPU utilisation. Test the hardware paths. Run it under load. Find what broke. Fix or work around it. And only then think about production.
Open source is sometimes described as the cheaper alternative to commercial media software.
After many years of operating streaming systems, I think that misses the point.
The real value of tools like FFmpeg is that they give us the ability to understand, measure and optimise almost every part of the media pipeline.
That is much more valuable than “free”.
Image generated with AI
