Streaming Learning Center Where Streaming Professionals Learn to Excel
Related Articles
On February 12, I interviewed David Ronca, formerly of Netflix and Meta, now RoncaTech, about the latest Beta version of his Video Codec Acid Test, or VCAT, which now includes VVC decode courtesy of Fraunhofer’s open-source VVeC decoder. The interview is available here on YouTube. Here’s an overview of the highlights we covered, with timestamps for the video.
To get a copy of VCAT, check out the release notes (with download link) here. You can contact Ronca on LinkedIn here.
Here are the notable points in the interview with the VCAT demo available below.
00:00:00 — Introduction. Jan introduces David Ronca: per-title encoding pioneer at Netflix, key figure in AV1 adoption at Meta, now running his own development company RoncaTech and the VCAT project.
00:00:32 — What VCAT is and where it came from. VCAT (Video Codec Asset Test) is a decoding benchmarking tool that encapsulates the manual device testing Meta did over five years to achieve 75% AV1 reach on Facebook and Instagram globally. The goal: make that level of benchmarking accessible to anyone without the pain.
00:01:29 — Why software decoding demands its own benchmarking approach. Hardware decoders announce their capabilities via levels. Software decoders don’t. A device with a 1080p screen may have no business software-decoding 1080p AV1 or VVC. Battery drain, thermal events, and system overload are the real metrics — and they require long-duration tests, not short synthetic ones.
00:05:04 — The global Android problem. US deployments are relatively easy — iPhones dominate, high-end Androids fill the gaps. Go global and you’re dealing with $100-$150 Android devices running software decoders on modest SoCs. That’s where VCAT earns its keep.
00:07:02 — Level-aware playback and the benchmarking case. The principle David applied at Meta: benchmark devices, establish what each class of device can actually handle, then enforce level-equivalent restrictions in software. VCAT gives any operator that same capability.
00:08:50 — Who else should be using this? Network operators evaluating which low-cost handsets to bundle, handset vendors wanting to prove their $150 device is competitive, and SoC vendors pitching to those handset vendors. Video quality is the dominant user experience metric, so video decode performance should inform device selection.
00:12:42 — How VCAT ended up open source under RoncaTech. Meta decided that VCAT didn’t fit into its open source portfolio. David took it on himself because he believes it matters to the ecosystem. Released under GPL, available as APK or source.
00:15:40 — Beta 4: what’s new. Two main additions: test vector management (shareable, standardized benchmark sets that any company can publish for their specific content profile) and VVdeC integration, enabling VVC software-decoding benchmarks for the first time.
00:22:27 — VVdeC vs. AV1 benchmark results. On a test device, AV1 (using the VVdeC-equivalent AV1 decoder) ran about 21 hours before hitting 15% battery. VVdeC drained roughly three times faster, with occasional thermal events. Fraunhofer acknowledged their decoder isn’t optimized — a private commercial VVC decoder would likely show better numbers.
00:27:27 — VCAT Demo. Showing test configurations, VVC playback, and telemetry captured during testing.
Commenting on how VVC playback compared to AV1, Ronca said that on his test device, Dav1d 1.5 delivered around 21 hours of playback before hitting 15% battery, while VVdeC 3.0 drained roughly three times faster. He noted occasional thermal events with VVdeC but no thermal shutdowns.
He added a caveat that when he published the results, a Fraunhofer engineer acknowledged that VVdeC is nowhere near as optimized as Dav1d, so the numbers weren’t a surprise — and that a commercial VVC decoder from a company like Tencent or TikTok would likely show considerably better performance.
00:34:00 — Future directions: encoder, GPU, and AI benchmarking. David sketched out a roadmap: a VCAT encoder variant to measure the cost side of the codec efficiency equation, GPU benchmarking for compositing and rendering scenarios, and AI codec testing—specifically, whether NPU-accelerated AI decoders deliver real battery-life gains or just paper bitrate savings.
00:44:31 — How to get VCAT. APK and GitHub source via David’s LinkedIn post. Test vector toolset available as Python scripts. Not on the Play Store due to file system access requirements. David is open to 30-minute conversations with anyone who has a use case — no NDA required.
Clarification on the ownership of VCAT. The original intent was to build on VLC Benchmark, and VCAT would remain under VideoLabs license. When VCAT was rewritten we needed a different ownership model. META decided vcat did not fit in their OSS portfolio. Since I wrote most of the code, they let me release under RoncaTech.