garageCube

Hi
I have a machine with a media disk array that gets reads ~5.5GB/s. I also have a test file that is close to max bitrate(intense colour noise and a smoothness indicator) at 8192*2160@50 HAP. I can smoothly play 3 of these in TouchDesigner, 4 occasionally chokes as it's pushing the disk limit. In MM I can only play one, a second copy playing in tandom gets seriously choppy with stalls of .5s etc, unplayable. This is both with and without the hardware decode box checked, with nothing else going on(just two standard quads).
My question is why such a drastic difference in playback performance?
As you can imagine the test file is enormous and not practical to send through, if you have a commercial TD license I can give you the project to render it out on your end.

Sorry I should have mentioned, I'm on latest win10 with latest quadro drivers, and P5000.

Are you using HapQ+Alpha ? That's the only codec that is not natively supported in 3.6. It will in 3.7.
If you're using Hap, HapQ or HapAlpha, MadMapper perfs should be optimal.
Do you have multiple GPUs ? MadMapper on Windows uses the GPU attached to the main display in Windows display settings.

Just HAP, with a single P5000 gpu(main display).

We're looking at your issue & keep you up to date.

The problem is when decoding speed is slower than playback speed. Touch Designer handles this case better, but I can see some frame drops in Touch. It's just getting back on track, while MM doesn't. We could reduce this effect for the case of "key frames" only codecs. Will be in MM 3.7.0.
At the moment, what you should do is recompressing your HAP files specifying the number of "chunks". You could set it to 8. The CPU side decompression of a single movie will then be spread over 8 threads (if you have a quad core with hyperthreading, the 8 thread will decompress in parallel).

Thanks Matt,
As you know you can get precise details on frame drops in TD and I'm not getting any at all. I'd say it's the pre-reading that TD does giving it an advantage here?
I had already tried chunked and non-chunked encoding, with no effective difference.

I PM you a beta build. We're also pre-reading the movie in a separated thread.