Right, HEVC playback with ~50 Mbit/s bitrate works using GPU decoding. You used 2 Media Player nodes to be able to use different decoding chip.
And you have an issue with 3 channels in second Media Player are not visible on second GPU outputs. We had similar problem with Touch Engine integration when it worked on one GPU and we have a mosaic setup and we see it’s output only on outputs of one GPU. The workaround was to use Multi GPU Image node as a bypass - it downloads the texture into memory and uploads again. It adds one frame latency and bus usage, but fixes the issue. Test patterns works because they are rendered on both GPU in a mosaic setup.
I don’t think updating the NVidia driver would help.
When you’re connecting with Team Viewer you see all 6 channels - Screenberry output must be composed through Desktop Window Manager in this case. You can force this without teamviewer by making Canvas size 1px smaller or bigger than the full desktop mosaic resolution.
We will look into this issue with GPU decoding and may fix for the next minor release 3.4.3.
Prores does not have accelerated decoding and it show low performance - that is expected. We only recently added hardware accelerated decoding on macOS for prores. It will be available in next major release 3.5. There is no available API for hardware accelerated prores decoding on other platforms.
My earlier suggestion to use the NotchLC codec still stands.
MPEG-2 plays 6 x 4096x2160 streams at 48 FPS while you need 60. We have setups where MPEG-2 plays 8K x 8K at 60, so this can be limited by your memory performance.
Check that the installed memory modules are configured to use all available memory channels (e.g. dual-channel, quad-channel, etc.). MPEG-2 decoding is sensitive to memory bandwidth. Also verify whether the system uses DDR5 or DDR4 memory.
It’s also preferable to have PCIe Gen4 for faster memory-to-GPU transfers. Finally, verify that the graphics cards are running with a PCIe x16 link width.
You can also try to use Display Output nodes instead of Canvas. Switch Canvas to offscreen mode, and connect all outputs to Display Ouptut node(s). You can try that with one Display Output and mosaic, or with mosaic disabled and 6 separate Display Output nodes. You can build/remove mosaic from Screenberry Panel via Displays Control, you just need to have Screenberry server running with administrative rights.
Another thing to test with MPEG-2: enable Async Long texture transfer mode in Media Player advanced parameters. This should have 20-30% performance improvement. It is known to have issues on multi GPU mosaic setups: frame tearing on jumps 3-4 frames back and forward on outputs of second GPU, but that some time fixed using “Force glFinish()” in Canvas node, or by outputing through Display Output, or by forcing to use Desktop Window Manager by changing its size. It is somewhat new not fully explored area for us, the behavior may vary between different NVidia drivers, so you can play with it.