Copy link to clipboard
Copied
Hi,
just watch the sneak peak on the Mercury Engine again.
I see that the filters are being CUDA accelerated as much as possible thus leaving the CPU to do the other work. In the video it showed how AME rendered MPEG2 faster because the effects filters ran on the CUDA GPU thus freeing up the CPUs (4 quad cores!!!!!) to do the AME work.
What happens if a AME Custom Renderer is using CUDA too? I know that the CUDA-Runtime will allow for this but has the dev team done any testing with this mixed mode? And if so, can you share what the impact is multiplexing the two (or more) CUDA jobs was?
CS5... We'll be happy to provide more details about it in another...3 days, 15 hours, 31 mins...
Now if we're talking about CS4, yes, you can develop an exporter plug-in that uses CUDA for accelerated encoding. Elemental Technologies did this with their Elemental Accelerator. And Divide Frame built a CUDA-accelerated importer with their GPU Decoder.
Copy link to clipboard
Copied
CS5... We'll be happy to provide more details about it in another...3 days, 15 hours, 31 mins...
Now if we're talking about CS4, yes, you can develop an exporter plug-in that uses CUDA for accelerated encoding. Elemental Technologies did this with their Elemental Accelerator. And Divide Frame built a CUDA-accelerated importer with their GPU Decoder.
Copy link to clipboard
Copied
Hi Zac,
can you elaborate now?
Copy link to clipboard
Copied
Our CUDA support in our accelerated renderer has been built to play well with others. We have considered providing a suite to allow exporters, when running in the same process as our accelerated render (i.e. when doing a direct export rather than queue up an encode to Adobe Media Encoder), to be able to access the rendered CUDA frames directly on the card, to avoid readback to memory. If you're interested in something like this, contact me offline using my email in the PPro SDK for the latest status.