-
Upgraded to Quadro K5000 Mac, and almost no speed improvement in AE CS6 or AE CC 2014?
-
Wow, that’s a lot of reflection-y surfaces!
If I had time, I could try to build something like this in C4D to test it, but it’s a bit hard to see what’s going on. I can’t even make out the shape of some of the geometry. Is there a download of that somewhere I could use as a reference?
– The Great Szalam
(The ‘Great’ stands for ‘Not So Great, in fact, Extremely Humble’)No trees were harmed in the creation of this message, but several thousand electrons were mildly inconvenienced.
-
-
Thanks Eric for the Octane link!! Duh me, someone told me about Octane a couple of
weeks ago, but I’d forgotten it was a plugin for C4D, though now I remember why: the
benchmark scene res (HD) is larger than the max pixel size of the plugin demo (1000×600)
so I can’t use the demo to test how C4D/CUDA would compare.Michael Szalapski writes:
> Wow, that’s a lot of reflection-y surfaces!It was designed to be a CUDA killer. 😀 One frame to use as a performance benchmark,
the whole animation as a stability test (kept seeing comments from people asking for
something that can really hammer a system to check that long duration renders won’t
knock their system over).With two Titans at max detail, the full 4 second animation would take maybe 2 days
to compute. I’ll be writing a page for it soon for my site.> If I had time, I could try to build something like this in C4D to test it, but it’s
> a bit hard to see what’s going on. I can’t even make out the shape of some of the
> geometry. Is there a download of that somewhere I could use as a reference?Here’s the aep:
https://www.sgidepot.co.uk/misc/cuda.101.zip
If you go through the usual RenderQueue setup, it’s preset to process just frame 96
at medium res/detail (or should be), ie. 16bpc, 8 levels (RayTrace3D options), HD 25fps.
Max detail is 32bpc, 10 levels, 50fps.Ian.
PS. I didn’t create the aep. Friend of mine did all the hard graft.
SGI Guru
-
Thank you for all the information! It’s been VERY HELPFUL!
I tried changing the “Render Multiple Frames” settings and that made significant improvement after some trial and error. My first attempt, I left it at 2GB per core, and allocated 12 cores to AE. This came back with about the same times, and took about a minute to initialize at start. When I gave it 18 cores at 6GB per core, it ran the fastest I have ever seen it with my test comp. Final time 2:43, which still sounds bad, but usually whenever the text is changed, it’s first render out of the change is around 5 minutes. The only time prior it ran that fast is when it re-ran an unchanged comp, which I assume was largely cached, and that’s why it ran faster?
Pulling up activity monitor during render showed about 12 cores running at 65 – 75%. Thank you so much for the help today. It’s already faster than it’s been, thanks to you folks!
-
The amount of ram per core/thread really is comp specific. Depending on the frame resolution, layers/complexity, and FX this fluctuates greatly. What works with ideally with 6GB per thread may not work so efficiently on another comp. This also can change the amount of cores used by the multiframe rendering versus the AE application and other processes currently provisioned by AE. As a general rule though the greater the frame res and amount of layers, the more ram per core you need.
Eric-ADK
Tech Manager
support@adkvideoediting.com -
This is why I’ve been hoping that the X99 chipset will up the max RAM to
128GB or more, but I fear it won’t.Ian.
SGI Guru
-
There are several reasons I disagree but primary among them are:
1. The overhead of launching and managing the several render processes (32 if maxed on my machine) generally results in diminishing returns.
2. AE is terrible at memory management. I prefer to limit the number of CPUs so that my computer doesn’t attempt to use all 128GB of ram across 32 render processes.
The Render Multiple Frames Simultaneously feature is a quirky workaround for not having a fully multithreaded renderer that sometimes causes more trouble than it’s worth.
Often I need 8GB (sometimes 16GB) of RAM dedicated for each background render process so generally the number of CPUs that can be used is limited by that, but even if it weren’t I hard cap the number of CPUs that can be used to half of my logical cores (32/2 = 16).
Darby Edelen
-
AE uses threads for more than just the Background Render processes of the multi-frame rendering. Leaving those cores available for AE doesn’t mean you need to use all of them for multiframe rendering. That is decided by how much ram the system has overall and how much you set for each thread. What is set for each core/thread is more decided by what is required for that comp than necessarily required to run all cpu cores/threads for the background render. Often running enough ram per core leaves more than enough cpu cores/threads available for the other processes AE uses. The point however is to let AE and the OS have whatever CPU cores available based on what is scheduled and moderated by the OS. The OS is far more efficient moderating that on average than we are since processing conditions constantly change in these applications. Its far easier to go back and adjust for a special case that is performing less than ideal than lose processing efficiency 100% of the time.
Eric-ADK
Tech Manager
support@adkvideoediting.com
Reply to this Discussion! Login or Sign Up