After Effects interactive performance, cores vs clock speed

[Aaron Kondziela]

I'm building a new workstation, on which I'll be editing and VFXing with PP and AE, working with R3Ds and some P2 stuff. I don't have a Red Rocket just yet (but it's in the plan). My primary concern is After Effects performance. The hardware choice I'm considering is a single Xeon, with very fast clock, or dual Xeons, with slower clocks. (Yeah I know bang per buck, and overclock, etc.. but Xeon it is - need ECC ram and have no OC plans)

I would like it to feel as responsive as it can while working. Is there an advantage for interactivity to having many more cores to handle multiple threads? Or would clock speed have the greatest impact on it, using threads only for the background rendering processes? I am guessing for interactive user experience, clock may be the more important, as AE uses multiple instances to run backround renders as separate threads. Of course, when render time comes around, I know that more cores will help chew through the frames (assuming all else in the system is optimized and there are no bottlenecks).

I'm particularly interested in anyone who's worked with (or coded!) CS6, since it may be different from what I can measure in 5.5. Does AE CS6 make use of many threads for the main application? Or is it more a single thread, with more cores primarily benefiting render time? Also, the R3D libraries that Adobe uses, do they multithread well and take advantage of many cores?

Thanks for any insights you can share!

 

[Ryan Schroeder]

You probably know this already, but in terms of rendering/RAM preview etc., you'll need the necessary RAM to use all the cores you end up choosing. It's recommended you have 2GB of available memory per core (or hyperthread) that you are rendering out. For instance, I run a Mac Pro with eight 2.4GHz cores. Ideally, to use all those cores, I should have (8 cores x 2 threads/core x 2GB memory) = 32GB memory. And that's just for AE to use - your system and other apps (especially if you run other Adobe apps simultaneously) can use up to 10GB, I've found. And 2GB is recommended - 3GB per render core is ideal! In this particular machine, I only have 26GB memory, and I still can't seem to use all my cores because of what I've mentioned before.

So, unless you can afford a ton of memory, I'd almost lean towards a way high clock speed and use all the cores you will be getting by filling out the RAM appropriately.

 

[PatrickFaith]

I would like it to feel as responsive as it can while working. Is there an advantage for interactivity to having many more cores to handle multiple threads? Or would clock speed have the greatest impact on it, using threads only for the background rendering processes? I am guessing for interactive user experience, clock may be the more important, as AE uses multiple instances to run backround renders as separate threads. Of course, when render time comes around, I know that more cores will help chew through the frames (assuming all else in the system is optimized and there are no bottlenecks).

When running AE on windows, i am using 12 threads and it looks like it scales nicely (u need atleast 2 gig per thread). What I can see from cs6, make sure you get a lot of fast SSD (the new caching approach really looks like a time saver to me ... but i haven't tested it yet).

 

[Aaron Kondziela]

Ryan - This is something I've considered, as with a dual 8-core xeon system, I'd need to toss 64GB of ram in there just to hit the minimum of 2 per thread. That's a lot of ram! But, ram is cheap these days, so I'm going to pack it in like a champ =)

Patrick - The 12 threads, are you using those for rendering final output, rendering with multiple background processes for previews, or just during use of the app in general? And I had the same idea with the SSD and the new caching. Thinking of one SSD for boot/apps, another larger one dedicated to global cache and other similar stuff, and an array for footage and projects and stuff. I'd love to go 100% SSD, but my goodness that'd get spendy! I need to do the math, and find out how much data the CPU arrangement can slurp up at full speed, and make sure the storage subsystems will have the necessary bandwidth. Really trying to eliminate the usual bottlenecks in this build. PCIe 3.0 for CUDA, also, if I can.

 

[Steve Sherrick]

Funny that this thread started today. The 16GB of RAM I have in my Mac just ain't cutting it working in After Effects or Avid FX. Got to max out my machine. Today was the tipping point.

 

[PatrickFaith]

Patrick - The 12 threads, are you using those for rendering final output, rendering with multiple background processes for previews, or just during use of the app in general? And I had the same idea with the SSD and the new caching. Thinking of one SSD for boot/apps, another larger one dedicated to global cache and other similar stuff, and an array for footage and projects and stuff. I'd love to go 100% SSD, but my goodness that'd get spendy! I need to do the math, and find out how much data the CPU arrangement can slurp up at full speed, and make sure the storage subsystems will have the necessary bandwidth. Really trying to eliminate the usual bottlenecks in this build. PCIe 3.0 for CUDA, also, if I can.

On the exporting the finished composite of AE to the output format (i.e. h264, png, exr) .... i have noticed the cpu balance changes a lot depending on the export options. Not sure why. For normal AE compositing, is where i noticed the thread distribution is real good. I do have two nvidea cards, so on some composites i don't see a heavy loading, but i think in a few cases a lot of the work is going out to cuda. I do think the load balancing to cude will be better with cs6, so I'm going to lean to more ssd and probably going to get multiple tesla cards. I do have the highest end intel processors (that can overclock), but i normally have the overclocking turned off ... just seems like my machine runs better(if i was some uber pc guy... understanding all those bios options ... i'm sure i'd have a different answer).

 

[Aaron Kondziela]

Patrick - thanks for the info, although you've probably contributed to the more rapid drain of my bank account now, I'll be looking at more cores... I used to do the overclocking and tuning thing a lot in the past, and I'm good with it. My priorities have changed though, and I go for stability over tweaked performance these days. Hence, no OC, and ECC ram. Last thing any of us needs is a crash in the middle of an important project!

I'll post my findings for everyone once I select the hardware, and run benchmarks when the machine is done. Throw a bunch of 4K awesomeness at it and see what happens!