At what point is it necessary to go from 64GB of RAM to 128GB?
I am aware at this point that 64GB will cover most casual motion graphics artists needs, but where exactly does the line lie where a creator should really be looking to go from 64GB to 128 GB? What would a project look like that would utilize 128GB? What would the rest of the system look like where 64GB is the bottleneck?
FWIW I am running on 32GB of DDR3 now, but as I look around and am building a "dream system" I'm finding most of the people talking about this issue are gamers, and where I find people talking about design uses they all seem to assume that you'll know when you need it, which is fine but I'd still like to know.
I've been looking at a Z390 Aorus Pro Wifi, with a 19-9900k and 64GB of RAM, but then I also love the thought of a Designare X299 with an i9-9980XE and 128GB of RAM.
And then there is the BIOS update that will allow the Z390 Aorus Pro Wifi to run 32GB sticks theoretically opening up the door to a i9-9900k (which is apparently the fastest AE rendering CPU) together with 128GB.
Anyway, enough rambling, hopefully you'll get where I'm going with this, anyone care to help illuminate this for me?
In my experience it's the bigger comp sizes (4k plus) that really need all that RAM. To be able to hold the frame in memory and all the assets that need to be integrate (but not necessarily all the layers in the comp since some of those will get shunted to the disk cache). There are times when depending on the effect and frame size, you just can't render on 32GB.
AE is both really good and bad at making sure no matter how much ram you have it can usually work with the disk to render - it can just take much longer when RAM is used up too quick. The bad is that with lots of ram you have lots of ram to get clogged. The memory purge has gotten much better over the years but Adobe has never been known for their elegant memory management. Even though we have a "clear all caches" button now, I find that even if I use it, then quit AE, I still can't run some other intensive program right after. There's still a hairball in the drain somewhere.
Finally, more RAM means more frames can be cached and played back in real time without rendering out to a file.
But it is a case of diminishing returns. Doubling the ram (after 16gb) doesn't make AE twice as marvelous, and the difference gets harder to detect the closer you get to 256GB. That being said, more RAM means more left over for something else running at the same time - say switching back and forth to C4D to a project that's linked and you're tweaking both AE and C4D.
Fantastic reply, I don't think I could've asked for anything more.
When I was running 16GB I was absolutely having more issues than I am now, I had one render that as you said in your first paragraph, just wouldn't render. It kept shutting down Media Encoder hard every time I tried. I ended up having to shut off all the other apps, and shut the computer down, and then I finally got it to render.
Even then it took 20 hours to finish, it was a 60 second clip. This was the point where I discovered rendering project comps/pre-comps individually with an alpha channel whenever possible instead of building a massive composition with a million pre--comps and layers upon layers of stacked up effects. Obviously in some cases you'll need a few pre-comps to be able to interact together under an adjustment layer or something. But I found that most times I can just render out things on their own and then build the final composition out of a bunch of clips with alpha channels.
Worth noting that clip I referred to was 3976x2840, I also try to keep everything I do in the realm of 1080p these days.
Now it just boils down to the Z390 or the X299 chipset and how 128GB (4x32GB) compares to 128GB (8x16GB). On the surface it looks (to my inexperienced mind) that combining the faster rendering capability of the 19-9900k with 128GB would be ideal, but does spreading the RAM out amongst 8 slots make the X299 configuration more ideal for AE and other motion graphics purposes?
Now I could be wrong here, because I'm not a bus architect, but I don't think spreading the ram out matters. Its all on the same bus. I would think fewer chips in fewer slots would allow for expansion in the future. (unless I missed where one chipset offers lots of bays and the other doesn't). But bigger chips are more expensive than smaller ones. At least until you throw the small ones out completely when you need those slots back and more Ram.
The big thing here is how else do you want to use the machine. And this comes down to cores. For 3D apps you want to allow enough ram so that each core can have enough RAM to take its part of the render. Watching C4D cook is really cool when you have a whack of cores. 12, 16 or 32 little renders going on simultaneously all over the image, each one being done on a separate core. But its only efficient if you have enough ram to support those cores.
For AE, the rule of thumb has always been: Faster CPU clock speed at the sacrifice of core numbers. Its common when you get a lot of cores in a chip that you don't get the top speed available in that chip set. I'm not sure if this is an issue with dumping heat, or the way the cores are optolithographied means the accuracies aren't as good so you have to run at a slower speed, or its just that the info from all those cores spitting things out all at once gangs up on the bus and you just have to slow it all down to get everyone on board. But based on experience and (what might now be old thinking) its better to have a faster overall speed than a slower speed and many cores.
Keep in mind too that much of what after effects does can be shunted to the video card. So lots of ram there too is best because compared to a video card's speed, your processor is standing still. Put the render on the fastest proc.
There's lots of discussions on the boards about what vid card is best. Some play well with after effects and some don't. Again this is a confusing mess because the pro cards (the Quadros) seem on the surface to be slower but in practice they tend to produce faster results, or should I say fa$ter. Whereas the GTX and Titanium cards: the 900s and 1000 series, look faster, with their gobs of cuda cores but can actually be slower, and worse, incompatible - I've got a 970 and 980 that work great but won't let me play smooth video out of cached AE. We've never solved it.
Then there's the drive. Ae must have fast drives as it's always shunting stuff to cache that it can't fit in ram - and it can never never never fit it all in ram. So consider an SSD somewhere in the mix. Some say use it for your disk cache some say use it for the OS and App since there are a lot of instructions getting passed around. I'm not sure, but keep in mind the SSD's do have a finite number of read write cycles. We do a lot of 3D here and I'd hate to think how many times Maya or C4D writes to the drive every render - billions I would guess. It would trash an SSD in short order I would imagine. So we just use 7200's and 10k drives while we wait for SSD's to mature (and get cheaper).
Another fantastic reply.
I also worry about SSD write capacity, I have every part of my AE projects on a different drive, which seems to have sorted a lot of the issues I was dealing with before when I was working off of one SSD, having moved from the dual GTX-770 in sli setup that this (second hand) computer came with the a GTX-1060 certainly didn't hurt either. I have my OS, Software, and project folders on on 500GB SSD, I have a dedicated current projects assets drive, also 500GB SSD, a dedicated current projects render output drive, 500GB, and a 250GB dedicated cache drive for my Creative Cloud apps only. Plus a 1TB 7200 HDD and a 4TB WD external for slow, and slower bulk storage.
Right now my workload isn't producing the kinds of read/write cycles that yours is but I have still had to empty my assets and output drives of older projects and put them in the external a couple times from hitting disk capacity. So my use is probably a bit higher than the average user.