-
CPU Upgrade doubles geekbench but has almost no effect on AE
hi all
I’ve just upgraded my MacPro from 2x E5520 to 2x X5680.
The geekbench and cinebench scores have both doubled, which was very exciting, but when I went to test my after effects comp it has only the most minimal speed increase.
Have I missed something ? Why would the benchmarks be so much better than AE.
My test comp consists of 128 masked layers and uses no plugins or filters.
I’ve gone from a render time of 1:32 to 1:16.
any suggestions greatly appreciated.../daniel
ps. I should also mention the OS and Apps are on SSD, both source and destination drives are superfast RAID and the machine has 32GB RAM (1066)
-
7 Replies
-
The processing of some things is single-threaded, so adding more CPU cores doesn’t speed them up. There are many multithreaded effects, so you will see a speedup when usong them. Also, you’ll see improvement when using Render Multiple Frames Simultaneously multiprocessing.
———————————————————————————————————
Todd Kopriva, Adobe Systems Incorporated
After Effects quality engineering
After Effects team blog
——————————————————————————————————— -
I’m NO EXPERT at all…but that was a 13% gain. I don’t think thats minimal for a 40% increase in clock speed.
You may see things rendering faster if you go for multprocessing. But unfortunatly I believe it won’t ever be close to half the time even though you have close to double the processing power, there are several other bottlenecks and software implementations needed for that.
-
so, I don’t have any effects whatsoever it’s just a single video file that has been cut up (masked) and placed around the frame (128 layers), no expressions, no blending modes nothing. I really thought the clock speed increase would result in a linearly proportional render time decrease
I hadn’t been expecting render times to be halved like in C4D as i don’t have the RAM to max the CPUs but I had expected a little more than 13%.
Interestingly, I’ve found launching separate instances of AE non-MP actually renders remarkably quicker than letting AE handle it through MP. My only concern is the disk access from 6 separate instances…
and finally why is the OSX version of AE so much slower than the Windows version. I’ve asked this before but i am still amazed by the following statistic:
MacPro 2x 3.3GHz , 32GB RAM, Super fast raids (>$15k) = render time of 1:15
Windows i7 960, 12GB RAM, internal 7200 drives (<$2k) = render time of 1:41 -
Finally why is the OSX version of AE so much slower than the Windows version. I’ve asked this before but i am still amazed by the following statistic:
MacPro 2x 3.3GHz , 32GB RAM, Super fast raids (>$15k) = render time of 1:15
Windows i7 960, 12GB RAM, internal 7200 drives (<$2k) = render time of 1:41—-
You’re comparing Apples to Pears. To compared AE in OSX vs Windows it has to be done on the same hardware. The i7 is MUCH faster for MHz on single-threaded operations than the MacPro or even current-gen Xeons.
-
[Erik Lindahl] “The i7 is MUCH faster for MHz on single-threaded operations”
hi Erik, that is very interesting information. at least I feel a little less harsh on my old mac pro. So, I’m thinking of selling the X5680s and just getting W5590s (keep the GHz but lose the cores) and spending the difference on RAM, i.e 96GB of 1333, that way I figure I can allot 6GB per process and have 12 cores for MP with fair amount of headroom. Does that make sense ?
-
[Daniel Crooks] “so, I don’t have any effects whatsoever it’s just a single video file that has been cut up (masked) and placed around the frame (128 layers), no expressions, no blending modes nothing. I really thought the clock speed increase would result in a linearly proportional render time decrease
I hadn’t been expecting render times to be halved like in C4D as i don’t have the RAM to max the CPUs but I had expected a little more than 13%.”
I might be wrong, but isn’t that taxing the hard drives more than the processors? The cpu is just applying the mask, but the hard drives are tasked with retrieving the file.
-
Yeah, could be I/O or single thread limitations.
Reply to this Discussion! Login or Sign Up