Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
FCPX: AMD vs NVIDIA
- Thread starter CuamckuyKot
- Start date
- Sort by reaction score
I've got a 3.2Ghz 6-Core and a Mac Nvidia 680 2Gb, fastest it seemed to do was 68 seconds.
Didn't matter if the project was on my spinning disk ZFS array or on a SATA-III SSD, ~68 seconds.
Running OS X 10.11.3, FCPX version 10.2.3, latest Nvidia web driver (as of posting), 16Gb RAM.
Didn't matter if the project was on my spinning disk ZFS array or on a SATA-III SSD, ~68 seconds.
Running OS X 10.11.3, FCPX version 10.2.3, latest Nvidia web driver (as of posting), 16Gb RAM.
Last edited:
Pls specify which OS X, FCP version and Nvidia driver, x GB RAM etc you are using. There are huge differences.
Last edited:
The original 2013 graph shows even the Radeon 5770 with a sizable lead over Iris and HD 4000. I'm curious if that still plays out now that we have Metal accelerating the latter GPUs...
[doublepost=1456874376][/doublepost]Just tested my system:
10.11.3 / FCP 10.2.3 / Mac Pro 5,1 6-core / 48 GB RAM
45 secs : Radeon 5770 (OEM/EFI model)
48 secs : Radeon 6850 (PC card)
Odd that the 5770 would beat out the 6850, but perhaps that's due to the latter being a PC card.
[doublepost=1456874376][/doublepost]Just tested my system:
10.11.3 / FCP 10.2.3 / Mac Pro 5,1 6-core / 48 GB RAM
45 secs : Radeon 5770 (OEM/EFI model)
48 secs : Radeon 6850 (PC card)
Odd that the 5770 would beat out the 6850, but perhaps that's due to the latter being a PC card.
Last edited:
Avid doesn't use GPU acceleration on Mac. I don't see much diff between my Nvidia that shopped with my 2009 and the Radeon R280X I installed.
And it does use very little on PC too - though Avid has always recommended Nvidia on PC.
I do. I ran the bench as instructed, using both Yosemite and El Capitan (on the same computer, but two different SSDs):
GTX 980
5,1 W3690 16GB
FCPX 10.2.3
OS X 10.10.5
Plextor M3
Nvidia 346.02.03f05
52.33 seconds
OS X 10.11.3
Samsung 830
Nvidia 346.03.05f02
27.20 seconds
El Cap is definitely a vast improvement.
That 780M time is also beating my 980 time by a significant margin.
Last edited:
My son is an editor at a big post house in NYC. Previously they used the cMP, but went to the nNP.
I asked him what they were using, and how they liked it. They use Avid.
He replied, "The work computers are 3.5 GHz 6-Core Intel Xeon E5 with 16 GB 1867 MHz DDR3 running 10.9.4 with a matrox mxo2 mini. We're running Media Composer 8.0. The system is OK."
I asked about video, and he said they use D500.
Not exactly a ringing endorsement, but in his world, predictibility and stability are key. Notice that they are still on 10.9, probably because of MC 8.
Thank you so much for responding. I guess nMP has some use. Apparently the nMP may arrive this summer! Thanks
Hopefully it will run better. Is your son planning to upgrade a new Mac Pro if it comes out this summer?
I'll be visiting there in a few weeks. If I go to the shop, I'll see if anybody knows. I'd guess that he wouldn't know anything about purchasing plans, because athough he's one of the senior editors, it's a big shop with a technical staff and all that. It's not a decision that he'd have any part of. He, and probably most of the other editors, are uninterested in what they cut with, provided it does the job for them.
That has always amused me (he's been in the business for years) because I, the amateur, am much more interested in the technical aspects of editing than he ever has been. When he was running his own small video production business, a long time ago, I built their editing machines (Supermicro boards, always). Then they got a G5, then he went to work for other people, and that was the end of my technical input into what he does.
Just want to confirm your thinking that a single 7950 completes the benchmark in about 29-30 seconds (2009 cMP single X5670 24GB RAM 10.11.4).
Interestingly enough though I had to plug my monitor into the 7950 (which is in Slot 1) to get FCPX to use it. Otherwise it was using my GTX 750Ti (benchmark at about 59-60 seconds with the latest Nvidia web driver). I think you said in a different thread that FCPX will use a GPU even if it's not plugged in to a monitor. Well maybe that is only the case if it sees two identical GPUs (like in the nMP)?
I also tried running the test with both cards plugged in to a monitor and still got the same result as with the 7950 only plugged in (~30 seconds). I haven't confirmed by actually pulling one card out and re-running the benchmark. But it seems like FCPX on my machine is only using 1 GPU at a time. Since the 7950 is in slot 1 if it is plugged in then FCPX uses it. If 7950 is not connected then the 750Ti in Slot 2 is used. If both cards are connected it still goes with the 7950 in Slot 1.
Obviously I should test a lot more before making that assumption. But it just seems to be a preliminary guess. Also, my 7950 was in PC UEFI mode (my 750 Ti has an EFI on it so I use it as my primary video card). I'm planning to keep the 7950 installed only for FCPX duties.
Thanks for the sharing.
My 2nd 7950 do not connect to any monitor. But not sure it if require same GPU, or same brand GPU, or a GPU that has native OSX support, etc.
Just to throw my results in. I'll start by saying I am remoting in from my office computer to my home iMac w/ a 980TI eGPU.
I ran the test hit my stopwatch 3 seperate tests and My average is 28 seconds.
Again this is on an eGPU setup over Thunderbolt2.
I'm wondering if an AMD card in the eGPU setup would make much of a difference.
Also been looking at the new Mac Pro 6core refurb unit if Apple doesn't announce or release an updated model in June.
I ran the test hit my stopwatch 3 seperate tests and My average is 28 seconds.
Again this is on an eGPU setup over Thunderbolt2.
I'm wondering if an AMD card in the eGPU setup would make much of a difference.
Also been looking at the new Mac Pro 6core refurb unit if Apple doesn't announce or release an updated model in June.
Interesting, so I assume there is a way to force OS X applications to use your 980ti eGPU as opposed to the build-in GPU?
Correct. The easiest way is I have an external monitor connected to the eGPU. Any app you start on the monitor connected to the eGPU auto uses the eGPU rather than the built in R9 395X card.
I work as an editor too, and will go along with your son's point of view. At the end of the day, what matters most is not the extra seconds one gains when exporting or rendering. What really matters is stability.
I just picked up a new Mac Pro yesterday and decided to revisit this thread and do another BruceX export and it DESTROYED what I already thought to be a fairly fast time from my 6700k iMac + GTX980TI eGPU setup.
8.9sec
XeonE5 6Core 3.5Ghz, 32GB 1866 RAM, Dual D500's.
I also didn't have my phone stopwatch nearby so this was clicking export, then clicking stopwatch on google, and stop right when quicktime popped to play the file. So it may even be slightly faster than this. I'll redo later with an actual stopwatch.
8.9sec
XeonE5 6Core 3.5Ghz, 32GB 1866 RAM, Dual D500's.
I also didn't have my phone stopwatch nearby so this was clicking export, then clicking stopwatch on google, and stop right when quicktime popped to play the file. So it may even be slightly faster than this. I'll redo later with an actual stopwatch.
Is there any particular reason why you decided to purchase a Mac Pro so close to WWDC? Is there strong reasoning not to expect a new model at the developer's conference?
I had been putting it off and putting it off and decided now was as good of a time as any. If nothing new is even referenced at WWDC then I'm fine with keeping this machine as it is very nice.
I am still well within my 14day return period with Apple and WWDC is only 8 days out from date of purchase, so I could take the full price I paid any put towards the equivalent new model.
Exactly that's what I was thinking
[doublepost=1465177938][/doublepost]
[doublepost=1465177938][/doublepost]
It's iffy either side. If people need it now.. Then it is what is.
...which, if we look at the 2013 WWDC announcement and what followed - could mean that the new model will ship on 31 December 2016, and arrive mid-January 2017.
If they even announced anything in the next week with a firm release date I'll go return this one and put the cash against the new model and make due until then, if they don't announce then I'm not that worried about it.
This machine will easily take me through several more years and it's AppleCared until 2019 should anything happen.
If I'm always waiting until the next thing is announced I'd almost never buy anything and if I had buyers remorse that bad then I probably shouldn't have bought it in the first place.
Just the way I am.
If the 5770 is an original Apple model, it's running at 5 GT/sec, if the PC card did not have the resistor mod it's running at 2.4 GT/sec which would explain the discrepancy.