Yes, definitely an interesting case. Wonder how the dynamic of Apple & AMD writing drivers passionately together looks like.
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.
Dual 2 two RX 580 Final Cut Pro X
- Thread starter VillasManzanill
- Start date
- Sort by reaction score
Yes, definitely an interesting case. Wonder how the dynamic of Apple & AMD writing drivers passionately together looks like.
I can't remember the results, but it was something very minor.
https://forums.macrumors.com/thread...cs-cards-for-cmp.2025989/page-2#post-24244163
Based on my past post, I had 2x390x and 1x390 running, and I got 22 seconds on BruceX
https://forums.macrumors.com/threads/dual-r9-390x-8gb-or-dual-rx-480-8gb-for-fcpx.2014114/
Based on Chriz_r's post, he got 19 seconds. He also posted that his dual RX 480 = 19 seconds, Single RX 480 = 22 seconds
Check out my benchmark for 3 graphic cardshttps://forums.macrumors.com/attachments/screen-shot-2017-01-28-at-10-26-56-pm-png.685942/
Funny I get 45000+ in luxmark only with 2 x 1080TI in our Mac Pro. No good for FCPX, but still great for DaVinci.
Neat Video is very compute-intensive -- on any machine. It is configurable to use all GPU, all CPU or any combination of the two. If you open the Neat Video interface in FCPX and select Tools>Preferences>Performance>Optimize Settings you can run a built-in benchmark that tests all combinations of CPU cores and GPU for best performance. If you take the recommended best combination in theory that will provide the fastest performance -- but it will be slow no matter what, this only makes it somewhat less slow.
There are two phases of export: rendering and encoding. A better GPU might help rendering -- IF the effect is GPU-based. Many effects claim to be GPU-based but examination with iStat Menus ver. 6 shows they are using little GPU resources. That is a good tool to check GPU utilization: https://bjango.com/mac/istatmenus/
The other phase is encoding, which cannot be accelerated by traditional GPU methods; it is mainly a CPU-bound activity. However if encoding to H264 or H265 Intel CPUs with Quick Sync will be much faster. Unfortunately the Xeon CPUs used in Mac Pros don't have Quick Sync so they can be slower at exporting to H264 than a MacBook Pro.
do you happen to have a dual rx 580 8gb benchmark for bruceX?
thanks for all your info. its helpful.
No, I don't have that, but below are the BruceX results I've run recently. BruceX is sensitive to test procedure. It should be run with an unrendered timeline (which means auto-render disabled), ProRes 422 output, pre-load QT Player, disable Spotlight indexing, disable Time Machine, recreate library each pass, and restart FCPX each pass, make at least three runs and take the average of group.
BruceX test, FCPX 10.3.4, mac OS 10.12.6:
2016 MBP i7 RP 460: 36.2 sec
2015 iMac 27 i7 M395X: 26.9 sec
2013 12-core Mac Pro, dual D700: 17.0 sec
2017 iMac 27 i7 RP 580: 15.8 sec
The author of BruceX initially said H264 then updated this to ProRes 422. That makes sense because an H264 export can become CPU-limited during encoding, which will vary greatly whether the computer has Quick Sync. Regarding other ProRes variants of I'm not sure how much difference it makes but it makes sense for everyone comparing results to use the same parameters and procedures: http://www.fcp.co/forum/hardware/18250-brucex-try-this-new-final-cut-pro-x-benchmark
[doublepost=1510650490][/doublepost]I have dual RX580, my bruceX test is: 8,74sec (background render off). It's very fast!
My system: 4,2 intel Core i7, 64 GB 3000 MHz DDR4 RAM, SSD, dual RX 580 8 GB, running High Sierra! (Hackingtosh ;-)
Flashed 2009 cMP, Dual Xeon x5690, Dual XFX RX480 8GB OC (with the itsage's Kext edits) BruceX: just under 12 seconds.
Thanks!
That’s what I been looking for.
Now it will be awes
those are amazing numbers.
Now let’s see if a real Mac Pro owner post his results for me to have a better picture of the performance in our old machines. Thanks again!
Just in case this helps anyone, I recently bought an 8gb RX580 for use in a 5,1 mac tower. It went in with no problems at all & and booted fine under OS10.12.6
However, i purchased it hoping it would help with FCPX & it offered absolutely no improvement from the older 3gb Radeon HD7950 I had in before. Like literally zero difference in either background timeline render or export. A worthwhile experiment, but I'm returning it.
I'm sure it would make a huge difference if you wanted to play games judging by the 3d benchmarks , but not so useful for FCPX unless you had a really old card to begin with
However, i purchased it hoping it would help with FCPX & it offered absolutely no improvement from the older 3gb Radeon HD7950 I had in before. Like literally zero difference in either background timeline render or export. A worthwhile experiment, but I'm returning it.
I'm sure it would make a huge difference if you wanted to play games judging by the 3d benchmarks , but not so useful for FCPX unless you had a really old card to begin with
Export is a CPU task, upgrade the GPU of course won't help (unless Apple unlock the GPU hardware video encoding ability).
Background rendering may be improved, but seems very depends on OS version (GPU driver version).
From memory, members here reported that a single RX580 can finish BruceX in ~15s (quite a few months ago), but a more recent report (after the M&S security patch) make the RX580 require ~35s to finish BruceX.
For 15s, that's about 2x 7950 performance (I've done this before).
For 35s, that's about a single 7950's performance. So, The RX580 may not able to do any better unless the job require more than 3GB of VRAM.
Run High Sierra and the latest FCPX, and then try again Things are way different in HS.
I can attest to this. FCPX 10.4 and latest High Sierra beta with a MSI RX 580 8GB "GamingX" yields a 19-20s BruceX. 10.13.x is a must.
[doublepost=1517338651][/doublepost]itdk, I also noticed that you upgraded from the 6-core to 8-core. Are you seeing faster export times?
Can you please try it again, as I see BruceX benchmark in High Sierra 10.3.3 is slower. What OS are you running?
FYI
https://forums.macrumors.com/thread...in-high-sierra-10-13-3.2103211/#post-25755445
Hi there, thanks for that info. I'm afraid I sent it back as I run quite a complicated rig with a lot of Pro Tools plugins too, & HS isn't 100% ready for primetime yet. Worth knowing though for when the price of the cards drops in 6 months. I might have another go, if I've not given cup & just bought a new iMac. cheers
Hi everybody.
I used to edit animatics for cartoon shows in FCPX. Animatics are build of many stills — jpeg images. Usually there are 2000+ images in 12 minutes + some 1080p videos, simple effects.
I have Mac Pro (Mid 2012)
2 x 3,46 GHz 6-Core Intel Xeon
96 ГБ 1333 MHz DDR3
Radeon RX 580 8gb Pulse (runs natively)
FCPX 10.4 high sierra 10.13.3
BruceX is 19 seconds.
FCPX and all library assets are stored at Kingston SHPM2280P2/480G (1200mb/sec Read-write).
But i have micro freezes while editing FCPX timeline. Micro pause after copy/pasting clip, micro freeze after triming and so on. In general, rendering performance are cool, but live timeline editing is poor. Turning off BG render does not change anything. FCPX does not make proxies for still images, but they are already light weight.
So i noticed that FCPX uses ALL graphic card memory (8gb) while working:
In this case i wonder, will second rx 580 pulse help my situation? Can FCPX use 16 gb of graphic memory from two cards? Does Macos natively support dual pcie gpus?
I used to edit animatics for cartoon shows in FCPX. Animatics are build of many stills — jpeg images. Usually there are 2000+ images in 12 minutes + some 1080p videos, simple effects.
I have Mac Pro (Mid 2012)
2 x 3,46 GHz 6-Core Intel Xeon
96 ГБ 1333 MHz DDR3
Radeon RX 580 8gb Pulse (runs natively)
FCPX 10.4 high sierra 10.13.3
BruceX is 19 seconds.
FCPX and all library assets are stored at Kingston SHPM2280P2/480G (1200mb/sec Read-write).
But i have micro freezes while editing FCPX timeline. Micro pause after copy/pasting clip, micro freeze after triming and so on. In general, rendering performance are cool, but live timeline editing is poor. Turning off BG render does not change anything. FCPX does not make proxies for still images, but they are already light weight.
So i noticed that FCPX uses ALL graphic card memory (8gb) while working:
In this case i wonder, will second rx 580 pulse help my situation? Can FCPX use 16 gb of graphic memory from two cards? Does Macos natively support dual pcie gpus?
it will use 2 cards natively like the nMP but i dont know if the memory is Independence, i assume not.
vega founders edition?
it may be the images are not as light as you hope?
dose istats let you see non apple cards vram use? iv not used it in ages last time i looked with my pc gtx770 no vram was shown
vega founders edition?
it may be the images are not as light as you hope?
dose istats let you see non apple cards vram use? iv not used it in ages last time i looked with my pc gtx770 no vram was shown
Hi everybody.
I used to edit animatics for cartoon shows in FCPX. Animatics are build of many stills — jpeg images. Usually there are 2000+ images in 12 minutes + some 1080p videos, simple effects.
I have Mac Pro (Mid 2012)
2 x 3,46 GHz 6-Core Intel Xeon
96 ГБ 1333 MHz DDR3
Radeon RX 580 8gb Pulse (runs natively)
FCPX 10.4 high sierra 10.13.3
BruceX is 19 seconds.
FCPX and all library assets are stored at Kingston SHPM2280P2/480G (1200mb/sec Read-write).
But i have micro freezes while editing FCPX timeline. Micro pause after copy/pasting clip, micro freeze after triming and so on. In general, rendering performance are cool, but live timeline editing is poor. Turning off BG render does not change anything. FCPX does not make proxies for still images, but they are already light weight.
So i noticed that FCPX uses ALL graphic card memory (8gb) while working:
In this case i wonder, will second rx 580 pulse help my situation? Can FCPX use 16 gb of graphic memory from two cards? Does Macos natively support dual pcie gpus?
I ran 2x HD7950 before. In FCPX, the VRAM should be mirrored. So, effectively same as single 8GB.
My experience so far. In FCPX, for time line, Nvidia GPU do better. For rendering, AMD GPU do better.
If you don't really need HiDPI, turn that off may significantly reduce VRAM usage.
[doublepost=1519821611][/doublepost]
Yes, slamjack's screen capture is actually the VRAM usage history from iStat.
Does Nvidia GPU use Metal in macOS? I thought that FCPX is optimised for AMD in general.
With Adobe video applications and NVIDIA (currently GTX 1080 FE), I am able to use CUDA, OpenCL, Metal and "Software Only" for GPU acceleration when it is available in an application. Premiere Pro particularly takes advantage of GPU with many effects. Other applications have less of a benefit. Not everything with Adobe is GPU accelerated. More applications should take advantage. Effects are added across applications with each update, however.
Adobe does have some implementation/optimization issues with GPU acceleration with Adobe Media Encoder CC 2018. Encodes from PPro & AE that are not "Dynamically Linked" do not benefit from GPU acceleration for compressions. This SHOULD be addressed by Adobe at some point, but has not yet. I've been dealing with bug reports and their video managers about this issue (along with NVIDIA support) for quite some time.
There are not nearly enough compression applications that benefit from GPU in macOS for encoding. Even Apple Compressor does not really utilize. Adobe blames Apple and NVIDIA for CUDA implementation restrictions, but offers no legitimate excuse or rationale for poor OpenCL or Metal support in AME. NVIDIA support says AME is not setup (programmed) to utilize properly.
Telestream has decided to basically move everything to cloud/server based vs. local for GPU accelerated compressions. Lightspeed Server is the only current application. Episode Pro is no longer being made or supported. (Would say the majority cannot afford these types of setups.)
Sorenson Squeeze does not recommend GPU acceleration and offers no benefits aside from CUDA, however they do not recommend using it (poor quality issues). The last time I tested, there was no support for OpenCL or Metal.
Would guess Adobe will start to support cloud encoding and rendering in the semi-near future. What it will take to get to that point and if media can remain local remains to be seen.