Dual R9 390x 8GB or Dual RX 480 8GB for FCPX?

Discussion in 'Mac Pro' started by chriz_r, Nov 10, 2016.

  1. chriz_r, Nov 10, 2016
    Last edited: Nov 27, 2016

    chriz_r macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #1
    If powering any of the dual cards isn't an issue (planning to externally power them,) which pair would be better suited for fcpx? With prices of used 390x going pretty low around my area, I can buy two of them cheaper than what i've bought 2 rx 480 8GB for. Im currently running dual rx 480 on my 5,1 dual 2.4ghz, and my brucex benchmark is only netting me 19 seconds. I've read of people getting single digits with two 390x but mostly are from hackintosh builds so i was wondering if any of you actually rocking two 390x's can chime in as far as the performance you're getting with them for video editing and other gpu intensive tasks?

    I know i'm getting pretty terrible geekbench scores on the cpu side but thats another topic/issue i'd have to address somehow...

    Thanks again in advance to all of you!

    UPDATE:

    head over to post 53 for the tests that i did:

    Dual R9 390x 8GB or Dual RX 480 8GB for FCPX?
     
  2. h9826790 macrumors G4

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #2
    Even though I believe that their performance may be similar in FCPX, I will go for the 480 (assume you know how to make it work) because it's a newer card, possible have better support in short future, and it draw much less power.
     
  3. chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #3
    Yes unfortunately i did my minimal research and just got really excited and forked out the money for two 480's and now currently have them on my cMP working with metal support and all that jazz, but sadly as far as fcpx goes, i'm only getting about 5-10% utilization off of them during rendering on 4K projects... Im really looking forward to seeing better support for them in the near future but as I use these for work, Im debating if i should sell them and get two 390X instead, pending more input for you guys on that cards performance.

    The power isnt an issue for me as I am able to have an external power source for both cards, and the electric bill is tied in for work use.

    So pending full utilization and driver support, is the 390X actually close to the performance of the RX 480?
     
  4. h9826790 macrumors G4

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #4
    Be careful with the 390x, not many review / reports to study because the driver was quit buggy. Not sure how's the situation now, hopefully someone have that and running Sierra can tell you more details. If the Nano can work, the 480 can work, may be 390X can run flawlessly now.
     
  5. Fl0r!an macrumors 6502a

    Joined:
    Aug 14, 2007
    #5
    The Hawaii cards should be the fastest option in OS X. Drivers are still buggy though. It migth work, but it's more likely to have various bugs (e.g. multi-display issues, dead ports, screen flicker, ...). Hard to say without testing, it varies among models.
     
  6. 840quadra Moderator

    840quadra

    Staff Member

    Joined:
    Feb 1, 2005
    Location:
    Twin Cities Minnesota
    #6
    Be curious to see how support is fleshed out in the future too. I picked up a pair of R9 280x cards on Amazon 2nd hand, and have been happy with my Render speed. That said, It would be nice to see an upgrade path from them down the road!

    What are you seeing in the BruceX benchmark test?
     
  7. theitsage Suspended

    Joined:
    Aug 28, 2005
    #7
    @840quadra IMO R9 280x is the best all around GPU for the Mac Pro tower. You flash it to get Boot screen then remove R17 to get 5GT/s. The dual vBIOS helps in case there's a screw-up.

    When I had a dual RX 480 in my 5,1, BruceX was right around 20s (post #36). This was 3 months ago when the drivers were not fully functional.

    The RX 480 has potential. We won't be able to exploit it fully unless Apple releases a Mac with Polaris 10. I'm very hopeful though. Until then, I'm using my triple Crossfire RX 480 in an Alienware.
     
  8. chriz_r, Nov 11, 2016
    Last edited: Nov 11, 2016

    chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #8
    Yeah I've been using the 480's and they 'work' but not sure how much they really work. Using fcpx 10.3 / 10.2.3 and rendering footage with gaussian blurs and color corrections and other effects, im only getting a few fps, but as it seems to me, the gpu processors themselves are not even budging and only the memory gets used up. Is that the norm?

    Ps. The CPU Usage is because of the proxy media transcoding, but it looks to me also that im only really getting 75%/25% between physical cores and logical core usage. Again, i dont have a baseline to compare it to so im hoping others with fully working systems and on Mac Certified gpus will be so kind as to chime in with their experiences on utilization.


    Screen Shot 2016-11-11 at 11.22.39 AM.png

    Im on my way to Microcenter to pick up their last stock of gigabyte 390x, i think its for $240 so i'm gonna try it on one of my hackintosh and see what kind of performance I get. My plan for the 480 was to just buy them now and hope for the best with support but current situation dictates me to try other options. Im only using single UW 34" monitor so hopefully i dont have any issues. The only thing i noticed so far with my dual 480 setup is that when setting them up, sometimes only hdmi works and then sometimes only DP works. But when it boots and shows the login screen, i haven had any issues otherwise.

    To be honest, i tested the 480s and im getting confusing results. I tried single and dual setups, and im scrathing my head, hopefully i just have some settings to tweak on my machine.

    cMP 5,1
    Dual 2.4GHz
    64GB RAM
    SSD system drive

    Results:
    Dual RX 480 = 19 seconds
    Single RX 480 = 22 seconds

    im using visiontek reference 480s clocked at 1333MHz

    Screen Shot 2016-11-03 at 3.23.45 PM.png
    Screen Shot 2016-11-03 at 3.15.06 PM.png
    Screen Shot 2016-11-03 at 3.14.09 PM.png


    Above are the geekbench gpu scores for each cards at slots 1 and 2 respectively.

    Yeah I also have my gaming rig and thinking of just transferring the 480s over to them as well. Btw thanks theitsage for your detailed notes and tutorial regarding the rx 480 running on mac pros. I've always been hesitant to get in the mac pro systems and stuck to my buggy unreliable hackintoshes but since i've read your thread and saw how much better support has been on macOS was, it got me off the fence and in the span of a month i now have two mac pros LOL

    I really appreciate everyone's input and help, and all the time and dedication that this community has put towards educating and sharing their experiences with these machines. Please keep them coming, as there are a lot more people out there that are just silently reading and learning (like i was for the past years) who are very grateful for all of you!
     
  9. 840quadra Moderator

    840quadra

    Staff Member

    Joined:
    Feb 1, 2005
    Location:
    Twin Cities Minnesota
    #9
    Yeah I have moved away from a hackintosh too. I actually purchased my pro, for a friend near Chicago, but, he found one locally so I was able to keep it for myself. I haven't booted my i7 hack since, and haven't looked back. Quite honestly, since hackintosh systems are a grey area (legally) I feel better about having a (more or less) supported Mac.

    Just wish apple better supported Cuda, as I really like my nVidia gaming cards that are essentially rotting away. I think I will put them back to work with Folding@home soon though, as I like supplementing my heat with MY GPU furnace.
     
  10. theitsage Suspended

    Joined:
    Aug 28, 2005
    #10
    I'm in the Twin Cities as well. My Mac Pro and Alienware Area 51 will keep things toasty this winter. :D
    --- Post Merged, Nov 11, 2016 ---
    Slow it down dude! Let us know how the R9 380X works.
     
  11. koyoot macrumors 603

    koyoot

    Joined:
    Jun 5, 2012
    #11
    That is because they are working on Baffin DeviceID, the RX 460 deviceID, and are behaving like that GPU. So ultimately, you get not around 6 TFLOPs of compute power from single GPU but at best 4 TFLOPs from both GPUs.

    We have to wait for final Ellesmere deviceID support in Kexts.
     
  12. chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #12
    Haha oh trust me, this is me slowing down. I got these systems for me and my editors and I take them to the office. My motto has always been "time saved is money saved" and it always holds true for editing. I've built x99 rigs for editing with premiere but fcpx has always been the consensus with me and the other editors with our workflow so we keep coming back to it, but hack into she's just have too much unreliability and iMacs aren't cutting it anymore. The nMP to me is just not in the budget and I prefer getting more performance on the cheap when I can.
     
  13. SoyCapitanSoyCapitan

    Joined:
    Jul 4, 2015
    Location:
    Geneva
    #13
    Because Apple is using Polaris they might be capping performance to help battery life. Even their APIs are staying on lower performing old versions to save battery life. If the full potential of OpenCL or OpenGL were realised on Sierra then Apple could never make claims that they have better battery life than a Windows laptop. This is the gimmick they use to help market their products and reviewers never question it.
     
  14. chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #14
    Ohh that actually makes a lot of sense and I haven't thought of it that way. In my head I thought the system would detect the cards as such BUT allow them to run at their raw potential. Then again, I must live in lalaland where things just work :/ lol
    --- Post Merged, Nov 11, 2016 ---
    True! Battery life has always been the name of the game when it comes to portable computers and that's what's really holding back most of our electronics and their developers. You would think that apple would just use lesser powered components all the time for that but being that they are typically behind on hardware implementation, it makes more sense now to use and deploy new tech on their systems and just enable and utilize it better in the future with firmware and driver updates.

    It's all a facade, just like car companies boasting top of the line parts and power that sip on gas, then get caught using software to trick the testers >_<
     
  15. h9826790 macrumors G4

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #15
    Just want to make sure your "few FPS" is not the 2.0 FPS shows in iStat menu, that FPS has nothing to do with your rendering FPS.

    Also, iStat menu in Sierra is not work very well yet. I only get the GPU loading with the latest build (version 5.30 build 704), anything before that unable to show GPU usage in Sierra (true for my 7950). However, that doesn't mean that your GPU is not working.
     
  16. chriz_r, Nov 11, 2016
    Last edited: Nov 11, 2016

    chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #16
    Yup, i noticed that sometimes it actually goes up to 10 or 20 or something silly, and that screenshot was only for that specific time i guess.. :/ I dont have a lot of experience with iStats but it was the only tool i had to instantaneously check the current status of the GPU, but it probably wasn't that good like you said.
    --- Post Merged, Nov 11, 2016 ---
    I tried! Lol well i have a little update:


    So, I got a hold of two cards, Gigabyte r9 390x 8GB for $247 and and a Diamond R9 Nano 4GB for $220 at Microcenter.


    The Nano has a single 8 pin while the 390x uses a 8+6 pin combo.

    R9 NANO

    First I tried powering the nano directly with a 6pin to 8pin cable, but it’s a no go because the power cuts and the system shuts of. Unplugging the cMP IEC (power cable) for a few seconds and plugging it back it allowed to do a system reset so I can try again. If I didn’t do this, the computer would not power up no matter the amount of button presses I do. Now for the next test, I switched to a dual 6pin to 8 pin and it works fine, computer booted up and went to the login screen, however, this is when I noticed the artifacts. The screen looked like it was refreshing and each time I moved the mouse or dragged windows around, I can see it stutter and leave “ghost images.”

    To fix the issue, I went on and edited the info.plist for Baffin at /System/Library/Extensions/AMDRadeonX4100.kext and added the identifier 0x73001002 and rebooted the system. Worked like a charm, with everything including Metal supported.

    Now time for the tests, I used luxmark, geekbench, and bruce-x 5k. No other programs or windows are running during each of the tests. For Bruce-X, I deleted the event from the library and created a new one each time I test it, and background render is disabled.

    The results are as follows:


    Luxmark: 11662
    Geekbench: 77179
    Bruce-X 5K: 25 seconds average over 5 tests

    Screen Shot 2016-11-11 at 4.50.21 PM.png Screen Shot 2016-11-11 at 4.51.50 PM.png

    Not too bad, close to what im getting with my single rx 480 setup. I just actually ran luxmark on single 480 card and came out to be 11483 so very comparable. However, the Nano was a little bit faster on bruce x by itself vs. the 480.

    R9 390X

    Now on to the R9 390X. I took out the Nano and replace it with the card. Tired powering it with a 6pin and a 6-to-8pin plug directly from the mini ports from the cMP and it boots. No need to edit any kext at all, and I’m wondering if its making use of the previous kext edits I did (including the one for the 480) and I can do normal low gpu tasks like watching youtube and browsing forums etc. But as soon as I fire up fcpx to try and render or try luxmark or geekbench, the display shuts off but the mac pro itself does not power off.

    Switched over the power setup to utilize the 2 mini 6pin as one 8pin, and one 2xSata from the drive bays to 1x6pin. Booted the cMP and resumed the tests. This setup worked. Like the Nano, I fired up luxmark, geekbench, and bruce-x on fcpx. Now there is one thing that messed up here, I was able to do geekbench and brucex with no issues, but when I try luxmark, the system shuts off. So either my dual sata ports aren’t supporting the 6 pin or the 2x6pin-to-8pin wasn’t… either way I wasn’t gonna let it keep running that way, so I externally powered it with another psu. Now everything else works, and I continued with the tests.

    The results are as follows:

    Luxmark: 16494
    Geekbench: 127195
    Bruce-X 5K: 19 seconds average over 5 test

    Screen Shot 2016-11-11 at 5.03.46 PM.png Screen Shot 2016-11-11 at 3.05.03 PM.png

    Holy balls its as fast as my dual rx480 setup! Its faster on single luxmark and definitely blazing fast for the geekbench score. The brucex result was pretty much the same with my dual rx 480s, so im wondering now if I have the 2nd card then I’ll probably get a bit more performance.

    Fun time?

    So for shitz and giggles I mounted both the Nano and 390x to ports 1 and 2 and powered them using the external PSU I already had out and ran the benchmarks ive been doing. The geekbench scores were the same, individual luxmark gpu scores were the same, and the combined luxmark obviously changed. Here are the results:

    Luxmark combined: 28250
    Bruce-X 5K: 22 seconds average over 5 tests

    Screen Shot 2016-11-11 at 5.20.51 PM.png

    Now for the brucex, sometimes it would run at 19 and sometimes up to 24seconds so to me, it seems as if fcpx is using whatever is faster and not utilizing both cards together, or whatever is on slot 1. Who knows? I had a hunch that fcpx doesnt work with different cards but I figured, hey why not try it myself while im at it?

    Screen Shot 2016-11-11 at 5.06.15 PM.png Screen Shot 2016-11-11 at 5.18.00 PM.png Screen Shot 2016-11-11 at 5.18.20 PM.png
    Now what do I get from all of this? Personal satisfaction and reason to move on with my life,… somehow. I’ll be returning the R9 Nano as I think that it requires more power by itself and I don’t get any significant performance boost out of it from my dual rx 480. I also checked the power draw from the two cards during stress test at luxmark, and on Idle i was getting about 150 and hitting 480 at full blast, a delta of about 330W for both cards total.

    IMG_1635.JPG IMG_1636.JPG

    So now, Im looking to find another r9 390x and will be powering it externally. Now my question to you guys is, should I get the same exact gigabyte version or can I get by with say an MSI version? Reason I ask is that I haven’t been able to find any two similar cards up for sale anywhere for a reasonable price. The other Gigabyte R9 390x that microcenter stocks is located an hour and a half from me, so with tolls and gas and my precious time, at $247+ tax it might cost me about $300. Amazon and other online/retail stores current have them for atleast $380 which doesn’t tickle my fancy. I like my discounted cards.




    So can I mix and match brands?
     
  17. h9826790 macrumors G4

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #17
    Thanks for the detail report.

    It seems the Nano can't perform, just like the 480, its driver is working in MacOS, but not performing.

    For OpenCL, any card works, no need to be the same card. My current setup consist of two different brand 7950 (actually one of them is the 280), no problem at all in FCPX. BruceX is 15s stable with internal power only. Since you have extra PSU, of course you can go something better than that.

    However, from your result. It looks like the 390x's performance is very close to an OCed 280X. For best compadibility and optimisation , you may actually consider to run 2x OC 280x with external power.

    Since the 280x has same device ID as the D700. It will be always supported until the nMP obsolete. That's actually an advantage over the 390x.
     
  18. chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #18
    So I basically asked the question because I read on one of theitsage's post that he was having some issues with card placement on the PCIe slots and he found out that the problem being the two un-identical cards that differ in their base clockspeed:

    So im just trying to be wary of not getting two different cards for that reason. I know I can always come back to the store and return it, but I dont always want to be that guy. haha!

    You're most welcome! It's the least that I can do to try and contribute somehow. I know its a long read :/
    Yes, the nano was a bit disappointing for me so i'm not gonna hold on to it much longer.

    Wow thats actually something for me to think about! I went with the 480 and 390x primarily because of the 8GB vram. Even though I know i'll be switching cards in a year or two after selling these, I want to slightly "future proof" my purchase and have no excuse when it comes to my editing, specially because im working on 4K footage and having more and more gpu intensive effects put on it. 1080P was fine with my old macbook pro retina setup, but when I switched workflows, everything had to change as it couldn't keep up.

    I just checked the local used 280x cards in my area, and it seems like i'll be able to get two of them for the price of one 390x, which is insanely great! However, the VRAM is only at 3GB from what it seems so i definitely have to do more research on them. Im really happy to hear that you have that setup that its pretty much the D700 drivers and the future support because of it. Barring that, would you be able to tell me your experience with 4K footage and a bunch of effects on FCPX and how your 280x setup performs? I know brucex is a good baseline test but when multiple streams of 4K footage starts being layered, my RX 480 starts to slow down...
     
  19. h9826790 macrumors G4

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #19
    I don't have the 280X, but a pair of different brand 280 (which is effective a pair of 7950). They can run at different clock without any problem (I personally tested it).

    And I can edit 4K video without any issue. I do that just for fun on my leisure time. So, nowhere near any pro. BruceX is 15s on my machine. If you want to know any test result. Just tell me all the details on how to perform the test and I can do that for you.

    Dual OC 280X can easily go 50% faster than a stock clock 7950 (that's what I am using now). If my dual 7950 can achieve 15s in BruceX, may be dual OC 280X can achieve around 10s.

    More VRAM always good, but the main point is if 3GB enough for the job. If yes, then everything is fine. IMO, if you want something future proof. 480 may be a good candidate. If you want something can perform right now. 280x is the way to go.

    Since it's so easy to swap cards in the cMP. I personally perfer get a performing card now, and upgrade later when there is some better choice avail. No need to get a pair of 480 now and hope the driver will be improved.
     
  20. chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #20
    With the power consuption of the 8+6 pin cards (especially two of them), I really do have to decide on external power or the psu mod...

    Its good to know that different cards can work, even with 7950/280 mixing. That'll really make it easy for me to hunt for used cards. The 280x cards and above are 8+6 pins, so power definitely has to be addressed for my system.

    I already have a pair of 480s in my main for a couple of weeks, both being reference 6pins and with dual 3.33 xeons, i'm now getting 13 seconds out of them so i think the bottleneck for me was the CPUs.
    --- Post Merged, Nov 14, 2016 ---
    I was able to get a hold of a 2nd 390x, so I did some testing and crunching number to compare it to the 480s, but mainly to find out how much power I'll need to run them both. My test system was a standard dual 2.4ghz mac pro 5,1 with a 5770 from factory. Basically I used this default config as my baseline, and then began swapping and running stress tests on the cpu/gpu/system in order to draw as much power from it as possible. Since I haven't done any permanent power modification on the tower yet, I opted for a standard 750W PC PSU with the paperclip trick to let it run. I'll be using a Kill-A-Watt meter to log how much power it's drawing at a particular time. To avoid having any extra power draw, the system is limited only to standard mouse, keyboard, cd drive, and an SSD for system drive. I removed any mechanical drives. The monitor was also plugged in to a different wall outlet on a separate wall, so only the cMP and the PSU will have the dedicated power.

    Here's what I tested, doing all > Idle / stress CPU / GPU / CPU+GPU benchmarking then logging watts individually

    1. Stock config, 5770 gpu on 6 pin internal
    2. Stock config, 1x R9 390X on 6pin, 8pin on external PSU
    3. Stock config, 2x R9 390X on 6pin, 8pin on external PSU
    4. Stock config, 2x RX 480 on 6pin internal

    Screen Shot 2016-11-14 at 2.31.46 AM.png

    Each watts column are what Ive measured individually on taking the highest -consistent- peak and average of 3 tries.
    The actual wattage on the end columns are taken directly by having both ePSU and MacPro on the same power splitter. I've already subtracted the idle power draw of the ePSU alone while the system is on but under no load whenever applicable.

    The last two table sets are actually the only important summary. The dual 390X's are pulling about 280W each card (PCIe+8pin+6pin) and by subtracting [GPU(PCIe+6pin)] from [GPU w/8pin] , im getting a consistent draw of ~150W on the ePSU 8Pin connector.

    From reading AndreeOnline's Pixlas mod http://forums.macrumors.com/threads/pixlas-4-1-mac-pro-mod.1859652/ and what i've tested so far, the dual 8-Pin connector will surely draw the full 300W with two of these cards on tow. At full system load, i was maxing out at 824W, and with the cMP PSU rated at 980W, its getting fairly close, but with my day to day workload for the system, my peak watts was only hitting around 560s so Pixlas mod might actually be OK, but your mileage may vary.
     
  21. sevichenko macrumors regular

    Joined:
    Oct 9, 2016
  22. itdk92 macrumors 6502

    itdk92

    Joined:
    Nov 14, 2016
    Location:
    Copenhagen, Denmark
    #22
    How would your dual-480 performance compare to, let's say, a 15" rmbp from 2015?

    Have you had any chance to try both machines (or another rmbp 15"?)
     
  23. theitsage Suspended

    Joined:
    Aug 28, 2005
    #23
    I would crush the MBP.
     
  24. itdk92 macrumors 6502

    itdk92

    Joined:
    Nov 14, 2016
    Location:
    Copenhagen, Denmark
    #24
    I just saw it's you itsage :)

    Thanks again for the help with the 480 via email!


    But don't you think it's still superior even considering the actual state of the drivers?

    I mean, I can edit and color grade multiple 4k proxy streams with no problem atm on the 2015 15" rMBP

    I just hope it doesn't get worse with the mac pro (while we wait for decent support)
     
  25. chriz_r thread starter macrumors member

    chriz_r

    Joined:
    Nov 10, 2016
    #25
    I second theitsage's comment! I don't have the 2015 rMBP but i'll see if I can borrow my buddy's laptop for a few hours and do some testing with benchmarks and real world performance using fcpx.

    With the dual 480's on the mac pro, i can comfortably work on 3 4k streams with background render on but on "best performance view" so i'd still need to get some transcoding done to prores. It can definitely handle more than that when I transcode and use proxy media. Sometimes the events I edit have 4 or 5 4K cameras rolling at the same time, so multicam on those i'd have to use proxy.

    With the dual 390X, ive been just working on multicam and grading directly on 3 4K streams, sometimes i'll have random stutter but still better than dual 480. I can transcode to prores and then edit more than 4 angles very easily at "best quality" view so I dont really have to use proxy. I can certainly edit directly on 4K streams natively when time is of the essence and i have to submit deliverables within a few hours. But when there's time, I let the computer transcode. I'm still testing these setups so I'll keep you guys updated.
     

Share This Page