Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Dual RX 480s in Mac Pro 5,1 running MacOS (w/ How-to)

theitsage said:
I'd get an RX 480. It runs with one 6-pin PCIe cable just like the RX 470.

The coupon is for in-game upgrade.
Click to expand...

So I have to still buy the game?

As for the RX480, I have read about it having power draw issues. The 6-pin is not enough or something and it is pulling more power from the PCIE slot to compensate. Is this still an issue?
[doublepost=1475762265][/doublepost]
theitsage said:
I found an open-box RX 480 8GB at Micro Center this morning and gave it a shot. Installed in my Mac Pro 5,1 which is running MacOS Beta 3, it can boot to the Desktop without any modification.

I recall reading there's built-in driver for this GPU but ID is not matched. Anything I should try?

Summary: RX 470 and RX 480 both work with acceleration and Metal support in macOS after adding PCI ID 0x67DF1002 to AMDRadeonX4100.kext. Other than no Boot Screen, you'll see UI during system update and in Recovery Mode. Kext editing needed after each OS update to reenable acceleration and Metal support.

View attachment 661380

Update 1: Installed GM macOS Sierra build 16A319 today (9/8/16). While the RX 480 still works after the kext mod, OpenGL crashes remain an issue. The RX 470 on the other hand is fully functional without boot screen.

Update 2: Installed 10.12.1 Beta tonight (9/21/16). Polaris support is definitely getting better. The Baffin personality which was in AMDRadeon4000.kext now has its own kext, AMDRadeon4100.kext. RX 480 is not crashing OpenGL apps anymore. This is the first time in nearly 2 months owning these RX 480 GPUs I was able to finish Unigine benchmarks.

Update 3: I received PMs for instructions so I wrote this Mac Pro Radeon RX 480 How-to.
Click to expand...

So, which card is more stable in MacOS? RX470 or RX480. I might be close to just getting either one and play around with it.

But still not sure. I'm concerned with RX480 not being stable as opposed to the RX470.

So, do you need to disable SIP then re-enable it afterwards?

Also, can I manually inject the Kext without using Kext Utility. What is Kext Utility doing. Can't that stuff be done in Terminal?

Thanks!
 
Last edited:
Power draw is no longer an issue. I'm using a pair of RX 480 GPUs in a 5,1 with the 6-pin cables and no issues in both macOS and Windows 10. Another 5,1 runs one RX 470. They are the same in terms of stability with 10.12.1 beta.

SIP has to remain disabled. I'm sure there are instructions out there on how to use Terminal to modify kext files.
 
theitsage said:
Power draw is no longer an issue. I'm using a pair of RX 480 GPUs in a 5,1 with the 6-pin cables and no issues in both macOS and Windows 10. Another 5,1 runs one RX 470. They are the same in terms of stability with 10.12.1 beta.

SIP has to remain disabled. I'm sure there are instructions out there on how to use Terminal to modify kext files.
Click to expand...

Cool. But, can you explain why SIPS has to be disabled? Is it for the editing of the GPU ID? What would happen if I don't disable SIP and proceed with editing the GPU ID and manually fix permission via terminal command?
 
namethisfile said:
Cool. But, can you explain why SIPS has to be disabled? Is it for the editing of the GPU ID? What would happen if I don't disable SIP and proceed with editing the GPU ID and manually fix permission via terminal command?
Click to expand...

SIP needs to be disabled because on reboot, OS X will check KEXT integrity and your modification won't match the checksum. It won't load, and as a result OS X won't boot.

That's assuming you're even allowed to modify the KEXT in the first place without SIP disabled.
 
Draeconis said:
SIP needs to be disabled because on reboot, OS X will check KEXT integrity and your modification won't match the checksum. It won't load, and as a result OS X won't boot.

That's assuming you're even allowed to modify the KEXT in the first place without SIP disabled.
Click to expand...

How can we maintain the original checksum in order to enable SIP again?
 
Really wish we see Metal apps soon because until then the power of Polaris (or any Metal compatible GPU) is completely untapped on macOS compared to Windows. How long do we have to wait until this API is truly prime time?
 
theitsage said:
Power draw is no longer an issue. I'm using a pair of RX 480 GPUs in a 5,1 with the 6-pin cables and no issues in both macOS and Windows 10. Another 5,1 runs one RX 470. They are the same in terms of stability with 10.12.1 beta.

SIP has to remain disabled. I'm sure there are instructions out there on how to use Terminal to modify kext files.
Click to expand...

How much can a cMP draw over a PCIe line without causing damage? 75W is the rated maximum. Benchmarks run on the RX480, constantly draws > 75 watts without a modified BIOS. With consecutive benchmark runs, voltage draw incrementally increases.
 
Last edited:
handheldgames said:
How much can a cMP draw over a PCIe line without causing damage? 75W is the rated maximum. Benchmarks run on the RX480, constantly draws > 75 watts without a modified BIOS. With consecutive benchmark runs, voltage draw incrementally increases.
Click to expand...

I don't have enough data to give you a good answer. The pair of RX 480s in my 5,1 ran many taxing hours in CrossFire under Windows 10. Sure, it's not the safest thing for these aging Mac Pro towers but we're modding them so safety is not top of mind.
 
theitsage said:
I don't have enough data to give you a good answer. The pair of RX 480s in my 5,1 ran many taxing hours in CrossFire under Windows 10. Sure, it's not the safest thing for these aging Mac Pro towers but we're modding them so safety is not top of mind.
Click to expand...

In Windows is not a problem, because the undated driver now reduce the power draw from the PCIe slot, and increase the power draw from the 6pin to maintain performance.

However, on OSX side, everything is based on the BIOS, that means the 480 will constantly draw >75W under stress.

AFAIK, the practical max of the cMP PCIe slot can deliver still an unknown.

If there is any protection like the mini 6 pin does also an unknown.

I personally tends to believe that the cMP is over engineered. So an extra 20-30% loading on the slot should not be a problem. However, there is no proof, no test, no record about this.

From memory, MVC did kill 1 or 2 PCIe slot, but can't remember how he managed to do it, and if that's power related.
 
Guys, I have a question for you all. Is it possible, that even there are Ellesmere traces in the system KEXT, the reason why the GPUs are not working is because they are not exactly the same thing what Apple can put in the computers?

There is a rumor that AMD might want to refresh Polaris lineup with GDDR5X versions and on 14nm SAMSUNG process, instead of GloFo.
 
This is the screen cap of info.plist and controller file inside AMD9500Controller.kext. I have been curious of what 67C0 might be. It's a Polaris 10/Ellesmere card but higher end than RX 470/480 (67DF).

Screen Shot 2016-10-07 at 08.57.44.png
 
Two possibilities - 67C0 could be a dual GPU or GDDR5X variant, most likely named RX 490.

I read an article a while back which mentioned earlier in development, AMD didn't finalize the Ellesmere name to Polaris 10 and in general was referred to as Baffin XT. Apple probably didn't bother to change this.

Baffin = Polaris 11 = 67E0, 67EF/RX 460, 67FF ==> for MBP
Ellesmere = Polaris 10 = 67C0, 67DF/RX 470/RX 480 ==> for Mac Pro/iMac
 
Truth is a little different.

There is upcoming quite a lot of release/refresh from AMD.
RX 465 - 1024 GCN core, GDDR5 GPU with possibly higher clocks than RX 460, and made on 14nm Samsung process.
Polaris 10, and 11 for embedded market are made on Samsung process.
RX 475 and RX 485 with GDDR5X, plus Dual Ellesmere XT Chip are due for refresh, and release, however dual chip will not have RX 490 branding.

RX 490 will be small Vega GPU, and RX Fury - big Vega.

And what is funnier, now people start to speculate, that Nvidia will refresh Pascal GPUs with higher clocks. In this industry nothing happens without a reason.

Remember that GloFo and Samsung synched the process? SemiAccurate was rumbling about when it was going into fruition, and had one big customer in the background requesting the sales of the IBM fab to GloFo, and synching the process from GloFo and Samsung. How come?

http://semiaccurate.com/2013/08/26/a-third-player-emerges-apples-foundry-plans/
http://semiaccurate.com/2013/12/17/apple-samsung-intel-foundry-plans/
http://semiaccurate.com/2014/04/17/semiaccurate-right-apples-foundry-plans/ with specific quote, that links to one of the articles on SA:
They did the same for the first part of the story too, the main customer for this 14nm synced process.
Click to expand...

Interesting time, really ;)

If the refreshed GPUs are made on Samsung process, and are slightly different than RX 480/470 - then we know why RX 480 and 470 does not work in macOS Out of the Box.
 
Last edited:
Barefeats test.

http://barefeats.com/sierra_rx480.html

The scale he extrapolated is completely hypothetical though and likely won't reflect how a Pascal would work on macOS because even Maxwell doesn't run as well as it should with the web driver.

Also, the Fury scores being so close to the 480 means the Mac drivers also suffer from the same issue the Nvidia web drivers do - the performance isn't optimised and doesn't scale upwards as it does on Windows.

Most likely also because OpenGL 4.1 is ****
 
koyoot said:
Guys, I have a question for you all. Is it possible, that even there are Ellesmere traces in the system KEXT, the reason why the GPUs are not working is because they are not exactly the same thing what Apple can put in the computers?
Click to expand...

Apple did that before, but in some other way: device IDs of 7xxx cards were recognized by 10.9.x as Dxxx cards until nMP release. Dxxx cards share device IDs (and GPUs) with some of 7xxx cards but drivers are recognizing each series differently.
BTW, AMD9500Controller at current (released) development stage (10.12.1 Dev Beta 3) doesn't contain any nMP-specific framebuffer personality (6x mDP + HDMI). But AMD9000Controller does. 2 complete and 1 incomplete personalities since 1st Sierra beta. This could mean something or nothing.
 
Slightly off topic. I really want RX470 at least. But what about RX460?

I basically just want an upgrade to my HD5770. But, how much is an RX460 an upgrade to the HD5770? I know it will have more VRAM. But, what else? Specifically, for FCPX and video production stuff.

I don't care if RX460 is slow for games (I already have a dedicated Windows PC for gaming). I just want a video machine workstation.

I'm asking about the RX460 because I read that the Gigabyte one specifically works OOB. No messing around with ID's, KEXT and whatnot.

Is this true?

If so, since it doesn't require a PCIE cable. Can I put an HD5770 and an RX460 together and have them work together?

Thanks!
 
namethisfile said:
Slightly off topic. I really want RX470 at least. But what about RX460?

I basically just want an upgrade to my HD5770. But, how much is an RX460 an upgrade to the HD5770? I know it will have more VRAM. But, what else? Specifically, for FCPX and video production stuff.

I don't care if RX460 is slow for games (I already have a dedicated Windows PC for gaming). I just want a video machine workstation.

I'm asking about the RX460 because I read that the Gigabyte one specifically works OOB. No messing around with ID's, KEXT and whatnot.

Is this true?

If so, since it doesn't require a PCIE cable. Can I put an HD5770 and an RX460 together and have them work together?

Thanks!
Click to expand...

I haven't seen any FCPX benchmark result for the 460 yet, so a bit hard to tell. FCPX really well optimised for the 5770 and 5870. But I guess 460 will be faster on most of the operation (for computation, the 460 can double the 5770's performance on some specific task), which only few task that the 5770 may able to match the 460 or even slightly faster.

I personally agree that a OOTB card is a better choice. No one know what Apple will do, even though they can drop the 460 support on the next update, but at least the chance to survive is higher than the card that never natively supported in MacOS.

And technically yes, the 5770 and 460 should be able to performance together in FCPX. Also, the 5770 can still provide you the boot screen, which is actually pretty good (if everything work as expected).
 
h9826790 said:
I haven't seen any FCPX benchmark result for the 460 yet, so a bit hard to tell. FCPX really well optimised for the 5770 and 5870. But I guess 460 will be faster on most of the operation (for computation, the 460 can double the 5770's performance on some specific task), which only few task that the 5770 may able to match the 460 or even slightly faster.

I personally agree that a OOTB card is a better choice. No one know what Apple will do, even though they can drop the 460 support on the next update, but at least the chance to survive is higher than the card that never natively supported in MacOS.

And technically yes, the 5770 and 460 should be able to performance together in FCPX. Also, the 5770 can still provide you the boot screen, which is actually pretty good (if everything work as expected).
Click to expand...

Has that been known to happen where a PC card works OOB and then it doesn't anymore?

But, the RX460 sounds really more tempting to me now after learning more about compatibility issues.
 
namethisfile said:
Has that been known to happen where a PC card works OOB and then it doesn't anymore?

But, the RX460 sounds really more tempting to me now after learning more about compatibility issues.
Click to expand...

It's happened before, yes.

But check here for comparison between 5770 and RX 460. No benchmarks, but you get a rough idea.
 
666sheep said:
Apple did that before, but in some other way: device IDs of 7xxx cards were recognized by 10.9.x as Dxxx cards until nMP release. Dxxx cards share device IDs (and GPUs) with some of 7xxx cards but drivers are recognizing each series differently.
BTW, AMD9500Controller at current (released) development stage (10.12.1 Dev Beta 3) doesn't contain any nMP-specific framebuffer personality (6x mDP + HDMI). But AMD9000Controller does. 2 complete and 1 incomplete personalities since 1st Sierra beta. This could mean something or nothing.
Click to expand...
Let me guess. One of them is Fiji related?
 
Draeconis said:
It's happened before, yes.

But check here for comparison between 5770 and RX 460. No benchmarks, but you get a rough idea.
Click to expand...

It's hardly any faster at all. LOL!

Do you think that it is worth it? Or should I wait and pray, sorta speak, that Polaris 10, will become fully accelerated and supported OOB in future Sierra update?

What do you think the chances are that future Sierra versions will fully support, say an RX470 without modifying anything. Just stick it in and go?
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back