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

[h9826790]

I just run the test again as control.

2x7950, Hotel scenes, 3380 (>1648x2)

And

All slots and Boost cable at 12V, slot 1 2.78A, slot 3 1.95A, boost A 6.04A, boost B 6.55A.

So, 12 x (2.8+2+6+6.6)=208.8W

Let's round it up to 210W.

So, each 7950 only takes around 105W.

Anyway, 3380 / 210 = 16 score / watt
And 1648 / 118 = 14 score / watt

Conclusion, the RX470's efficient is actually lower then a tweaked 7950.

For the PCIe fan, should able to do the same tick with AMD card. I did that quite a few times with my 7950. But now I find the lazy method, which is use MacsFanControl and setup a PCIe fan profile like this.

So, the PCIe fan will now spin up and down base on the PCIe ambient temperature. I found this works very well in real world. When the GPU works hard, it will warm up the PCIe compartment, and then the PCIe fan will spin up accordingly.

With this setup, the PCIe fan will always work as expected, and never stay at 1500RPM with no reason.

 

[SoyCapitanSoyCapitan]

I just run the test again as control.

2x7950, Hotel scenes, 3380 (>1648x2)
View attachment 661920
And
View attachment 661922
All slots and Boost cable at 12V, slot 1 2.78A, slot 3 1.95A, boost A 6.04A, boost B 6.55A.

So, 12 x (2.8+2+6+6.6)=208.8W

Let round it up to 210W.

So, each 7950 only takes around 105W.

Anyway, 3380 / 210 = 16 score / watt
And 1648 / 118 = 14 score / watt

Conclusion, the RX470's efficient is actually lower then a tweaked 7950.

For the PCIe fan, should able to do the same tick with AMD card. I did that quite a few times with my 7950. But now I find the lazy method, which is use MacsFanControl and setup a PCIe fan profile like this.
View attachment 661930
So, the PCIe fan will now spin up and down base on the PCIe ambient temperature. I found this works very well in real world. When the GPU works hard, it will warm up the PCIe compartment, and then the PCIe fan will spin up accordingly.

With this setup, the PCIe fan will always work as expected, and never stay at 1500RPM with no reason.

Yeah but remember you have tweaked the 7950 and therefore you can even tweak a Polaris. Some people already doing that. The other key to remember is that a new card is future proof if there are updated APIs that are not supported by an older GPU. I had the option to buy a 280 for just 90 bucks or a 470 for 200. I chose the latter because I want to see if Metal will potentially benefit more from newer chips.

I just let the system idle completely and now the fans are down to around 900RPM. This appears to be the minimum even shortly after you boot. It's very silent anyway.

 

[h9826790]

Yeah but remember you have tweaked the 7950 and therefore you can even tweak a Polaris. Some people already doing that. The other key to remember is that a new card is future proof if there are updated APIs that are not supported by an older GPU. I had the option to buy a 280 for just 90 bucks or a 470 for 200. I chose the latter because I want to see if Metal will potentially benefit more from newer chips.

I just let the system idle completely and now the fans are down to around 900RPM. This appears to be the minimum even shortly after you boot. It's very silent anyway.

Sure if I have to buy another card now. I will pick the RX470 but not the R9 280. More RAM, better architecture, support the newer software / standard / API...

Also good to know that the 470 is a very quiet card. I hate loud cooler.

And I totally agree that we can tweak the Polaris. However, I am surprised a little tweaked 7950 can match the stock RX470. Also, I really want to know how much the RX470 can be tweaked. It has to cut the power consumption by 15% in order to match the tweaked 7950's efficiency.

 

[handheldgames]

Great benchmarks! Now.. some serious thoughts on power usage.
With the help of Luxmark, iStats Menu 5.0 and the Amps to Watts Calculator, it's easy to see the XFX RX 480's power draw exceeds the cMP's maximum rated amount of 75 Watts. The RX 470 and 7950 appear to stay within specification. Running any benchmark, looking beyond a temperature gauge can reveal some interesting findings.

RX 480 Luxmark ~ power usage
PCIE0: 6.23A @ 12.06v = 75.13 Watts (max is 75watts)
PCIE Boost A: 5.6A @ 12.06v = 67.54 Watts

XFX RX 470 Luxmark ~ power usage
PCIE0: 4.65A @ 12.06v = 56.07 Watts
PCIE Boost A: 5.31A @ 12.06v = 64.09Watts

Gigabyte 7950 Luxmark ~ power usage
PCIE0: 2.9A @ 12.12v = 34.04 Watts
PCIE Boost A: 6.1A @ 12.12v = 73.93 Watts
PCIE Boost B: 3.5A @ 12.12v = 42.42 Watts​

For those that are inclined to dig into the details of rx 480 bios editing tools, etc...., head over to Overclock.net for details on RX480 / 470 / 460 Windoze bios editing tools, etc. A RX 480 bios with lowered PCIE0 power draw is available at the Overclocking Guide.

At idle, the RX 480 PCIe power draws around .~47A @12.12v. whereas the 7950 draws ~1.1A. Demonstrating the power efficiency at idle the new GPU's bring to the table. Idling aside, running a 3D Development environment with the RX 480, I see a consistent 2.88V on PCIe0 and 3.42V from BoostA. Well within specs.

 

[namethisfile]

I'm coming from an HD5770. Should I get the RX 470 right now? Or wait for Sierra 12.1? I'm not running the Dev kit. And, will be using my cMP 2010 for FCPX and Motion5....

Thanks!

PS--also, will the stock 6-pin PCI cable that is for the HD5770 fit or reach the RX470 since I notice that the 6-pin connector for the apple HD5770 is in a different location than the on RX470.

 

[h9826790]

I'm coming from an HD5770. Should I get the RX 470 right now? Or wait for Sierra 12.1? I'm not running the Dev kit. And, will be using my cMP 2010 for FCPX and Motion5....

Thanks!

PS--also, will the stock 6-pin PCI cable that is for the HD5770 fit or reach the RX470 since I notice that the 6-pin connector for the apple HD5770 is in a different location than the on RX470.

That cable perfectly fit for the RX470.

However, if you use this machine to work, I personally think it's better to wait for bit longer. Its too early to use RX470 on a productive computer which running an non officially supported OS.

If there is any serious bug which affect FCPX / Motion 5 hasn't been discover yet. Then it may give you big trouble.

If you only use FCPX for fun, then of course not a big issue.

 

[SoyCapitanSoyCapitan]

I wouldn't advise anyone to buy an unsupported GPU or beta driver for professional work. We already saw how badly it affected people using the Nvidia web driver's neverending beta support for Maxwell. It also hurt people who bought 290X.

Having said that, the 470 is cheapish and if it doesn't work out you can refund it or you can sell it with minimum loss.

 

[SoyCapitanSoyCapitan]

Removed the GT120. Now the RX 470 shows the full progress bar when booting the system.

GFXBench still doesn't work and is one of the worst made pieces of software I have seen on a Mac. Scroll bars disappear from it, the log file says it can't set or fetch a configuration to run tests, and I have to force quit the app to close it every time, even if I did nothing. So I am going to disregard this as a decent test for anything and will look for another Metal test.

 

[namethisfile]

That cable perfectly fit for the RX470.

However, if you use this machine to work, I personally think it's better to wait for bit longer. Its too early to use RX470 on a productive computer which running an non officially supported OS.

If there is any serious bug which affect FCPX / Motion 5 hasn't been discover yet. Then it may give you big trouble.

If you only use FCPX for fun, then of course not a big issue.

I wouldn't advise anyone to buy an unsupported GPU or beta driver for professional work. We already saw how badly it affected people using the Nvidia web driver's neverending beta support for Maxwell. It also hurt people who bought 290X.

Having said that, the 470 is cheapish and if it doesn't work out you can refund it or you can sell it with minimum loss.

Yeah.... That's what I was thinking... wait a bit longer... wait for the upcoming macs to see if it will feature AMD Polaris cards... wait for future Sierra updates that support said cards... and so on and so forth.

With that said, some of you are running RX series cards here. And, I'm curious, if it is indeed workable with apps like FCPX and Motion5 and whether or not it is stable.

You guys are running benchmark test. But, what about just regular tests of working a project on FCPX and how it works there? How fast is it there?

Any of you guys have FCPX?

Thanks!

 

[SoyCapitanSoyCapitan]

Yeah.... That's what I was thinking... wait a bit longer... wait for the upcoming macs to see if it will feature AMD Polaris cards... wait for future Sierra updates that support said cards... and so on and so forth.

With that said, some of you are running RX series cards here. And, I'm curious, if it is indeed workable with apps like FCPX and Motion5 and whether or not it is stable.

You guys are running benchmark test. But, what about just regular tests of working a project on FCPX and how it works there? How fast is it there?

Any of you guys have FCPX?

Thanks!

Unfortunately not and I haven't worked with Premiere or After Effects for a while on a Mac and prefer to use the Windows versions because Media Encoder is a lot faster on Windows (like 4x faster with the same hardware)

 

[namethisfile]

Unfortunately not and I haven't worked with Premiere or After Effects for a while on a Mac and prefer to use the Windows versions because Media Encoder is a lot faster on Windows (like 4x faster with the same hardware)

What about FCPX? I hear AMD cards are better for FCPX anyway.

 

[SoyCapitanSoyCapitan]

What about FCPX? I hear AMD cards are better for FCPX anyway.

Yes they are because Apple and AMD have worked closely to optimise FCPX for OpenCL and Radeons. Let's see what happens when FCPX integrates Metal. You are in the middle of a transitional period at the moment so it is best to hold off purchases until you see what is released.

 

[AndreeOnline]

Hopefully I'll get my card (RX 480) tomorrow and I'll do BruceX (FCPX) and Standard Candle (DaVinci Resolve) plus some actual test work in those apps.

I'll also check Cinebench R15, because I work quite a bit in C4D and that should be the best benchmark for that. My tests will be in MacOS. We'll see what happens.

 

[namethisfile]

Hopefully I'll get my card (RX 480) tomorrow and I'll do BruceX (FCPX) and Standard Candle (DaVinci Resolve) plus some actual test work in those apps.

I'll also check Cinebench R15, because I work quite a bit in C4D and that should be the best benchmark for that. My tests will be in MacOS. We'll see what happens.

Cool. Looking forward to your update!

 

[h9826790]

Hopefully I'll get my card (RX 480) tomorrow and I'll do BruceX (FCPX) and Standard Candle (DaVinci Resolve) plus some actual test work in those apps.

I'll also check Cinebench R15, because I work quite a bit in C4D and that should be the best benchmark for that. My tests will be in MacOS. We'll see what happens.

No, Cinebench is a well known software that can't benchmark the GPU at all. That GPU part is basically benchmarking your CPU's single thread speed. Unless you put an extremely poor GPU in, the result are always the same with the same CPU, but not GPU.

 

[namethisfile]

No, Cinebench is a well known software that can't benchmark the GPU at all. That GPU part is basically benchmarking your CPU's single thread speed. Unless you put an extremely poor GPU in, the result are always the same with the same CPU, but not GPU.

Isn't there an OpenGL test in Cinebench?

 

[h9826790]

Isn't there an OpenGL test in Cinebench?

Yes, however, that test is very poorly written, which end up heavily CPU single thread bounded. Therefore, unable to properly measure the GPU performance.

 

[koyoot]

For graphics benchmarking I suggest using... games. World of Warcraft, Hearthstone, League of Legends, etc.

 

[namethisfile]

Yes, however, that test is very poorly written, which end up heavily CPU single thread bounded. Therefore, unable to properly measure the GPU performance.

Yes. But, whether or not it is an antiquated benchmark that is poorly written, would it show comparatively, what an RX470 card can do on that test, compared to an HD5770, anyway?

Or, do you think it's pointless?... That an RX470 will perform so badly in that the benchmark that the result will be inconclusive? Such as shown in some cases, where new GPU's don't show any gain or even losing performance because of newness-factor, or driver issues, or whathaveyou?

 

[h9826790]

For graphics benchmarking I suggest using... games. World of Warcraft, Hearthstone, League of Legends, etc.

I think that's benchmarking the overall system more then specifically the GPU.
[doublepost=1475425019][/doublepost]

Yes. But, whether or not it is an antiquated benchmark that is poorly written, would it show comparatively, what an RX470 card can do on that test, compared to an HD5770, anyway?

Or, do you think it's pointless?... That an RX470 will perform so badly in that the benchmark that the result will be inconclusive? Such as shown in some cases, where new GPU's don't show any gain or even losing performance because of newness-factor, or driver issues, or whathaveyou?

As I said, the 470's performance in Cinebench is totally meaningless, because the result will be completely CPU bounded.

For OpenGL benchmarking, Unigine Valley / Unigine Heaven are much better tools.

 

[SoyCapitanSoyCapitan]

No, Cinebench is a well known software that can't benchmark the GPU at all. That GPU part is basically benchmarking your CPU's single thread speed. Unless you put an extremely poor GPU in, the result are always the same with the same CPU, but not GPU.

Cinebench is screwy on the Mac, mostly down to the crappy old GL stack.

The Unigene stuff is also heavy on CPU but not as much. It's basically Apple's fault here for not having modern or mature APIs.

 

[AndreeOnline]

No, Cinebench is a well known software that can't benchmark the GPU at all.

In the case of Cinebench, I'm not primarily interested in benchmarking the GPU. I want to see what effect, if any, I get by swapping from a 280X to a RX 480.

Yes, however, that test is very poorly written…

I see this comment a lot. Perhaps because people always run benchmarks with preconceived results in mind? Every benchmark is utterly meaningless, unless it translates to real world performance.

Cinema 4D is known for not being able to "take advantage" of all your hardware when it comes to viewport rendering (it does use all cores linearly when rendering out image of course). It's single threaded and whatnot. Well, that's what it is and Luxmark won't change that. I'm running Cinebench in the assumption that it might have some relevance to actual viewport performance. And some numbers I've seen from Polaris cards are quite an improvement on what I get from my 280X. I'll report back when I've done the actual comparison.

At any rate, the 8GB RAM surely won't hurt when doing actual work.
[doublepost=1475426298][/doublepost]

For graphics benchmarking I suggest using... games. World of Warcraft, Hearthstone, League of Legends, etc.

Problem with that is, it has no relevance whatsoever to what I want to do.

 

[h9826790]

In the case of Cinebench, I'm not primarily interested in benchmarking the GPU. I want to see what effect, if any, I get by swapping from a 280X to a RX 480.

Because you said "that should be the best benchmark", that's why I said don't use it as a benchmark. However, if what you mean is actually just use it to check the effect, then of course nothing wrong.

 

[handheldgames]

The Good...
1) The RX480 and RX470 are stable. Drivers are improving, although performance is limited, bugs exist. Will support for these cards be pulled? Highly unlikely.
2) Apple / AMD is updating the drivers and OpenGL lib's every release. A new hardware release is pending.
3) Open CL / Metal rendering appears to be solid and fast.
4) Drivers are built into the OS and initialize at the start of the boot process, welcoming back the apple menu.
5) My personal favorite... With Updates to the OS, just update a KEXT and reboot. Nvidia driver headaches begone!

The Bad... (in limited instances)
1) OpenGL has issues. MSAA applied to still images destroys the frame buffer. Disabling MSAA for still screen rendering makes the display workable, although reflections are trashed.

 

[namethisfile]

The Good...
1) The RX480 and RX470 are stable. Drivers are improving, although performance is limited, bugs exist.
2) Apple / AMD is updating the drivers and OpenGL lib's every release. A new hardware release is pending.
3) Open CL / Metal rendering appear solid and fast.
4) Drivers are built into the OS and initialize at the start of the boot process, welcoming back the apple menu.
5) My personal favorite... With Updates to the OS, just update a KEXT and reboot. Nvidia driver headaches begone!

The Bad... (in limited instances)
1) OpenGL has issues. MSAA applied to still images destroys the frame buffer. Disabling MSAA for still screen rendering makes the display workable, although reflections are trashed.

So, what you're saying is that RX470 and RX480 is stable enough for daily use and for working in FCPX?

Maybe not so much gaming. But, is stable enough for daily use and video editing work in FCPX?