In my hands right now is a Sapphire RX 470 4GB OC Edition. Obviously I can't OC it but I'll test it out fully tonight with a bunch of pressure tests, if the drivers work.
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 RX 480s in Mac Pro 5,1 running MacOS (w/ How-to)
- Thread starter theitsage
- Start date
- Sort by reaction score
Awesome! I don't have many pro software to provide a good feedback on overall performance.
Good thing I'm starting prep early; shouldn't I be able to boot into recovery with a 280X? Like in the guide theitsage boots to recovery with the RX 480, right?
I realize I don't get the normal boot screen, but I expected the recovery mode to eventually pop up. But I get booted to desktop regardless of if I try cmd+r or opt+cmd+r.
I do have an original nvidia 120 something card, but that takes DVI and I don't have an adapter right now.
I realize I don't get the normal boot screen, but I expected the recovery mode to eventually pop up. But I get booted to desktop regardless of if I try cmd+r or opt+cmd+r.
I do have an original nvidia 120 something card, but that takes DVI and I don't have an adapter right now.
This card looks sweet. It's a metallic version of the reference design that suits the cMP perfectly. It's only one 6 pin connector. I'm downloading the 10.12.1 beta now.
Attachments
That a 470? Only card I could find with that same look was a 470 on amazon.
It's 470. In my opinion, it's a better value and better looking card than the reference 480.
The backplate is cool but makes for a very tight fit in the PCIe Slot 1. I took the backplate out.
[doublepost=1475261219][/doublepost]
Not sure why. I was able to enter Recovery Mode the very first time I tried the RX 480.
Probably not important if your card is in slot 1 and the backplate would otherwise just be sitting flat on the CPU compartment divider.
I'm probably not looking hard enough, but I just looked at Sapphire's site and the only card I can find that has that physical configuration is either the Sapphire Radeon RX 470 4G D5 or RX 480 4G / 8G D5. Do you have a link to the card, scsc?
OK, here's the report.
I followed theitsage's instructions to install the Sapphire RX 470. Yes, the backplate means you can't use slot 1 unless you remove the plate.
At first I reenabled SIP and when I rebooted my desktop wasn't accelerated. I scratched my head and wondered why. I tried the DVI port and the three DP ports to see if that made any difference. None.
Then I saw theitsage didn't reenable SIP. So I disabled it, rebooted and now I have an accelerated desktop on Sierra. You have to keep SIP disabled.
So I've started benchmarking. First of all our favourite Unigene Valley got a score of 1537 at Extreme HD ultra settings 1080p.
Then I ran the Photoshop Speed Test Action that we have on this forum. Photoshop identifies the GPU as a Baffin Prototype. Make sure you have all the advanced settings enabled including OpenCL and then restart the app.
The speed test action can cause Nvidia's web drivers to crash if OpenCL is enabled so I was interested to see how these beta Polaris drivers work. The speed test completes in 11 seconds on a dual X5650. That's about one second faster than the GTX980 was doing it before the web drivers took a quality nose dive. I went a step further and resized the test image to 20,000 pixels wide. The RX 470 handled it easily.
Now I'm downloading Luxmark and GFXBench for Metal testing.
[doublepost=1475268648][/doublepost]Luxmark results.
Luxball 10213 (almost the same as a GTX970)
Neumann 5862 (around the same as 280X)
Hotel 1648 (around the same as 380X)
All scenes passed validation, which is better than what the Nvidia web driver does sometimes.
[doublepost=1475270351][/doublepost]Geekbench openCL compute. Faster than a 780Ti, the same as a single D700, but much slower than the Windows Polaris driver.
I followed theitsage's instructions to install the Sapphire RX 470. Yes, the backplate means you can't use slot 1 unless you remove the plate.
At first I reenabled SIP and when I rebooted my desktop wasn't accelerated. I scratched my head and wondered why. I tried the DVI port and the three DP ports to see if that made any difference. None.
Then I saw theitsage didn't reenable SIP. So I disabled it, rebooted and now I have an accelerated desktop on Sierra. You have to keep SIP disabled.
So I've started benchmarking. First of all our favourite Unigene Valley got a score of 1537 at Extreme HD ultra settings 1080p.
Then I ran the Photoshop Speed Test Action that we have on this forum. Photoshop identifies the GPU as a Baffin Prototype. Make sure you have all the advanced settings enabled including OpenCL and then restart the app.
The speed test action can cause Nvidia's web drivers to crash if OpenCL is enabled so I was interested to see how these beta Polaris drivers work. The speed test completes in 11 seconds on a dual X5650. That's about one second faster than the GTX980 was doing it before the web drivers took a quality nose dive. I went a step further and resized the test image to 20,000 pixels wide. The RX 470 handled it easily.
Now I'm downloading Luxmark and GFXBench for Metal testing.
[doublepost=1475268648][/doublepost]Luxmark results.
Luxball 10213 (almost the same as a GTX970)
Neumann 5862 (around the same as 280X)
Hotel 1648 (around the same as 380X)
All scenes passed validation, which is better than what the Nvidia web driver does sometimes.
[doublepost=1475270351][/doublepost]Geekbench openCL compute. Faster than a 780Ti, the same as a single D700, but much slower than the Windows Polaris driver.
Attachments
Last edited:
Something new to macOS? The last few seconds of booting progress shows on this unflashed card meaning the driver intitialises earlier than Nvidia's web driver.
GFXBench does nothing. It made me download almost 300MB of scene data and it won't run any tests or give me any messages to indicate anything happened after I press 'Start'
GFXBench does nothing. It made me download almost 300MB of scene data and it won't run any tests or give me any messages to indicate anything happened after I press 'Start'
Attachments
Last edited:
Looks the same as yours. But tests don't do anything. Is the app broken or is it the driver?
Attachments
GPUTest/Furmark/Tessmark does nothing either. Click on the buttons and nothing happens. Again, this seems to be an application issue because even with software rendering checked the app still does nothing.
Just tested it, all tests working fine with my 7950. So, something wrong. but not the application itself. Anyway, I am also on the latest 10.12.1 beta.
I can't even see where the 300MB downloaded content was saved. Maybe that's the issue. I searched all library and system folders and can't see it anywhere.
Redid Luxmark to have solid benchmark on the same MacOS and Luxmark version to compare against incoming RX 480.
These are still from my 280X—a card that's a few years old now. Compared to SoyCap's RX470, my card still comes out 20-37% faster in Luxmark. I will try to focus on real apps once I get my card, even if I'll do the synthetic tests as well.
I'll test the cards individually, but my goal will be to have both cards installed in the end.
By the way: regarding starting Mac Pro in recovery—I realised I needed to have keyboard connected directly to Mac Pro and not via display as it was.
These are still from my 280X—a card that's a few years old now. Compared to SoyCap's RX470, my card still comes out 20-37% faster in Luxmark. I will try to focus on real apps once I get my card, even if I'll do the synthetic tests as well.
I'll test the cards individually, but my goal will be to have both cards installed in the end.
By the way: regarding starting Mac Pro in recovery—I realised I needed to have keyboard connected directly to Mac Pro and not via display as it was.
/Users/[your user name]/Library/Containers/net.kishonti.gfxbench.metal.osx/Data/Library/Caches
Yeah your drivers are more mature but they should be roughly the same. The benefit of the 470 is the 120w TDP and can easily install two of you need.
[doublepost=1475322575][/doublepost]
Ok I have cleaned the app and caches from my system and downloading a fresh copy to see what happens. I might also need to remove the GT120 in case that is causing confusion.
[doublepost=1475324770][/doublepost]The issue with GFXBench Metal is that it thinks I am trying to run a benchmark on a virtualised or custom OS (see main window) and then it can't fetch a compatible API or configuration from the server to run the tests (see log window). So it is an application problem. They should just send the damn data at the GPU just like other tests would and let the GPU deal with it if it works or not.
Attachments
Interesting, I may try to boot into recovery mode with my 7950 when boot from the PC ROM.
Of course, the 470 is not that powerful, AFAIK, it's main purpose is to improve efficiency, but not raw power. However, what confuse me a bit is the actual performance (and power draw) now in OSX looks like roughly the same as my tweaked 7950.
I keep my 7950 at stock 800/1250MHz, and down volt them to improve cooling and power efficiency.
As you can see, the Luxmark result is more or less the same as the 470, which is no big deal. However, the 2x 7950 total power consumption is just about 220W (110W per card) during the stress test.
I know TDP 120W doesn't mean the card will really draw 120W during the benchmark, however, I really want to know how much power the 470 actually draws during the Luxmark test (or stress test). If it still draw something like >100W, then there is almost no performance and power efficiency improvement from the 7950 to RX470 (in MacOS, at the moment). If it's actual power draw is <90W, then I will say the upgrade may be still worth. At least I will get more VRAM, and difference in power consumption is observable.
Of course, it may still hardly consider as an upgrade, but more like side stepping.
Last edited:
Let me know which app that one is to measure the power draw and I'll run with Luxmark.
That's Hardware Monitor. However, iStat can do the same thing. As long as you can get the voltage can current (both slots and the 6pins). It's very easy to work out the power draw.
The 470 is installed in slot 2. See if you can work out the power draw while Luxmark Hotel was rendering. It looks like just under 10 A from the slot and 6 pin cable, so less than 120W.
BTW, even when the system fans hit 1600RPM the GPU is barely audible under this heavy Luxmark render.
BTW, even when the system fans hit 1600RPM the GPU is barely audible under this heavy Luxmark render.
Attachments
Last edited:
Sure. Obviously the card is installed in slot 2. So,
Slot 2, 12.11V x 4.38A = 53W
PCIe Boost A, 12.12V x 5.36A = 65W
Therefore, the card actually drawing 53+65W = 118W.
That's what I thought, so it is on the specs exactly.
Another observation, with the Polaris installed the system fans are higher than normal and if you put them under load then the fans will stay around 1500RPM and won't go down.
With Nvidia we had a trick to use CUDA Z to reset the fans. I am unaware if we have a trick like that with the Radeons.