Contribute
Register

macOS Sierra PB: Need testers for new AMD Radeon drivers!

Status
Not open for further replies.
I've been watching this thread since beginning. GTX 1070 was my primary preference to change my dGPU, but with the progress showed here, decided to get a RX 480 instead of wait for Pascal support from lazy Nvidia. This is my first decent graphics card and I am very satisfied. Finally, I retired my good old Geforce 210.

I have a XFX RX 480 RS 8Gb running on 10.12.2 with latest Clover version 3961. Just set ATI FakeID = 0x67EF1002 and Inject ATI = true to properly work. Using DVI-D cable in a LG IPS236 1080p monitor. Haven't tested HDMI, but it should work fine as usual. My iGPU HD 2500 is connected to same monitor with a HDMI cable.

My motherboard is Gigabyte GA-H61M-DS2H. On stock bios version F5, bios doesn't even load with RX 480. On F7, card is recognized only in legacy mode. Just in the latest and beta bios F8a, which improve high-end display card compatibility, it worked on UEFI VBIOS.

I did the changes proposed by Fl0r!an and Mork_vom_Ork to unlock the full potential from Polaris 10.

What worked to improve performance


- Remove CU limit, change CU number and change init from Baffin to Ellesmere. Improves Luxmark v3.0 in 27,62%! Thanks a lot!!! :clap:
Screen_Shot_2016_12_17_at_00_31_57.png

RX_480_Unlocked_without_Power_Containment_Disabled.png


Unigine has not improved. Score remains around 1240.
image.png


- Changing PP_DisablePowerContainment(AMD9500Controller.kext) to 1 improve Luxmark performance a little, but Unigine crash the system between scene 8 and 11(tested three times). The crash happened because, besides it disable the "power containment", it turns off the fan RPM control, that doesn't raise speed, keeping slow RPM as the GPU wasn't heating up. So, temperature rises rapidly(VBIOS limit is 90ºC, 194ºF), the RX 480 screen goes black, but not HD2500(which freezes after this) and the VBIOS make the fan spin at full speed. After this, the only way out is press reset button.

Mystery: If the kext file is changed, the behavior is like described above, but if patched on-the-fly with Clover, the fan control works normally, however performance is decreased(lower than normal).

Kext file changed (Unigine crash)
Power_Containment_Disabled_Kext_Changed.png


Clover on-the-fly patch
Power_Containment_Disabled_Kext_Patch2.png

Power_Containment_Disabled_Kext_Patch.png


What doesn't work

- Injecting Intel HD Graphics 2500, we can get only CI, not QE. The only way to inject without boot to black screen(HD 2500) is deleting AppleIntelHD4000Graphics.kext, leaving just AppleIntelFramebufferCapri.kext loaded, which is needed for full resolution(in this case, can get full resolution even if it doesn't load). But injecting iGPU, even if only the framebuffer loads, it drastically reduces the performance of the RX 480 on Unigine Heaven.

Screen_Shot_2016_12_17_at_09_30_09.png


- Delete AppleIntelFramebufferCapri.kext, because it gets loaded, even without HD 2500 injection. But it made no difference.

- Initialize without the HDMI cable on the HD 2500(Blind boot). Same performance on Unigine and Luxmark. I see a ghost monitor with 1600x900 resolution and if plug the cable into the monitor after the system loads, there is no video signal.

- Initialize without the two cables, DVI and HDMI, and connecting DVI to RX 480. Idem.

- Different FakeID's E0, EF and FF provides same performance(I read several pages ago that switching could improve performance).

- Same performance using legacy or UEFI VBIOS, changing the switch position on the board side.

- Patching AMDRadeonX4100.kext with hexadecimal editor or Clover kext patch method, made no difference too.

Bruce X Benchmark 5k (FCPX 10.3.1)*

Original AMDRadeonX4100.kext w/ 16 CU: 29 seconds.
Patched** AMDRadeonX4100.kext w/ 36 CU: 29 seconds too.

*MacOS Sierra is installed on a WD Blue 1TB HDD, but the master file was exported to Intel 540s 120Gb SSD.
** 36 CU and init function to Ellesmere

The DisableMonitor app, mentioned here, is good for not losing your mouse cursor in iGPU screen.

Thanks to Gigamaxx, Fl0r!an and Mork_vom_Ork! :thumbup:
 
Last edited:
I've been watching this thread since beginning. GTX 1070 was my primary preference to change my dGPU, but with the progress showed here, decided to get a RX 480 instead of wait for Pascal support from lazy Nvidia. This is my first decent graphics card and I am very satisfied. Finally, I retired my good old Geforce 210.

I have a XFX RX 480 RS 8Gb running on 10.12.2 with latest Clover version 3961. Just set ATI FakeID = 0x67EF1002 and Inject ATI = true to properly work. Using DVI-D cable in a LG IPS236 1080p monitor. Haven't tested HDMI, but it should work fine as usual. My iGPU HD 2500 is connected to same monitor with a HDMI cable.

My motherboard is Gigabyte GA-H61M-DS2H. On stock bios version F5, bios doesn't even load with RX 480. On F7, card is recognized only in legacy mode. Just in the latest and beta bios F8a, which improve high-end display card compatibility, it worked on UEFI VBIOS.

I did the changes proposed by Fl0r!an and Mork_vom_Ork to unlock the full potential from Polaris 10.

What worked to improve performance


- Remove CU limit, change CU number and change init from Baffin to Ellesmere. Improves Luxmark v3.0 in 27,62%! Thanks a lot!!! :clap:
Screen_Shot_2016_12_17_at_00_31_57.png

RX_480_Unlocked_without_Power_Containment_Disabled.png


Unigine has not improved. Score remains around 1240.
image.png


- Changing PP_DisablePowerContainment to 1 improve Luxmark performance a little, but Unigine crash the system between scene 8 and 11(tested three times). The crash happened because, besides it disable the "power containment", it turns off the fan RPM control, that doesn't raise speed, keeping slow RPM as the GPU wasn't heating up. So, temperature rises rapidly(VBIOS limit is 90ºC, 194ºF), the RX 480 screen goes black, but not HD2500(which freezes after this) and the VBIOS make the fan spin at full speed. After this, the only way out is press reset button.

Mystery: If the kext file is changed, the behavior is like described above, but if patched on-the-fly with Clover, the fan control works normally, however performance is decreased(lower than normal).

Kext file changed (Unigine crash)
Power_Containment_Disabled_Kext_Changed.png


Clover on-the-fly patch
Power_Containment_Disabled_Kext_Patch2.png

Power_Containment_Disabled_Kext_Patch.png


What doesn't work

- Injecting Intel HD Graphics 2500, we can get only CI, not QE. The only way to inject without boot to black screen(HD 2500) is deleting AppleIntelHD4000Graphics.kext, leaving just AppleIntelFramebufferCapri.kext loaded, which is needed for full resolution(in this case, can get full resolution even if it doesn't load). But injecting iGPU, even if only the framebuffer loads, it drastically reduces the performance of the RX 480 on Unigine Heaven.

Screen_Shot_2016_12_17_at_09_30_09.png


- Delete AppleIntelFramebufferCapri.kext, because it gets loaded, even without HD 2500 injection. But it made no difference.

- Initialize without the HDMI cable on the HD 2500(Blind boot). Same performance on Unigine and Luxmark. I see a ghost monitor with 1600x900 resolution and if plug the cable into the monitor after the system loads, there is no video signal.

- Initialize without the two cables, DVI and HDMI, and connecting DVI to RX 480. Idem.

- Different FakeID's E0, EF and FF provides same performance(I read several pages ago that switching could improve performance).

- Same performance using legacy or UEFI VBIOS, changing the switch position on the board side.

Bruce X Benchmark 5k (FCPX 10.3.1)*

Original AMDRadeonX4100.kext w/ 16 CU: 29 seconds.
Patched** AMDRadeonX4100.kext w/ 36 CU: 29 seconds too.

*MacOS Sierra is installed on a WD Blue 1TB HDD, but the master file was exported to Intel 540s 120Gb SSD.
** 36 CU and init function to Ellesmere


I can't get any difference with patches on Clover. Do you apply the reboot without caches and after normal (twice)?
thanks.
 
I've been watching this thread since beginning. GTX 1070 was my primary preference to change my dGPU, but with the progress showed here, decided to get a RX 480 instead of wait for Pascal support from lazy Nvidia. This is my first decent graphics card and I am very satisfied. Finally, I retired my good old Geforce 210.

I have a XFX RX 480 RS 8Gb running on 10.12.2 with latest Clover version 3961. Just set ATI FakeID = 0x67EF1002 and Inject ATI = true to properly work. Using DVI-D cable in a LG IPS236 1080p monitor. Haven't tested HDMI, but it should work fine as usual. My iGPU HD 2500 is connected to same monitor with a HDMI cable.

My motherboard is Gigabyte GA-H61M-DS2H. On stock bios version F5, bios doesn't even load with RX 480. On F7, card is recognized only in legacy mode. Just in the latest and beta bios F8a, which improve high-end display card compatibility, it worked on UEFI VBIOS.

I did the changes proposed by Fl0r!an and Mork_vom_Ork to unlock the full potential from Polaris 10.

What worked to improve performance


- Remove CU limit, change CU number and change init from Baffin to Ellesmere. Improves Luxmark v3.0 in 27,62%! Thanks a lot!!! :clap:
Screen_Shot_2016_12_17_at_00_31_57.png

RX_480_Unlocked_without_Power_Containment_Disabled.png


Unigine has not improved. Score remains around 1240.
image.png


- Changing PP_DisablePowerContainment to 1 improve Luxmark performance a little, but Unigine crash the system between scene 8 and 11(tested three times). The crash happened because, besides it disable the "power containment", it turns off the fan RPM control, that doesn't raise speed, keeping slow RPM as the GPU wasn't heating up. So, temperature rises rapidly(VBIOS limit is 90ºC, 194ºF), the RX 480 screen goes black, but not HD2500(which freezes after this) and the VBIOS make the fan spin at full speed. After this, the only way out is press reset button.

Mystery: If the kext file is changed, the behavior is like described above, but if patched on-the-fly with Clover, the fan control works normally, however performance is decreased(lower than normal).

Kext file changed (Unigine crash)
Power_Containment_Disabled_Kext_Changed.png


Clover on-the-fly patch
Power_Containment_Disabled_Kext_Patch2.png

Power_Containment_Disabled_Kext_Patch.png


What doesn't work

- Injecting Intel HD Graphics 2500, we can get only CI, not QE. The only way to inject without boot to black screen(HD 2500) is deleting AppleIntelHD4000Graphics.kext, leaving just AppleIntelFramebufferCapri.kext loaded, which is needed for full resolution(in this case, can get full resolution even if it doesn't load). But injecting iGPU, even if only the framebuffer loads, it drastically reduces the performance of the RX 480 on Unigine Heaven.

Screen_Shot_2016_12_17_at_09_30_09.png


- Delete AppleIntelFramebufferCapri.kext, because it gets loaded, even without HD 2500 injection. But it made no difference.

- Initialize without the HDMI cable on the HD 2500(Blind boot). Same performance on Unigine and Luxmark. I see a ghost monitor with 1600x900 resolution and if plug the cable into the monitor after the system loads, there is no video signal.

- Initialize without the two cables, DVI and HDMI, and connecting DVI to RX 480. Idem.

- Different FakeID's E0, EF and FF provides same performance(I read several pages ago that switching could improve performance).

- Same performance using legacy or UEFI VBIOS, changing the switch position on the board side.

Bruce X Benchmark 5k (FCPX 10.3.1)*

Original AMDRadeonX4100.kext w/ 16 CU: 29 seconds.
Patched** AMDRadeonX4100.kext w/ 36 CU: 29 seconds too.

*MacOS Sierra is installed on a WD Blue 1TB HDD, but the master file was exported to Intel 540s 120Gb SSD.
** 36 CU and init function to Ellesmere
Could you help me? If I set the values that I mentioned above - meanwhile I figured out where are they and how to edit them - it doesn't take any effect. In system report the device id remain 67DF (470 default id) instead 67EF.
 
Could you help me? If I set the values that I mentioned above - meanwhile I figured out where are they and how to edit them - it doesn't take any effect. In system report the device id remain 67DF (470 default id) instead 67EF.

You should use Clover Configurator.
 
Luxmark score on 10.12.3 beta
Card is a red devil with an overclock on the core to 1370.

In clover I inject ATI with 460 ID
I also applied the 3 kext to patches to unlock power and CU's.


Untitled.jpeg
 
With recent developments, it would be really nice if any of RX460 owners could do a BruceX test.
 
I did.
My config.plist currently looks like this: (Attached)
So the situation is the following right now:
I updated Clover to the latest build. Now it recognizes FakeID and etc but it hangs while booting. The display which is connected by HDMI turns off, the one with DVI-D freezes. Same thing happens with only one of them connected. When the screen goes off, the GPU fan speed goes up to 100% for 1 second.
(The all CU unlocker patches are not even in the boot.plist right now)
Please help me :)
 
I can't get any difference with patches on Clover. Do you apply the reboot without caches and after normal (twice)?
thanks.

I just noticed improvement on Luxmark benchmark, not Unigine.

If you've changed the kexts, you'll must repair permissions and rebuild kext cache. I use Kext Wizard app for this.

Type on terminal(may take a few minutes):
Code:
sudo rm -r /System/Library/Caches/com.apple.kext.caches
sudo touch /System/Library/Extensions && kextcache -u /

This will remove old caches and rebuild them.

Patching kexts with Clover is better, because it'll probably survive future macOS updates and you'll not have to worry after this.

Btw, did not reboot without caches. I wish you luck!
 
Last edited:
So the situation is the following right now:
I updated Clover to the latest build. Now it recognizes FakeID and etc but it hangs while booting. The display which is connected by HDMI turns off, the one with DVI-D freezes. Same thing happens with only one of them connected. When the screen goes off, the GPU fan speed goes up to 100% for 1 second.
(The all CU unlocker patches are not even in the boot.plist right now)
Please help me :)

If you are using the latest Clover, you may be working at cross purposes ... in that the newer Clover versions have a lot of the info so that you do not have to declare it as a fake version.

But for the moment, can you get to the EFI and main partition of your drive? On the main partition along with System and Applications, etc. you probably have a folder called ---> EFI-Backups

Inside of this you will find copies of each version of Clover you have installed along with all the necessary stuff. Decide which version you had that was working, open your EFI partition and copy (boot file and EFI folder) to the same places on your EFI partition. Reboot and you would be back to where you started when it was working. BEFORE you start make sure you are familiar with where everything is. Take a couple of looks (it is easy to do, but getting in a nervous hurry has consequences).

I do not know that you need the FixDisplay_0100 key or the FBName key any longer with all of the recent Clover changes and the latest Ellesmere patching, but Gigamaxx and the others can give you a better idea about that.
 
Status
Not open for further replies.
Back
Top