GPU Performance Issue, (Somewhat better after sleep)

[walidur234]

I'm extremely frustrated with my GPU performance.

I've been working on a project on Premiere Pro 2023( latest version), and it's a struggle to play 1080p footage smoothly. let alone effects, color grade applied footage. So I tried to benchmark my Hackintosh, and the GPU perform very poorly. in Mojave, Catalina I used RadeonBoost.kext which help to perform better since I switch to Monterey, especially in Ventura I noticed a major performance drop (I didn't run the GPU benchmark in Monterey). I can't play a 4k video on youtube smoothly, it's making the whole mac laggy.

Here are Geekbench [Version 5.4.5 (503938)] results:

Spoiler: GEEKBENCH 5 - ALGREY'S PATCH

Spoiler: GEEKBENCH 5 - SHANEEE'S PATCH

Spoiler: GEEKBENCH 5 - SHANEEE'S PATCH [After Sleep]

System Spec:
CPU: AMD RYZEN 3700x
MOBO: MSI b450m Mortar max
GPU: SAPPHIRE NITRO+ Radeon RX 580 8GB
Ram: Corsair 8+8+16 /3200
PSU: Corsair 550/650watt (I forget the actual rating)

OS: macOS Ventura 13.0.1 (22A400) | Opencore 0.8.6
SMBios: MacPro7,1


EDIT: Bluetooth need to be turned off and on after sleep, otherwise won't work.

 

[Edhawk]

@walidur234 the IOReg is not openable. I see this message when I try to open it.



Can you download the IORegistryExplorer 2.1 dmg from this Github site and create a new copy of your IOReg.

GitHub - vulgo/IORegistryExplorer: Apple IORegistryExplorer 2.1

Apple IORegistryExplorer 2.1. Contribute to vulgo/IORegistryExplorer development by creating an account on GitHub.

github.com

 

[Edhawk]

@Jo-Toku which graphics card and system are you using?

Those scores are similar to the scores I get from a GTX 780Ti, not an AMD RX580.

 

[walidur234]

@walidur234 the IOReg is not openable. I see this message when I try to open it.

View attachment 8733

Can you download the IORegistryExplorer 2.1 dmg from this Github site and create a new copy of your IOReg.

GitHub - vulgo/IORegistryExplorer: Apple IORegistryExplorer 2.1

Apple IORegistryExplorer 2.1. Contribute to vulgo/IORegistryExplorer development by creating an account on GitHub.

github.com

ok, here's the file

 

[Edhawk]

OK, there is something strange going on with your USB configuration.

UTBMap.kext/Contents/Info.plist - showing the 2 x USB controllers and 15 ports you have activated.

None of the ports are set with the connector type (255) which would need to be applied to a connector for a Bluetooth module.

The IOReg shows your Bluetooth module is connected to PO12 on the PTXH controller, and it is set with the Connector Type (3) which should only be used for a USB3 port connector.

Only 2 x USB ports on the XHC0 controller are showing as having anything connected.

1. AppleUSB20XHCIPort@60100000 - USB Keyboard
2. AppleUSB20XHCIPort@60300000 - USB Gaming Mouse

None of the other possible 6 x USB ports on the XHC0 connector are showing anything connected.


None of the possible 14 x USB ports under the PTXH controller are showing because the ports in the UTBMap.kext have been set with the wrong port name. You have used HS01-HS06 and SS01-SS03. When the ports should be named POT1-POT9 and PO10-PO14. As shown in the screenshot from your IOReg below.

PTXH USB controller showing port names, and BRCM20702 Bluetooth module connected to port PO12.

As you have 2 x USB controllers PTXH and XHC0, you are able to activate up to 15 ports on EACH controller.

You do not have to limit yourself to a total of 15 ports across both controllers. That is a common Intel limitation, it doesn't apply when you have more than 1 x USB controller.

The screenshot below shows what I would expect the Port naming convention to look like for your PTXH USB controller.

Only the PO12 connector type is known to be correct, as it activates your Bluetooth module.

When you look at creating a new UTBMap.kext, which you need to do. Ask for help if you are unsure about what you are doing.

Your current UTBMap is not working, so I would remove it and see what difference it makes to how your ports are defined.
I would then follow the USB configuration guide over at Dortania and see if a better and correct solution can be created.

Why should you USB map | OpenCore Post-Install

 

[Edhawk]

Now to review your Display connection!

These are the details for your current display connection, as shown in your IOReg.
ATY, AMD, RadeonFramebuffer@3

This tells us you are using a HDMI connector from connector 3 (4th port) on your RX580.
The IOReg also tells us your system is using the correct AMD9500Controller.kext.

What it also tells us is your Display is not registering any Data against the connector.

In comparison, my Sapphire Nitro+ RX580, using the same Framebuffer, but with a DisplayPort connector shows the following:


While the av-signal-type and device-type are different, HDMI vs DP, the rest should be very similar.

So from this I read that something is lacking or preventing the display data from being generated.


Power Management & Radeon Sensor kexts.
One major difference between our setups is that I am not using the AGPMInjector.kext or RadeonSensor.kext. Both of these kexts are Kernel extensions with executable code. This can have an impact on how your system runs.

Have you tried booting the system with the AGPMInjector.kext and/or RadeonSensor.kext disabled?
What are the GeekBench scores like with these two kexts removed or disabled?


Optional SSDT.
Also as your dGPU works better after a sleep, it could be that the initialisation of your RX580 is not working from a cold boot.
Previously we used a Clover option or SSDT-Radeon-Deinit.aml table to fix this issue. For this fix we need the ACPI path of the RX580. Looking in your IOReg it shows the following:

IOService:/AppleACPIPlatformExpert/PCI0@0/AppleACPIPCI/GPP8@3,1/IOPP/GFX0@0/ATY,AMD,RadeonFramebuffer@0/AMDFramebufferVIB/AGPM

The ACPI path above, converts to (SB.PCI0.GPP8.GFX0) when looking to create or edit an SSDT for the card.

I have attached an edited copy of the SSDT-Radeon-Deinit-GPP8-GFX0.aml for you to try, if disabling or removing the two kexts doesn't help improve the GB scores.

 

[walidur234]

Now to review your Display connection!

These are the details for your current display connection, as shown in your IOReg.
View attachment 8750 ATY, AMD, RadeonFrameBuffer@3

This tells us you are using a HDMI connector from connector 3 (4th port) on your RX580.
The IOReg also tells us your system is using the correct AMD9500Controller.kext.

What it also tells us is your Display is not registering any Data against the connector.

In comparison, my Sapphire Nitro+ RX580, using the same Framebuffer, but with a DisplayPort connector shows the following:
View attachment 8751

While the av-signal-type and device-type are different, HDMI vs DP, the rest should be very similar.

So from this I read that something is lacking or preventing the display data from being generated.


Power Management & Radeon Sensor kexts.
One major difference between our setups is that I am not using the AGPMInjector.kext or RadeonSensor.kext. Both of these kexts are Kernel extensions with executable code. This can have an impact on how your system runs.

Have you tried booting the system with the AGPMInjector.kext and/or RadeonSensor.kext disabled?
What are the GeekBench scores like with these two kexts removed or disabled?


Optional SSDT.
Also as your dGPU works better after a sleep, it could be that the initialisation of your RX580 is not working from a cold boot.
Previously we used a Clover option or SSDT-Radeon-Deinit.aml table to fix this issue. For this fix we need the ACPI path of the RX580. Looking in your IOReg it shows the following:

IOService:/AppleACPIPlatformExpert/PCI0@0/AppleACPIPCI/GPP8@3,1/IOPP/GFX0@0/ATY,AMD,RadeonFramebuffer@0/AMDFramebufferVIB/AGPM

The ACPI path above, converts to (SB.PCI0.GPP8.GFX0) when looking to create or edit an SSDT for the card.

I have attached an edited copy of the SSDT-Radeon-Deinit-GPP8-GFX0.aml for you to try, if disabling or removing the two kexts doesn't help improve the GB scores.

I have tried with disabled these kexts (AGPMInjector.kext or RadeonSensor.kext) and later removed these kexts, but the same low GB score.

So I have installed the SSDT you have attached, and now the GB score is awesome

 

[walidur234]

OK, there is something strange going on with your USB configuration.

View attachment 8746 UTBMap.kext/Contents/Info.plist - showing the 2 x USB controllers and 15 ports you have activated.

None of the ports are set with the connector type (255) which would need to be applied to a connector for a Bluetooth module.

The IOReg shows your Bluetooth module is connected to PO12 on the PTXH controller, and it is set with the Connector Type (3) which should only be used for a USB3 port connector.

Only 2 x USB ports on the XHC0 controller are showing as having anything connected.

1. AppleUSB20XHCIPort@60100000 - USB Keyboard
2. AppleUSB20XHCIPort@60300000 - USB Gaming Mouse

View attachment 8748

None of the other possible 6 x USB ports on the XHC0 connector are showing anything connected.


None of the possible 14 x USB ports under the PTXH controller are showing because the ports in the UTBMap.kext have been set with the wrong port name. You have used HS01-HS06 and SS01-SS03. When the ports should be named POT1-POT9 and PO10-PO14. As shown in the screenshot from your IOReg below.

View attachment 8747 PTXH USB controller showing port names, and BRCM20702 Bluetooth module connected to port PO12.

As you have 2 x USB controllers PTXH and XHC0, you are able to activate up to 15 ports on EACH controller.

You do not have to limit yourself to a total of 15 ports across both controllers. That is a common Intel limitation, it doesn't apply when you have more than 1 x USB controller.

The screenshot below shows what I would expect the Port naming convention to look like for your PTXH USB controller.

View attachment 8749 Only the PO12 connector type is known to be correct, as it activates your Bluetooth module.

When you look at creating a new UTBMap.kext, which you need to do. Ask for help if you are unsure about what you are doing.

Your current UTBMap is not working, so I would remove it and see what difference it makes to how your ports are defined.
I would then follow the USB configuration guide over at Dortania and see if a better and correct solution can be created.

Why should you USB map | OpenCore Post-Install

Finally, GPU and USB has fixed. At first, I tried the Dortania USB guide but unfortunately, maybe I messed up and made a wrong USB-map.kext(which work with 1USB port). Later on, tried with the Hackintool, and made a USBPort.kext. It works as expected, and Bluetooth automatically works after sleep.

Again, I've attached the IOReg file, Kext and OC kext(ss).
Thank You a lot @Edhawk

*Without AppleMCEReporterDisabler.kext System won't boot (MacPro7,1 smbios).

 

[walidur234]

Further testing the GPU, I figure out that If I open any of Adobe's products and close it/or not it makes GPU perform less otherwise it's performing as expected.
I tried to close all Adobe products (after opening) with Activity monitor, and run the GB but same low score. I think Adobe is the main culprit, what do you think?

 

[Jo-Toku]

@Jo-Toku which graphics card and system are you using?

Those scores are similar to the scores I get from a GTX 780Ti, not an AMD RX580.

Radeon RX 580. Is the full Desktop card jammed into a laptop but tgp is regulated for thermals. For reference under Windows 10 I get 40+ k on OpenCL.

I am on a project that is kicking my ass. As soon as I have more time I will upload my EFI

Basic Specs of the laptop

CPU: Intel I7 7700hq

GPU: RX 580

Display: 17" 1080p

HDD: 1TB NVMW - Crucial

HDD2: HP's 2.5 drive disconnected, it was bugging down my boot up with Mac OS

Memory: 24GB Dddr4

 

[Edhawk]

@walidur234 Adobe apps not working well on an AMD Hack is not unusual as the macOS versions are looking for an Intel CPU. See the thread linked below.

AMD Adobe Fixes

Below you'll find a list of the AMD Adobe commands to fix some applications. Usage: Open terminal and run the below commands as required. Note: File path changes in each version release. So if the file is not found you can find the specified file manually and replace the path in the below...

forum.amd-osx.com

 

[Jo-Toku]

Here is my EFI.

Basic Specs of the laptop

CPU: Intel I7 7700hq

GPU: RX 580

Display: 17" 1080p

HDD: 1TB NVMW - Crucial

HDD2: HP's 2.5 drive disconnected, it was bugging down my boot up with Mac OS

Memory: 24GB Dddr4

 

[Edhawk]

You do recall that even natively supported dGPU‘s don’t work in laptops with switchable graphics, yes? This is the same for Intel and AMD laptops.

Does your HP have a Mux switch?

 

[tomnic]

@walidur234 Adobe apps not working well on an AMD Hack is not unusual as the macOS versions are looking for an Intel CPU. See the thread linked below.

AMD Adobe Fixes

Below you'll find a list of the AMD Adobe commands to fix some applications. Usage: Open terminal and run the below commands as required. Note: File path changes in each version release. So if the file is not found you can find the specified file manually and replace the path in the below...

forum.amd-osx.com

Adobe apps work perfectly with AMD hackintoshes, you pointed to a very old thread. Just use AMDFriend following this link:

Release Version 1.0.2 · NyaomiDEV/AMDFriend

Changes Nothing much. I am now sure it works well in Ventura, especially if you have screwed up environments thanks to Homebrew (dang it). Oh also now there's a binary release. Grab it below!

github.com

Detailed info here: https://www.macos86.it/topic/5489-tutorial-for-patching-binaries-for-amd-hackintosh-compatibility/

 

[Jo-Toku]

You do recall that even natively supported dGPU‘s don’t work in laptops with switchable graphics, yes? This is the same for Intel and AMD laptops.

Does your HP have a Mux switch?

This laptop doesnt have optimus, it only uses the Radeon. No Mux required. The Legion's 6600m and the Area 51m with 5700m dgpu work at full speed, however it could be due to Monterey?

 

[walidur234]

Adobe apps work perfectly with AMD hackintoshes, you pointed to a very old thread. Just use AMDFriend following this link:

Release Version 1.0.2 · NyaomiDEV/AMDFriend

Changes Nothing much. I am now sure it works well in Ventura, especially if you have screwed up environments thanks to Homebrew (dang it). Oh also now there's a binary release. Grab it below!

github.com

Detailed info here: https://www.macos86.it/topic/5489-tutorial-for-patching-binaries-for-amd-hackintosh-compatibility/

Yes, I use "NyaomiDEV/AMDFriend" patch.
But if run Adobe's products and then close it/or not it makes GPU perform less (Geekbench score: ~36k[metal] ) otherwise it's performing as expected(Geekbench score: ~55k[metal] ). I tried to close all Adobe products (after opening) with Activity monitor, and run the GB but same low score.

 

[Edhawk]

@Jo-Toku As your laptop has an AMD discrete GPU it wouldn't have Nvidia's Optimus.

That doesn't mean your HP laptop doesn't pass the dGPU to the external port, as most laptops with Switchable graphics do.

You will need to look a bit deeper to clarify if the Intel IGPU and AMD dGPU are truly separate and if they can be used independently in macOS or for that matter in Windows.

What is the exact Model number and name of your HP laptop?

 

[Jo-Toku]

@Jo-Toku As your laptop has an AMD discrete GPU it wouldn't have Nvidia's Optimus.

That doesn't mean your HP laptop doesn't pass the dGPU to the external port, as most laptops with Switchable graphics do.

You will need to look a bit deeper to clarify if the Intel IGPU and AMD dGPU are truly separate and if they can be used independently in macOS or for that matter in Windows.

What is the exact Model number and name of your HP laptop?

The model number is: 17-AN012DX


Here is a link. https://www.bestbuy.com/site/omen-b...n-fiber/5902586.p?skuId=5902586&intl=nosplash

I also get the same issue with Mojave and my Acer Helios 500. The performance is half with the Vega 56. The Helios laptop uses the desktop CPU Ryzen 7 2700 (No IGPU with the CPU). The Vega 56 is the same sku in the Apple Imac Pro. There is an injector that allows to overclock it within OSX but is not stable. The Helios is my daily driver, but is at the shop for the moment.

Definitely recommend to snag one while you can find them. Even at half the performance of the Vega (until I can find a solution that is stable) still very powerful. at full power it gets just a tad lower than my 5700m

 

[Edhawk]

The model number is: 17-AN012DX

Do you dual boot macOS and Windows 10/11 on this system?

If yes, when you access the AMD Radeon graphics app does it display the same options as is seen in this link about AMD switchable graphics?

https://www.amd.com/en/support/kb/faq/dh-017

 

[Jo-Toku]

Do you dual boot macOS and Windows 10/11 on this system?

If yes, when you access the AMD Radeon graphics app does it display the same options as is seen in this link about AMD switchable graphics?

https://www.amd.com/en/support/kb/faq/dh-017

I have windows 10. The IGP HD630 is not accessible under windows.

 

[Edhawk]

The IGP HD630 is not accessible under windows.

Is that due to a Bios setting? As that is not normal for a HP Windows laptop.

I was looking around at the specs for your HP and I can't find any information regarding the graphics being Hybrid/switchable etc., other than it contains a discrete RX580 8GB.

The Kaby Lake CPU definitely contains a HD630 IGPU, as it isn't an 'F' series processor.