Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Status
The first post of this thread is a WikiPost and can be edited by anyone with the appropiate permissions. Your edits will be public.
This mod sure won't able to activate any Nvidia card's hwaccel. That GTX650 has nothing to do with hwaccel, just cosmetic error same as "N/A" if no other GPU installed.

But it's good to know that now allow 3,1 to watch some demanding videos that was not playable, also has much better H264 encoding speed.
The goal of the post was to show that while using both cards doesn't let the RX 580 show video but the computer is able to access it for hardware encoding. So if you need boot screens and an 'encoder' card you could get by with having both cards installed and just not use the RX580 for any display attachments. But the display won't get any accelerated decoding from the RX card as the display would be on the GTX card. You might be able to get accelerated decode from handbrake for example while it is hardware encoding.

When it is a solo RX580 card its great, just lacks boot screen (expected).
[doublepost=1560722920][/doublepost]
Airplay not working is a known issue. This function require HEVC hardware encoding, which we still have no idea how to get it work yet.
It is curious though that the non-apple iMacPro1,1 is able to airplay without HEVC encode claiming that it works.
Maybe that is just a side effect of it being a hackintosh or a VideoProc bug?
 
  • Like
Reactions: h9826790
The goal of the post was to show that while using both cards doesn't let the RX 580 show video but the computer is able to access it for hardware encoding. So if you need boot screens and an 'encoder' card you could get by with having both cards installed and just not use the RX580 for any display attachments. But the display won't get any accelerated decoding from the RX card as the display would be on the GTX card. You might be able to get accelerated decode from handbrake for example while it is hardware encoding.

When it is a solo RX580 card its great, just lacks boot screen (expected).

You can get an alternative boot screen on 3,1 via RX580.

There is a customised rEFInd which allow RX580 to display via UEFI GOP.

https://forums.macrumors.com/thread...-output-protocol.2089181/page-7#post-26378798
[doublepost=1560723162][/doublepost]
It is curious though that the non-apple iMacPro1,1 is able to airplay without HEVC encode claiming that it works.
Maybe that is just a side effect of it being a hackintosh or a VideoProc bug?

I am not 100% sure if the iMac Pro has HEVC hardware encoding or not. My understanding is YES, it has HEVC hardware encoding via the Vega.

BTW, may I know what is non-Apple iMac Pro?
 
You can get an alternative boot screen on 3,1 via RX580.

There is a customised rEFInd which allow RX580 to display via UEFI GOP.

I am not 100% sure if the iMac Pro has HEVC hardware encoding or not. My understanding is YES, it has HEVC hardware encoding via the Vega.

BTW, may I know what is non-Apple iMac Pro?
A Hackintosh ;) the rx 580 I’ve been swapping between the “iMacPro1,1 smbios” Hackintosh and my 3,1 for testing before I buy another card. When the 580 is in the hack it has working airplay, when it is in the 3,1 it doesn’t work.

Is there a way to do more to pretend a cMP is an iMac Pro and get more of those features? Perhaps leveraging Clover to do it?

I’ll have to look into the GOP thread more. Thanks.
 
A Hackintosh ;) the rx 580 I’ve been swapping between the “iMacPro1,1 smbios” Hackintosh and my 3,1 for testing before I buy another card. When the 580 is in the hack it has working airplay, when it is in the 3,1 it doesn’t work.

Is there a way to do more to pretend a cMP is an iMac Pro and get more of those features? Perhaps leveraging Clover to do it?

I’ll have to look into the GOP thread more. Thanks.
Watch out with Clover. Unless you have a way to recover/reflash your chip or something similar to a MATT card do not try it. Maybe opencore will provide safer alternative. Or you can try virtual machine and GPU pass-trough, though 3.1 lacks SSE4.2 I believe.

Actually for 3.1 it might be safe as @DearthnVader has been using clover on his machine
 
  • Like
Reactions: Ludacrisvp
A Hackintosh ;) the rx 580 I’ve been swapping between the “iMacPro1,1 smbios” Hackintosh and my 3,1 for testing before I buy another card. When the 580 is in the hack it has working airplay, when it is in the 3,1 it doesn’t work.

Is there a way to do more to pretend a cMP is an iMac Pro and get more of those features? Perhaps leveraging Clover to do it?

I’ll have to look into the GOP thread more. Thanks.

I see. We never really change the cMP to iMacPro definition. This is the fundamental difference.

I used Clover on my 5,1 before, it works. However, I really don't know how bad it can corrupt the BootROM. Really won't recommend anyone doing this (unless you can hardware flash a clean BootROM to recover the logicboard.

At this moment, the safest way I believe is QEMU + PCIe pass through. However, I think that's too extreme if all we get is just HEVC hardware encoding.
 
  • Like
Reactions: Reindeer_Games
So took a look at the Console>System Reports and see Kernal gpuRestart errors.

Mon Jun 17 07:52:21 2019

Event: GPU Reset
Date/Time: Mon Jun 17 07:52:21 2019
Application: Final Cut Pro
Path:
Tailspin: /Library/Logs/DiagnosticReports/gpuRestart2019-06-17-075221.tailspin
GPUSubmission Trace ID: 0
OS Version: Mac OS X Version 10.14.5 (Build 18F132)
Graphics Hardware: Radeon RX 580
Signature: 0

Report Data:

GPU Log Version: 1

Restart Channel: 7 UVDDec


Some sort of driver issue? In this morning's case.. the sound continues to play, but I'm unable to stop it and regain control of the app or system. I did the Command + CTRL + press the power button to shutdown.
 
So took a look at the Console>System Reports and see Kernal gpuRestart errors.

Mon Jun 17 07:52:21 2019

Event: GPU Reset
Date/Time: Mon Jun 17 07:52:21 2019
Application: Final Cut Pro
Path:
Tailspin: /Library/Logs/DiagnosticReports/gpuRestart2019-06-17-075221.tailspin
GPUSubmission Trace ID: 0
OS Version: Mac OS X Version 10.14.5 (Build 18F132)
Graphics Hardware: Radeon RX 580
Signature: 0

Report Data:

GPU Log Version: 1

Restart Channel: 7 UVDDec


Some sort of driver issue? In this morning's case.. the sound continues to play, but I'm unable to stop it and regain control of the app or system. I did the Command + CTRL + press the power button to shutdown.

My guess would be Driver related.
On the 3rd i'm not sure what i was doing but I had 20 of these logged between 15:28 and 15:31 with all of them coming from the Messages application.

Code:
Mon Jun  3 15:31:42 2019

Event:               GPU Reset
Date/Time:           Mon Jun  3 15:31:42 2019
Application:         Messages
Path:               
Tailspin:            /Library/Logs/DiagnosticReports/gpuRestart2019-06-03-153142.tailspin
GPUSubmission Trace ID: 0
OS Version:          Mac OS X Version 10.14.5 (Build 18F132)
Graphics Hardware:   PowerColor Red Devil RX 580
Signature:           0

Report Data:

GPU Log Version: 1

Restart Channel: 7 UVDDec
 
  • Like
Reactions: Chung123
So took a look at the Console>System Reports and see Kernal gpuRestart errors.

Mon Jun 17 07:52:21 2019

Event: GPU Reset
Date/Time: Mon Jun 17 07:52:21 2019
Application: Final Cut Pro
Path:
Tailspin: /Library/Logs/DiagnosticReports/gpuRestart2019-06-17-075221.tailspin
GPUSubmission Trace ID: 0
OS Version: Mac OS X Version 10.14.5 (Build 18F132)
Graphics Hardware: Radeon RX 580
Signature: 0

Report Data:

GPU Log Version: 1

Restart Channel: 7 UVDDec


Some sort of driver issue? In this morning's case.. the sound continues to play, but I'm unable to stop it and regain control of the app or system. I did the Command + CTRL + press the power button to shutdown.


I own a MP6.1 and i had about 4 years of trouble with the GPU restart error - exactly the same symptoms as you describe - UI freezes but sound continues to play requiring a force reboot.

There is a huge thread about this issue on this forum, eventually it was fixed in a 10.14. OS update from Apple so I assume it is a driver related problem?
 
I have the same:
Sun Jun 16 23:20:01 2019

Event: GPU Reset
Date/Time: Sun Jun 16 23:20:01 2019
Application: Final Cut Pro
Path:
Tailspin: /Library/Logs/DiagnosticReports/gpuRestart2019-06-16-232001.tailspin
GPUSubmission Trace ID: 0
OS Version: Mac OS X Version 10.14.5 (Build 18F132)
Graphics Hardware: Radeon RX 580
Signature: 0

Report Data:

GPU Log Version: 1

Restart Channel: 7 UVDDec

---THE STATE OF THE DRIVER---

AMDRadeonX4000_AMDEllesmereGraphicsAccelerator PCIe Device: [5:0:0] State: ENABLED
TotalVideoRAMBytes: 0x0000000200000000 (8589934592)​
....

I'll probably go and replace the RX580 with Sapphire pulse Vega 56 - I see no reported problems on that one, but what I understand is that for it to work in Mac Pro I need to do the Pixas mod?
 
I have the same:
Sun Jun 16 23:20:01 2019

Event: GPU Reset
Date/Time: Sun Jun 16 23:20:01 2019
Application: Final Cut Pro
Path:
Tailspin: /Library/Logs/DiagnosticReports/gpuRestart2019-06-16-232001.tailspin
GPUSubmission Trace ID: 0
OS Version: Mac OS X Version 10.14.5 (Build 18F132)
Graphics Hardware: Radeon RX 580
Signature: 0

Report Data:

GPU Log Version: 1

Restart Channel: 7 UVDDec

---THE STATE OF THE DRIVER---

AMDRadeonX4000_AMDEllesmereGraphicsAccelerator PCIe Device: [5:0:0] State: ENABLED
TotalVideoRAMBytes: 0x0000000200000000 (8589934592)​
....

I'll probably go and replace the RX580 with Sapphire pulse Vega 56 - I see no reported problems on that one, but what I understand is that for it to work in Mac Pro I need to do the Pixas mod?

Vega 56 most likely can stay within the cMP's power envelope (mini 6pin only), especially if you mainly use it for FCPX (not that demanding).

The key is to balance the power draw. e.g. do NOT use single mini 6pin -> single 8pin. But DUAL mini 6pin -> single 8pin -> DUAL 6+2pin (2 cables combination).
 
Hi ! for compatibility reasons I need to stick with High Sierra. Does hardware acceleration work natively with 10.13.6 ? If not can I use this method ?
 
exports a bit faster
I am seeing roughly 2x faster. Admittedly, not a complete apples-to-apples comparison, but on the same MP/x5680/32GB/RX580, I have an SSD with 10.13.6 and another with 10.14.5.
10.14.5 has @h9826790 's lilu accelerator adjustment. On the FreeSolo mp4, it generally performed at 250-330 fps with HB set to use h.264 VideoToolkit. About 27 min to complete.
10.13.6 with stock drivers, and the same video, but set to plain h.264, HB performed at 120-180 fps. About 52 min to complete.
HB_FreeSolo-VT.pngFreeSolo_HB13.6_noVTK.png
 
Last edited:
Someone may or may not be interested in this but did a quick benchmark between my 'iMacPro1,1' hackintosh and my real 3,1. This seems relevant as I'm sure there are still going to be those that are more strongly considering a hack replacement of their older mac due to the cost of the 7,1.

The hack has a lot more power on paper but for this video encoding test between them they aren't that far apart.
In my opinion this modification for HW accel + netkas's SSE4.2 patch really allows the 3,1 to become relevant again.

Sure, this isn't a 1:1 true test due to different video cards.

hack = RedDevil 580 8Gb in OC mode, Single Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz (6c/12 threads) w/ 32Gb DDR4 2133 ECC
3,1 = Sapphire Pulse 580 4Gb which claims "OC" on box, Dual Intel(R) Xeon(R) CPU X5482 @ 3.20GHz w/ 32Gb DDR2 800 FB ECC

Both read/write from SATA3 SSD, with the 3,1 doing it via the onboard SATA2 chipset.

I used handbrake to re-encode the same video file from h264 to h264, there was only about 35fps difference between the machines. The source video was 16:9 at 1920x1080 being cropped to 4:3 with 1440x1080 resolution (removing black pillars), 45 min duration and 6.6Gb in size (very high bitrate), output was 875.5Mb. Audio was 2 channel stereo AAC in passthru mode.

In the screen shot I was using screensharing and the real mac is the background / bottom progress bar with the hack on the top. During encode CPU was doing about 250% and GPU processor was about 33% on the hack, but on the 3,1 the CPU was about 450% and GPU processor was about 95%.
Screen Shot 2019-06-21 at 10.43.41 AM.jpg


Hack:
Screen Shot 2019-06-21 at 11.16.59 AM.jpg


3,1:
Screen Shot 2019-06-21 at 11.16.24 AM.jpg


Compare that to what CPU was able to do...
Using just over 1000% CPU on the hack and about 750% CPU on the 3,1
(only change was output filename and video encoder from toolbox to CPU)
Screen Shot 2019-06-21 at 11.08.07 AM.jpg


Using hardware acceleration from this thread allowed both machines to be faster by 100fps.
I'm sure other settings may reveal different results (or other source material), but this was an easy group of real data points to share.
 
Oh well, it was good while it lasted...

On my MP4,1->5,1 with RX580 running macOS 10.14.5, I got smooth playback of 4k video, very speedy encoding in apps like MacX Video Converter and approximately 3x faster exports from iMovie... only that they systematically hung at the end. I've disabled the nvram arguments for now. Any hope that this gets fixed/stabilized at one point?
 
  • Like
Reactions: Chung123
Oh well, it was good while it lasted...

On my MP4,1->5,1 with RX580 running macOS 10.14.5, I got smooth playback of 4k video, very speedy encoding in apps like MacX Video Converter and approximately 3x faster exports from iMovie... only that they systematically hung at the end. I've disabled the nvram arguments for now. Any hope that this gets fixed/stabilized at one point?
If I interpret the recent posts around this correctly it sounds like the RX580 is more stable if you do the 'alternate method' (hex editor) vs using Lilu+WEG+nvram args.
There hasn't been any reports of instability / freezing with Vega or newer using HW accel, so if you want to buy a newer GPU ... :)
 
  • Like
Reactions: Chung123
If I interpret the recent posts around this correctly it sounds like the RX580 is more stable if you do the 'alternate method' (hex editor) vs using Lilu+WEG+nvram args.
There hasn't been any reports of instability / freezing with Vega or newer using HW accel, so if you want to buy a newer GPU ... :)


I'll have to explore that 'alternate method' then. The original post made me think it was limited to macOS 10.13 though.

Regarding Vega... I guess at some point I'll have to stop spending money on this 10y-old cMP ;), and honestly, the RX580 has all the power I really need.
 
Oh well, it was good while it lasted...

On my MP4,1->5,1 with RX580 running macOS 10.14.5, I got smooth playback of 4k video, very speedy encoding in apps like MacX Video Converter and approximately 3x faster exports from iMovie... only that they systematically hung at the end. I've disabled the nvram arguments for now. Any hope that this gets fixed/stabilized at one point?
I'll have to explore that 'alternate method' then. The original post made me think it was limited to macOS 10.13 though.

Regarding Vega... I guess at some point I'll have to stop spending money on this 10y-old cMP ;), and honestly, the RX580 has all the power I really need.

The alternate method should be more stable for RX580, but still not 100% stable.

IMO, it isn't that hard to buy a low cost used Vega 56. And if you sell the RX580, the overall extra cost should be quite low. And since no Vega users reported any freeze. If you want to utilise this function. Get a used Vega card should be one of the good option.

But anyway, I also suggest you try the alternate method first.
 
  • Like
Reactions: alchemistics
For info, just tested 10.14.6 Beta (18G48f), hwaccel still works, nothing more, nothing less. But not sure about stability yet (mainly for RX580).

There is no update of the AppleGVA.framework, so, no need to re-do any file mod (if using the alternate procedures).
Screenshot 2019-06-22 at 4.33.26 PM.png
 
Just found out a relative simple (and clear) way to check if you are really using hardware decoding / encoding (not just activated, but are you really using it at that particular moment).

Open Terminal and enter (at the moment that your GPU is actually performing some video task)
Code:
ioreg -l |grep \"PerformanceStatistics\" | cut -d '{' -f 2 | tr '|' ',' | tr -d '}' | tr ',' '\n'|grep 'HW\|Command'

Here are some examples.

1) Nothing is using hardware encoding / decoding. Everything associated with VCE or UVD is 0.
HWAccel not in use.png



2) Both hardware encoding and decoding are working (transcoding a HEVC video into H264 in FFMpeg)
FFMpeg HEVC to H264.png


Commands Submitted / Completed associated with some VCE and UVD channels are not at 0
FFMpeg HEVC to H264 VideoToolBox copy.png



3) Only hardware encoder is working. In this case, I use Handbrake to convert the same video (in Example 2) from HEVC to H264.
Screenshot 2019-06-24 at 1.53.50 AM.png


However, due to Handbrake unable to utilise the HEVC hardware decoder. So, only H264 hardware encoder was working. Some VCE Channels shows having activity. But all UVD Channels stay at 0 command.
Handbrake H264 VideoToolBox copy.png


Noted: Due to the very demanding HEVC source video, even my CPU usage was way higher in case 3 (to assist the transcoding), but performance still about 30% lower than case 2. However, in some case, if the CPU can decode fast enough to feed the GPU, intentionally only use hardware encoding may have better performance.

4) Only hardware decoder is working (Preview HEVC video in Finder). Only some UVD channels receive commands.
Preview HEVC copy.png
 
Last edited:
Just found out a relative simple (and clear) way to check if you are really using hardware decoding / encoding (not just activated, but are you really using it at that particular moment).

Open Terminal and enter
Code:
ioreg -l |grep \"PerformanceStatistics\" | cut -d '{' -f 2 | tr '|' ',' | tr -d '}' | tr ',' '\n'|grep 'HW\|Command'

Hi, I am really confused after this post.
I have Mac Pro 5.1 with Sapphire RX Vega 64 (reference one). Few days ago I activated hardware acceleration wit Lilu and WhateverGreen kext. Exactly same way as it on first post. I was not quite easy because I had to repeat it five times and after fifth attempt it starts working.
Judging by VideoProc I thought that it is working correctly. FPC X export is much faster and GPU is much more stress (but not for 100%).
I have to admit that Compressor is not using GPU... And It looks like that there is no change for compressor.
So after post post #273 I tray new way how to check if it works. And base on that it doesn't but base on VideoProc it does. Any idea where is problem?

Thanks so much for any help.
 
Hi, I am really confused after this post.
I have Mac Pro 5.1 with Sapphire RX Vega 64 (reference one). Few days ago I activated hardware acceleration wit Lilu and WhateverGreen kext. Exactly same way as it on first post. I was not quite easy because I had to repeat it five times and after fifth attempt it starts working.
Judging by VideoProc I thought that it is working correctly. FPC X export is much faster and GPU is much more stress (but not for 100%).
I have to admit that Compressor is not using GPU... And It looks like that there is no change for compressor.
So after post post #273 I tray new way how to check if it works. And base on that it doesn't but base on VideoProc it does. Any idea where is problem?

Thanks so much for any help.

Just checked Compressor can utilise both VCE and UVD
Compressor copy.png


This is the encode setting I use
Screenshot 2019-06-24 at 3.43.28 AM.png


And you MUST run that terminal command DURING encoding. It's a command to check if the hardware acceleration engine is actually working at that moment. Video Proc is to check if hardware acceleration is available at that moment.
Screenshot 2019-06-24 at 3.44.00 AM.png
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.