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.
MP 1,1-5,1 Activate AMD hardware acceleration
- Thread starter h9826790
- WikiPost WikiPost
- Start date
- Sort by reaction score
- Status
First of all thanks for the work you do here.
I’m running a cMP 4,1 > 5,1 with a MSI RX Vega 56 Air Boost OC. It’s powered by two mini-6-Pin-to-6-Pin cables connected to a EVGA PowerLink and running perfectly fine (BIOS 2).
I applied the hex-edit-method under 10.14.5, what worked fine regarding video playback. Only when I tried to connect a 4K screen, the UI of Mac OS hung every second for some frames.
After the update to 10.14.6 the hex-edit was disabled. When I tried the 4K screen now, the UI ran smoothly.
Might this behavior be frame buffer related?
I’m running a cMP 4,1 > 5,1 with a MSI RX Vega 56 Air Boost OC. It’s powered by two mini-6-Pin-to-6-Pin cables connected to a EVGA PowerLink and running perfectly fine (BIOS 2).
I applied the hex-edit-method under 10.14.5, what worked fine regarding video playback. Only when I tried to connect a 4K screen, the UI of Mac OS hung every second for some frames.
After the update to 10.14.6 the hex-edit was disabled. When I tried the 4K screen now, the UI ran smoothly.
Might this behavior be frame buffer related?
Is it possible to provide the exact steps about how to reproduce the issue?
[doublepost=1563982502][/doublepost]
The hex edited AppleGVA.framework will be replaced during 10.14.6 update. You have to apply the mod again (if you haven't backup the modded file).
I am not sure if that's framebuffer related. From memory, you are the first one to report this issue.
I confirmed that hwaccel (at least RX580) can shows artifacts for Socket's video in FCPX. He send me the source video, and I can easily reproduce the issue.
However, at this moment, it seems the video itself has codec parameters error and missing keyframe. May be hardware decoding is more sensitive to codec error. May be hwaccel is not quite compatible to this particular codec. Need more data to find out what's the issue.
There are few unknown values shows in media info. I wonder if this is the problem.
I tried another 10 different videos, frame to frame edit doesn't shows any artifacts.
So, it can happen, but we don't know what exactly required. However, the chance should be low.
N.B. If you want to know how to turn hwaccel ON / OFF like it shows in the video, please check post #1 Q:24
However, at this moment, it seems the video itself has codec parameters error and missing keyframe. May be hardware decoding is more sensitive to codec error. May be hwaccel is not quite compatible to this particular codec. Need more data to find out what's the issue.
There are few unknown values shows in media info. I wonder if this is the problem.
I tried another 10 different videos, frame to frame edit doesn't shows any artifacts.
So, it can happen, but we don't know what exactly required. However, the chance should be low.
N.B. If you want to know how to turn hwaccel ON / OFF like it shows in the video, please check post #1 Q:24
I ran VideoProc in a new iMac Core i5 3.7 GHz equipped with a Vega 48. In the settings I see only the Intel graphics, while the Vega 48 is totally absent. I e-mailed their technical support about it, but so far they stay silent. Anyone knows what is going here? In their site they say that VideoProc supports AMD Radeon HD 7700 series or higher.
In the meantime I tried the jellyfish video posted previously and I can say that in ABC Player, Movist and Blu-ray Player Pro it plays flawlessly with minimal CPU effort. I see the Intel graphics and the Vega going both up to 16-18% in iStat Menus. I am not sure about it but the obvious explanation here is that these video players offload the decoding to the GPUs.
It's simple:
If your Mac has a Intel processor that supports Quicksync, it's Quicksync that do the decoding (or encoding, when supported). Quicksync is always used, even if you have a Mac that have a GPU capable of decoding/encoding.
If your have an iMac Pro, that have a Xeon without Quicksync support, the encoding and decoding is offloaded to the GPU.
If your have an iMac Pro, that have a Xeon without Quicksync support, the encoding and decoding is offloaded to the GPU.
I don't understand. Is not a Vega 48 much more powerful than the Intel HD graphics? Why not use both, like the video players that I mentioned previously? Totally ignoring the Vega makes no sense, unless I am missing something here.
Only Apple can answer why they chose to do this way.
Only iMac Pro supports GPU offload encoding and decoding. Btw MP7,1 will have to support GPU decoding/encoding too, since the Xeon family used don't have Quicksync too.
But ... VideoProc is not Apple software, it is Digiarty software. They say that they support basically all kinds of modern GPUs and still their software does not detect the Vega 48. For the time being, their technical support does not answer the question about it.
If you (or anyone else here) have an iMac Pro, could you please try VideoProc and see what it does report in the settings for hardware acceleration? There is a trial version with only some limitations regarding video encoding, but it shows all settings.
AFAIK, VideoProc uses the Apple framework, Videotoolbox. All apps that use Videotoolbox will have the same behaviour.
This is interesting. But it raises questions about Final Cut Pro X. People say so often (example) that it relies heavily on the GPU and the Vega 48 is no exception. Perhaps Apple does not use its own Videotoolbox for FCPX? I am still confused.
It's really complex to understand since Apple exposes the hardware in different ways/frameworks.
Videotoolbox is the Apple framework for encoding and decoding that even FFmpeg is using now, it's the "Apple blessed way" for all apps to have hardware decode (and encoding when the hardware supports it), but FCPx needs more than just Videotoolbox and Apple opted to use the GPU as a METAL compute device with FCPx.
So I randomly decided to buy a Radeon VII and installed a fresh version of 10.14.6. I've done this AMD hardware acceleration and so far, Premiere Pro (latest version) feels really great now. It plays back DCI 4K 10-bit Canon XF-AVC files at full high quality without issues. Before, I felt like the machine struggled a bit just on playback. It even plays Canon Cinema RAW light files back without dropped frames too! I couldn't even play a single stream before in the timeline. Not sure if this has to do with the new GPU or the hardware acceleration or both.
But regardless, the new Radeon VII with this acceleration really has been noticeable.
But regardless, the new Radeon VII with this acceleration really has been noticeable.
Last edited:
PowerMike,
Did you try the HW Accel Hack with your Vega Frontier?
How are you powering the Radeon VII? Pixlas mod?
Yeah I did, and it worked pretty well too on the Frontier. But the Radeon VII just feels faster and more fluid. Just more than what I was expecting TBH. I guess the upgraded Vega20 to 7nm does make a difference after all.
I power via the 2 6-pins to 8-pin and the 4 SATA ports combined into another 8-pin. Then both 8-pins go into an EVGA power link connected to the card, since I already had one. Really powers the GPU safely and keeps it all internal. You could get away with 3 SATA to one 8-pin and still be safe. But I don't use my internal SATA ports now, so just using them all for power.
I power via the 2 6-pins to 8-pin and the 4 SATA ports combined into another 8-pin. Then both 8-pins go into an EVGA power link connected to the card, since I already had one. Really powers the GPU safely and keeps it all internal. You could get away with 3 SATA to one 8-pin and still be safe. But I don't use my internal SATA ports now, so just using them all for power.
So far, my system has been pretty stable with 10.14.6. (Crossing fingers) No locks up yet.
On a related note for FCPX users-- I Googled: Final Cut Pro X + Google Chrome where user reports the Google Chrome browser monopolizes the VideoToolbox from Final Cut Pro X and not playing nice? (Wonder if this could be related to issues I had previously? ).
On a related note for FCPX users-- I Googled: Final Cut Pro X + Google Chrome where user reports the Google Chrome browser monopolizes the VideoToolbox from Final Cut Pro X and not playing nice? (Wonder if this could be related to issues I had previously? ).
So what is the most recent step-by-step to get this hack installed on my Mac Pro?
I have a question regarding whether this would work on my setup:
- MacBook Pro (13-inch, 2016) with Touchbar
- eGPU containing RX Vega 64
My OBS recordings of desktop + camera via Elgato is very laggy/delayed, and therefore I want to enable hardware encoding in OBS, using the encoder option "Apple VT H2644 Hardware Encoder", but OBS doesnt seem to recognise my Vega 64, eventhough my graphic card shows up in System Information and I can use it for other things.
Any answers would be highly appreciated!
- MacBook Pro (13-inch, 2016) with Touchbar
- eGPU containing RX Vega 64
My OBS recordings of desktop + camera via Elgato is very laggy/delayed, and therefore I want to enable hardware encoding in OBS, using the encoder option "Apple VT H2644 Hardware Encoder", but OBS doesnt seem to recognise my Vega 64, eventhough my graphic card shows up in System Information and I can use it for other things.
Any answers would be highly appreciated!
Just did a little test in Catalina DP5. The board-id injection method still works. But again, only H264 hwaccel work flawlessly, HEVC hwaccel is still completely disabled.
But the funny thing is. If I replace the AppleGVA.framework by the 10.14.6 version, I can achieve exactly the same result (with NO board-id injection). I know it's not a good way to do it. And I have no idea if this has any adverse effect yet (especially this is the only framework that has 32bit code). I just did this few minutes ago. No stability test yet. But at least good to know if Lilu method fail, we still can have some kind of backup method to make it work.
But the funny thing is. If I replace the AppleGVA.framework by the 10.14.6 version, I can achieve exactly the same result (with NO board-id injection). I know it's not a good way to do it. And I have no idea if this has any adverse effect yet (especially this is the only framework that has 32bit code). I just did this few minutes ago. No stability test yet. But at least good to know if Lilu method fail, we still can have some kind of backup method to make it work.
It's safe to install the 10.14.6 Supplemental update. Nothing about AppleGVA.framework will be replaced. No precautionary action required (of course, you should still have proper backup).
OS build version is now 18G87
OS build version is now 18G87
Now with a fresh installed EFI RX 580 in my Mac Pro I have a question.
My macOS 10.14.6 is (yet) totally unpatched, and VideoProc of course indicates no hardware acceleration:
But when I play h.264 footage e.g. with VLC it indicatesno some CPU usage and the RX 580 does all the work:
Does that mean h.264 hardware decode acceleration is working without any patching in macOS 10.14.6 ?
My macOS 10.14.6 is (yet) totally unpatched, and VideoProc of course indicates no hardware acceleration:
But when I play h.264 footage e.g. with VLC it indicates
Does that mean h.264 hardware decode acceleration is working without any patching in macOS 10.14.6 ?
Last edited:
no clearly vlc sits at 7% cpu usage. Your GPU is working to display things but it's not actively decoding.
decoding 1080p isn't very taxing even for old cpus. Try playing a 4K x264 clip.
I don't have 4K h264 footage, but 4K HEVC/h265:
Clearly no hardware acceleration (but working fine).