MP All Models Tired of low Geekbench scores? Use RadeonBoost!

[DSM2.Hackintosh]

@tsialex You know yourself that a 4.1 can also be flashed on 5.1 and Catalina can also be used with a 4.1/5.1. So why remove it?

@miztahsparklez @IndioX And just by the way OC injection works without any issues.

No matter Kext injection or Device Properties.

 

[tsialex]

@tsialex You know yourself that a 4.1 can also be flashed on 5.1 and Catalina can also be used with a 4.1/5.1. So why remove it?

@miztahsparklez @IndioX And just by the way OC injection works without any issues.

No matter Kext injection or Device Properties.

My suggestion to remove MacPro4,1 is just that it's useless.

A MP4,1 flashed with a MP5,1, what we call MP4,1>5,1, identifies itself as a MP5,1. The only difference, and what we use to identify a flashed one, is the SMC version (immutable outside replacing the SMC micro controller from a real MP5,1 board):

early-2009 = 1.39f5
mid-2010/mid-2012 = 1.39f11

A MP4,1 with MP4,1 firmware don't support APFS, don't support 1333MHz RAM, don't support Westmere Xeons and have vulnerable microcodes.

 

[Dayo]

A MP4,1 with MP4,1 firmware don't support APFS

How about those with modified ROM to add APFS?
I have done this on my cMP3,1. I presume (Not sure) that it can be done on a cMP 4,1. Can it not?

 

[CMMChris]

Ah, true that, didn't take that into account
I guess I will keep it in there tho. Doesn't do any harm. Apple haven't removed their AGPM configs for the 4,1 from Catalina either.

 

[tsialex]

How about those with modified ROM to add APFS?
I have done this on my cMP3,1. I presume (Not sure) that it can be done on a cMP 4,1. Can it not?

Sure, but why risk it?

The MP4,1>5,1 upgrade process is older, brings lots more benefits and most were forced to do it back in the Sierra days.

Obviously some very insane person could still be running a hackintoshified MP4,1 loading APFSJumpStart/NVMe/etc modules from the EFI partition and hacking macOS. If someone has the knowledge to keep a MP4,1 running this way, I bet that he is doing it for the ride and not because he can't upgrade to MP5,1 firmware.

 

[Dayo]

If someone has the knowledge to keep a MP4,1 running this way, I bet that he is doing it for the ride and not because he can't upgrade to MP5,1 firmware.

Most likely.

Can you see any penalty with leaving this in the kext apart from nobody really needing it except one or two "insane" people?

 

[tsialex]

Most likely.

Can you see any penalty with leaving this in the kext apart from nobody really needing it except one or two "insane" people?

I don't think that will be any trouble, just that most developers love to delete no useful code.

 

[Tizucumacu]

New version attached in first post. Vega 64 owners please check if fans still behave abnormal.

No difference in Vega 56 fan issue. The performance is ok. I can only silent the fans by running an HEVC video.

 

[CMMChris]

Ok, driver / firmware bug then. Can't fix that.

 

[Tizucumacu]

Ok, driver / firmware bug then. Can't fix that.

I've found kext who stopped the fans, maybe you can compare/use the plist inside. Thanks for your work.

 

[Luca Maddalena]

On my hack with MSI Vega 56 8GB Boost OC WaterCooled with EK block BUT not overclocked there's basically no performance improvement.
Some Geekbench OpenCL/Metal are slightly worse, others slightly better.

Final Cut Pro X ProRes Export for BruceX is slower with the Kext:
- Yes Kext: 10.5
- No Kext: 9.5
Final Cut Pro X H264 High Quality 4k export for BruceX is faster with the kext:
- Yes Kext: 23.0
- No Kext: 24.5

LuxMark identical.

In my case: Useless.

 

[CMMChris]

New version 1.2 added in first post. Vega10 support has been removed (Vega 56, Vega 64, Vega Frontier) due to broken fan control (yes, I was wrong about it not happening on a Hackintosh).

Reason:
People have observed fan control being broken when using this kext like it was the case prior to the release of macOS 10.14.5 Mojave. You remember? 10.14.5 was the macOS release that fixed the fan issues with Vega 56, 64 and Frontier cards. But you might also remember that exactly with this release the Geekbench scores have become lower as well.

Thanks to the developers of Whatevergreen we now know that Apple disabled the loading of the SMU Firmware to PC graphics cards with this release of macOS. macOS does now check for the device property "ATY,EFIVersion" before uploading the SMU firmware to the graphics card. Adding it back causes the SMU firmware to be loaded again and is the main reason for the increase in Geekbench scores. Anyhow, it also causes the Vega10 fan bug to come back.

It really looks like Apple never fixed the Vega10 fan control issues which apparently are caused by a bug inside the SMU firmware. So they simply did a dirty hack, disabling the firmware for all PC cards and thus essentially crippling them in some way.

Since I cannot fix this issue and a silent card is more important than Geekbench scores, I decided to drop Vega10 support for RadeonBoost.
If you are pissed about the way Apple does handle bug fixing these days, open bug reports about this issue.

 

[h9826790]

On my hack with MSI Vega 56 8GB Boost OC WaterCooler with EK block BUT not overclocked there's basically no performance improvement.
Some Geek bench are slightly worse, other slightly better.

Final Cut Pro X ProRes Export for BruceX is slower with the Kext:
- Yes Kext: 10.5
- No Kext: 9.5
Final Cut Pro X H264 High Quality 4k export for BruceX is faster with the kext:
- Yes Kext: 23.0
- No Kext: 24.5

LuxMark identical.

In my case: Useless.

It looks like you benchmarked in the wrong direction.

GB OpenCL / Metal, not GB CPU.

BruceX is more bottlenecking on encoding, not a Metal / OpenCL process. Especially true on the H264 exporting test, even with hardware encoding.

Luxmark is relavant, but we already know that's highly optimised in macOS, and this kext won't unlock anything for it.

Of course, this kext can be useless for your workflow. I won't argue that, you are the only one that know what you really need. But we also already proved that this kext definitely can unlock something. And few usages are proved can be benefited from it.

 

[h9826790]

Just found one of the fundamental difference after applying RadeonBoost. It's the Device to Device memory bandwidth.

This may explain why Vega (including Radeon VII) benefit from this kext more than Polaris etc (because the of HBM)

Without Radeon Boost

With Radeon Boost

I tested this a few times. Consistent result.

In fact, when I ran those Compute Performance tests, that Device to Device Memory Bandwidth can go even higher with Radeon Boost.

If the HBM's performance was artificially locked. That of course can cause some performance degradation for certain usages.

Not sure if this is one of the main difference. But if it is, this can explain why some usages / benchmarks that are not memory bandwidth sensitive (device to device) show no benefit from this kext.

 

[Brian Y]

What's the issue with Vega 10 fan control? If you can send me some technical details, I'm happy to take a look (I have a Vega Frontier and I'm comfortable with driver hacking if you think this is fixable).

 

[CMMChris]

@h9826790 I can confirm this. Interesting observation!

@Brian Y I explained it here: https://forums.macrumors.com/thread...st.2231366/page-4?post=28381763#post-28381763
Apple never fixed the fan control bug. All they did was disabling the SMU firmware upload for ALL PC cards. The fan control bug of Vega10 seems to be inside the SMU firmware. It's a shame it took them more than a year to release such a dirty hack "fix" in 10.14.5.

 

[leoaf79]

Just found one of the fundamental difference after applying RadeonBoost. It's the Device to Device memory bandwidth.

This may explain why Vega (including Radeon VII) benefit from this kext more than Polaris etc (because the of HBM)

Would you please explain that? Which device are you talking about? Why Polaris doesn’t do that too? Is it a different architecture?
Thanks.

 

[h9826790]

Would you please explain that? Which device are you talking about? Why Polaris doesn’t do that too? Is it a different architecture?
Thanks.

Vega use HBM2, Polaris use DDR5

 

[CMMChris]

Polaris also doesn't need SMU Firmware. It needs SMC Firmware instead which is loaded by "PP_EnableLoadFalconSmcFirmware" which it seems is the part of my kext that gives a boost to some Polaris users.

 

[CMMChris]

New version 1.3 added to the first post.

 

[Ludacrisvp]

So a few things changed between benches ... but still the same version of macOS and the same hardware.
Updated to a newer beta SSE4.2 emulator from @Syncretic
Added RadeonBoost
Updated WEG/Lilu versions

RX580 4Gb Pulse on the cMP 3,1

GB5 scores:
Metal Before 40,845
Metal After 48,462

OpenCL Before 38,830
OpenCL After 43,246

Moved to version 1.3 and stats are slightly better, metal is now just over 50k
(ignore that my cMP3,1 has an identity of vmware7,1 and the iMacPro1,1 board... the board ID is for AMD accel and the model type i was doing some testing with a long time ago and never changed it)

Metal:

VMware7,1 vs VMware7,1 - Geekbench

OpenCL:

VMware7,1 vs VMware7,1 - Geekbench

 

[Dayo]

Geekbench 4.3.3 OpenCL (See Post 69)
v1.2: 134,446
v1.3: 129,365

Seems like a reduction for me.

 

[Ludacrisvp]

Geekbench 4.3.3 OpenCL (See Post 69)
v1.2: 134,446
v1.3: 129,365

Seems like a reduction for me.

What numbers do you see with GB5?

 

[CMMChris]

@Dayo Make sure to test with a freshly booted machine and run multiple tests in a row. Geekbench scores never are the same.

 

[Dayo]

@Dayo Make sure to test with a freshly booted machine and run multiple tests in a row. Geekbench scores never are the same.

Yeah. Machine freshly booted to inject the kext.

What numbers do you see with GB5?

Not run v5 before as I had done some benchmarks on my machine with v4.3.3 about a year ago before I started going hardcore on upgrades and keep using it to compare.

Anyway, I downloaded GB5 ... Benchmark (Right Side) is v1.2:

OpenCL Comp Link: https://browser.geekbench.com/v5/compute/compare/786553?baseline=786488
Metal Comp Link: https://browser.geekbench.com/v5/compute/compare/786579?baseline=786707