Unusual high benchmark score for a Vega 64 on Mojave 10.14.6 (93500 Metal on GB5)

[ozamal]

I've been optimising my system for the past few nights... had issues with freezing on screensavers which I eventually fixed up today!!

However I have added the RadeonBoost.kext (v1.0) to my system along side KGP Vega 64 SSDT and the benchmark results went off the charts without any weird fan behaviours (Powertable GPU OC and Fan control implemented in the SSDT taken from VGTab and OC to 1680Mhz GPU and 1050Mhz vram).

My system:
My system is based on KGPs X99 iMacpro 1,1 Mojave
CPU 6950x oc at 4ghz
Asus Deluxe II Mobo
64gb RAM @ 3200 Mhz
Sapphire Vega 64 Liquid
Mojave 10.14.6

I saw that the developers of RadeonBoost removed support for Vega 56 and 64 cards because of this fan and power management issue, but with the EFI and my Sapphire Vega 64 Liquid cooled edition I can't notice much of a fan noise difference, as a matter of fact 45dB before applying the kext and 46dB after applying the kext. But the benchmark went up 34% for open CL on Geekbench 5, and 42% for Metal. Valley score remained about the same at 3500 on Ultra 2K setting and Open GL Luxmark at about 33000.

The only thing that is weird is that the GPU Core clock on HW monitor appeared to be halved when using the kext, but obviously the benchmark don't reflect that and there must be a conflict somewhere with the FakeID sensor. GPU temp remain unaffected in Luxmark running the GPU at 100% with temp around 70 degree in both case (with ambience temp at 30 degrees right now).

Thought I would share... this puts the Vega 64 almost on par with a Radeon VII which seems strange... also seems super stable.

 

[pastrychef]

It looks about right...

Here's my Vega 56 (RadeonBoost.kext installed. Custom PowerPlay Table in use. No GPU SSDTs needed.):


Here's my Radeon VII (RadeonBoost.kext installed. Custom PowerPlay Table in use. No GPU SSDTs needed.):


Edit:
Updated Vega 56 Geekbench scores. I previously had wrong version of RadeonBoost installed in that system.

 

[ozamal]

@pastrychef do you happen to have v1.1 of radeonboost.kext? i can't find it anywhere. Also with your powertable, are you just using the kext exported by VGTab? I've noticed that your Luxmark score indicate the actually GPU core speed whilst mine are always showing 945Hz...

 

[pastrychef]

@pastrychef do you happen to have v1.1 of radeonboost.kext? i can't find it anywhere. Also with your powertable, are you just using the kext exported by VGTab? I've noticed that your Luxmark score indicate the actually GPU core speed whilst mine are always showing 945Hz...

No, I do not have version 1.1. But I do have the codeless version of 1.0 (attached below). This the only one I have that works with Vega 56/64.

I used VGTab then extracted the PowerPlay Tables and inject them via config.plist > DeviceProperties.

 

[ozamal]

No, I do not have version 1.1. But I do have the codeless version of 1.0 (attached below). This the only one I have that works with Vega 56/64.

I used VGTab then extracted the PowerPlay Tables and inject them via config.plist > DeviceProperties.

thanks mate! Pardon my ignorance but any particular advantage in using the codeless version?

 

[pastrychef]

thanks mate! Pardon my ignorance but any particular advantage in using the codeless version?

I feel it's less likely to break since it's just an info.plist.

 

[ozamal]

I feel it's less likely to break since it's just an info.plist.

Thanks mate! One last question if you don't mind. There is a massive difference between GPU Temperature and GPU Die temperature reported by iStats. Valley seems to display GPU Die temp, but the target temperature in VGTab target the regular GPU Temp it seems (my target temp is 65 degrees)... I can tell by monitoring the GPU fans in terminal. This mean in Valley the GPU get to about 80 degrees during load with my OC settings... is this normal?

Also I've tried the devices injection in clovers instead of the KGPs SSDT which incorporates it, it works but I still have the GPU Clock display stuck at 840Hz which is exactly 1/2 of my OC (1680)... any idea on how I could fix that?

 

[pastrychef]

Thanks mate! One last question if you don't mind. There is a massive difference between GPU Temperature and GPU Die temperature reported by iStats. Valley seems to display GPU Die temp, but the target temperature in VGTab target the regular GPU Temp it seems (my target temp is 65 degrees)... I can tell by monitoring the GPU fans in terminal. This mean in Valley the GPU get to about 80 degrees during load with my OC settings... is this normal?

Also I've tried the devices injection in clovers instead of the KGPs SSDT which incorporates it, it works but I still have the GPU Clock display stuck at 840Hz which is exactly 1/2 of my OC (1680)... any idea on how I could fix that?

I don't trust the temps reported by Valley. It has never been accurate for me.

 

[djbuddha]

No, I do not have version 1.1. But I do have the codeless version of 1.0 (attached below). This the only one I have that works with Vega 56/64.

I used VGTab then extracted the PowerPlay Tables and inject them via config.plist > DeviceProperties.

Before and after. YOU'RE A KING FOR THIS!!! THANK YOU!!!

Now stress-testing it with OBS and hardware acceleration turned out while running an advanced scene on my Sapphire Nitro+ Vega 64 on my 10980XE Hackintosh LOL

Final Cut Exports are NOTICEABLY faster!!

 

[djbuddha]

Update .... so running 2 instances of OBS using hardware accelerated encoding ... both OBS's are recording (one is streaming and recording, the other is screen recording a browser window of the stream) using the GPU, no dropped frames outputting at 1080/60fps at 8500kbit video / 192kbit audio. My temps are sitting around 50 degrees on the Vega 64 GPU and the computer is using just under 14% usage (with a lot of other apps running), with 3 monitors attached. No hangs, no crashes after about 45 minutes. THIS IS INSANITY.