23d1

4 minutes ago, fabiosun said:

I build it from here:

https://github.com/acidanthera/OpenCorePkg

inside the zip file you can find

build_oc.tool

it compiles for you debug and release latest opencore master version

 

Of course. 🤦🏼‍♂️

Thanks! Gonna boot into bare metal with logging to file on debug. Brb.

22 minutes ago, fabiosun said:

if I have had this problem I will try to use an OpenCore debug version and I would post here log ...

Second option..

Send me your 3990x 🙂 🙂

 

Where do you guys find the beta releases? Right now I've just snagged what I could from posts here to got onto 0.7.5, the github repo is still on release 0.7.4. Figured I'd drop in the debug version. Here's a debug I generated with 0.7.4 a couple of tweaks back, but not sure if that one had DummyPowerManagement enabled (though I still have the same results with and without);

opencore-2021-10-13-230144.txt.zip

CPUInfo.txt.zip

4 minutes ago, fabiosun said:

config-2.plist.zip 6 kB · 0 downloads

and this is previous yours with correct value for dummy

I do not know if that quirk is useful or not for you because to some users solve some freeze after few minutes of OS usage

 

this is good

 

and this previous one posted (no)

Weird, well... I wonder if disabling Above 4G Encoding would do anything, it did allow me to finally pass through my 6800XT to VM when launching via Qemu/Virt-Manager through arch. Haven't booted bare metal since.

3 minutes ago, fabiosun said:

check if that you posted here (quoted one before) give the same

if so..you have a problem 😛

 

I hope there's a problem with the config, because I doubt anyone will solve any issues with 3990x anytime soon.

I suspect it just hasn't been tested thoroughly, seeing as most results in the Geekbench 5 database are within the same range.

config.plist.zip

2 minutes ago, fabiosun said:

@23d1is that the previous you posted?

 

are you sure Do you have only a booting EFI in your system ?

I have downloaded again it and it is on true

If you used @DriftwoodEFI he has tha quirk on /yes/true 

 

Yes, I just mounted the EFI to check and grab that screenshot. Could it be my ProperTree not showing the right values? Everything else seems to be in order...

Just now, fabiosun said:

in this you have it enabled

Hmm, what the... Here's how it looks on my end;

1 minute ago, fabiosun said:

great advice

I see many of you (ASRock creator users) use this quirk

But, if it not useful for you you should avoid that quirk

I've never had it enabled, should I try to enable it?

11 hours ago, Rox67er said:

Do you have DummyPowerManagement enabled? (Should be disabled)

 

 

Mine is working just fine (Same motherboard)

 

While running Cinebench, I do run a low noise profile on my AIO so running a bit hot

Yes, DummyPowerManagement is disabled.

3 hours ago, fabiosun said:

 

Thanks, yes—the values I have and the MMIO Whitelist is all correct.

2 minutes ago, fabiosun said:

@23d1have you calculated your exact MMIO whitelist or are you using @driftwood one?

I'm using driftwoods one, which I now realize might be a mistake as I probably have some different settings in general. Do you have a link to info on how to calculate?

In any case, I ran some Geekbech 5 benchmarks, and I seem to land at 885 single core and 25326 multi core, which is consistent with other results I see for 3990x ranging from 20k to 32k. So there seems to be a general (pretty large) performance loss with the 3990x on macOS compared to win/linux. Which would point to a need for a kext or something to tap into the full performance...

39 minutes ago, fabiosun said:

@23d1

no extra idea..it is weird

here is pretty simple

PBO disabled or auto max tdp under 280 w

PBO enabled (smart overclocking and max tdp about 375 watt till temperatures is moderate)

 

we should compare Bios option and behavior with other Asrock creator user like @Driftwood 🙂 )

 

Yes, very very strange, I agree. I can snap some shots of my settings when I get a moment for comparison.

I actually used @Driftwood's EFI shared in this thread as a starting point and built on that with AMDPowerManagement and so forth.

I'm beginning to think more and more that it's a kernel thing that would need to be patched or kexted somehow. It seems that it just defaults to factory safe settings when in macOS, as I'm not even seeing any change in clocks in macOS when having PBO enabled (I thought I did at first, but that was through booting as KVM). Super strange.

19 minutes ago, fabiosun said:

@23d1

1-SSDT-PLUG.aml.zip 645 B · 0 downloads

try to declare this SSDT in your ACPI/add section (copy it in ACPI folder)

and see if something happens

Config is fine , i think is a problem related to your bios (version/settings)

In window your CPU scale well (min/max frequency and TDP i mean)

Thanks! Will try it out and report back.

Yes, in Windows and Linux it scales perfectly, and the increase in performance with PBO is insane, but staying stock for now until I swap my Noctua for Icegiant tonight.

Edit;

No change.

7 hours ago, fabiosun said:

@23d1could you post your config.plist?

 

Here it is.

config.zip

7 hours ago, fabiosun said:

maybe I was not clear

I would like to see this:

 

during Cinebench test with PBO enabled

 

it is a bios problem

AMD app if you do not modify basic feature or option is not responsible of lack of performance

I can capture it, however it stays static at 2.9Ghz/2.9Ghz, only thing that changes is temp (barely) and wattage, but wattage stops at 250W. This is true for both with and without PBO enabled. Not sure what bios setting might help change that, if anything. It seems like a kernel level issue, but I could be wrong of course.

9 hours ago, fabiosun said:

To have PBO enabled you must have a proper cpu cooling system

but to prove a thing you can try this:

put it enable in bios

boot osx

open AMD power gadget tools

run cinebench cpu benchmark one time (off multiple tests)

check tdp with amd power gadget max tdp value i mean

You will be astonished 

check the same with auto or disabled for PBO option

in poor cpu cooler condition PBO in auto or disabled produce pretty same tdp value

as side note

a huge tdp value produces tremendous temperatures and a tiny improvement in benchmark

 

Yeah, on the Noctua right now after discovering it worked better than the AIO I had originally, but getting an Icegiant Prosiphon Elite tomorrow (hadn't realized it was released and a lot smaller than the prototype). In any case, I doubt it'll do anything for my current predicament with performance. Still getting 2.9GHz at a max of 250W even with PBO, so it seems to be an issue with 3990x and power management kexts if nothing else. Basically untapped performance. I don't even know what's possible anymore in terms of masking the CPU as something else, but I assume the only way is through KVM. There should be an increase/boost showing up, but no dice. I think that there is essentially no power management in macOS with 3990x.

Here are a couple of illustrative vids (watching these I realize it's time to swap the monitor);
 

Same results pretty much.

2 minutes ago, fabiosun said:

Could you post a screen grab of your PBO bios settings?

thank you 

I actually set it to the default Auto (Off), as my temps were getting crazy. It did increase the speed in general (with the same percentages of performance loss on bare metal, 30% and KVM, 15%). I had it set to the mid level (3), which was the default for enabled.

13 hours ago, 23d1 said:

Yeah, no difference in performance.

I posted an issue to the kext's github;

https://github.com/trulyspinach/SMCAMDProcessor/issues/142

Doesn't seem to be overly active development, so we'll see if it takes...

Hopping back in this thread to see if anyone has some info in regards to stalling on "PCI Configuration Begins" in the kernel when booting with a passed through GPU in Big Sur? Would love to be able to boot via qemu as well as bare metal, which is working. I seem to be getting about 15% better performance via qemu kvm. Single GPU passthrough with all the hooks and everything is working, launching from within my Arch install, so it's odd to get stuck in the kernel.

I tried setting the boot flag npcie=0x2000 but did nothing.

Edit;

Disabling "Above 4G Encoding" in BIOS resolved the issue. Strangely, my KVM is functioning better/faster than my bare metal boot. Just some kinks to sort out and pass through all the proper ports and devices.

6 hours ago, tomnic said:

We need to find out why your cpu stays at 2.9 GHz... have you tried removing all the AMD PowerManagement stuff?

Yeah, no difference in performance.

39 minutes ago, fabiosun said:

@23d1you know , I loved Proxmox way 🙂

but results in optimal situation was very similar and a bit lower than bare metal for all of us

 

The advantages of proxmox were others, greater compatibility with Adobe apps ... but today I don't see valid reasons to still use that kind of virtualization (albeit advanced and fantastic) 

I hear ya. I would probably be running KVM if I could get past the pci message in kernel, just stops there when passing through the 6800XT, just before booting into the UI. Bare metal seems to work pretty good in general.

9 hours ago, tomnic said:

 

Probably you're not boosting with PBO... try to set it to enabled in bios if it is on auto.

Yes, I tried PBO, but it's the same difference in percentage, basically. So there's quite a loss of performance in bare metal compared to VM. At least 10-15%. And 30% compared to Win/Linux...

Strange. I wonder if it's due to KVM being on a fully threaded 64 core as opposed to single threaded, and Linux is more efficient in terms of offloading to more threads in the background so to speak. I currently get about 20-30% loss of performance in Cinebench R23.

On 10/12/2021 at 12:58 AM, fabiosun said:

KVM should be a bit less fast than Bare Metal or, in an ideal condition, kvm has similar values

 

 

Bare metal; around 43000 points

KVM; around 53000 points

Windows; around 63000 points

Edit;

It does seem as though the CPU isn't using any boost or power management, it just sits at 2.9GHz and the fans don't even spin up when under load.

I wonder if this is somehow due to not running at 128 threads, though probably not. I am loading AMDRyzenCPUPowerManagement.kext and SMCAMDProcessor.kext with AMD Power Gadget app... Frequency just stays at 2.9GHz...

3 hours ago, tomnic said:

 

Patched the offending libraries, libtbb.dylib and libtbbmalloc.dylib, they call intel.fastmemset.A and intel.fastmemcpy.A, specific for true intel cpus... pointed to intel.fastmemset.V and intel.fastmemcpy.V, for generic AVX enabled cpus, like our glorious AMD... boom, it worked

maya mod.zip 333.13 kB · 1 download

Oh, gotcha—that's awesome. Will check later, but pretty sure those are the exact libs/issues I remember from fine combing the error report. Would also explain why it works under proxmox when using Penryn (haven't tried beyond 2020 version in KVM yet though).

Thanks!

As a side note; I noticed I get about 20k less points in Cinebench compared to fully threaded 64 core in Windows, and about 10-13k less compared to KVM. Anyone have any insights I to the accuracy of that? Feels like a pretty steep loss of performance.

2 hours ago, fabiosun said:

 

@23d1

thanks @tomnic for patched library 🙂

 

Did you run into the same crash at first (I can share the report if not), and how on earth did you solve it? 🙂

7 hours ago, fabiosun said:

No I don’t think so

it should be a driver efi but it does not exist 

Makes sense as not a lot of people are running the 3990x in general, and specifically macOS...