trx40 [Discussion] - TRX40 Bare Metal - Vanilla Patches

[fabiosun]

RC

 

 

 

[Ploddles]

Strange how it changed between me grabbing a screen shot with it saying Beta and what it says now, within a matter of minutes.

 

It still isn't showing up in Software Update for me but I am downloading the full thing now.

 

EDIT

 

Even stranger, it has now gone back to Beta 2 😲

Edited November 5, 2020 by Ploddles

[Driftwood]

Hi chaps. All good here with BS RC. 

 

[Ploddles]

New Supplemental Update to Catalina just released, taking it up to 19H15.

[iGPU]

 

Updated without issue. My sequence was BS ß9 to 11.01 a few days ago, and then today to 11.01 RC (I skipped ß10).

 

I use the latest OC compile (release v064; now 6a81462) run daily with, of course, the latest, correct, config settings. Kexts are always latest versions.

[iosengineer]

On 11/4/2020 at 9:03 AM, Ploddles said:

On 11/4/2020 at 8:24 AM, fabiosun said:

@Ploddles

 

Could you try and see if your system starts well also without RebuildMemoryMap?

It should start also without others two quirks but I am curious about RebuildAppleMemoryMap because I see it on many trx40 config..but it is not mandatory..I think :9

Thank you

Read more  

Yes, it boots fine with all 3 disabled. 

 

I have a potentially-interesting finding for you, @Ploddles & @fabiosun:

• First off, I was able to boot with Ploddles' config! Woo-hoo, it's undoubtedly an improvement over my minimally-guide-adapted fork of one of my Intel configs, lol (I'm glad to never use it again).
• Interestingly though, I tried from the start with these three options disabled (AvoidRuntimeDefrag, RebuildAppleMemoryMap, SyncRuntimePermissions), and it did not boot successfully.
  • It did get farther than the ultra-early failure I was seeing before [the Err(0xE) one], but fails on the trusty x86_validate_topology!
  • I have not yet isolated which of the three flags is required to solve this (running out of time for tonight).

Additionally, I've read the excellent iGPU tutorial on MMIO Whitelisting—a true community service, thank you @iGPU! I hope your recovery has been brisk and total.

 

That said, even though I like running OpenCore DEBUG and have the boot logs at hand, I've been changing the PCIe config of the machine almost every day this week. In the next couple days, I'll finalize the configuration and MMIO is my top priority before anything else TRX40-Hackintosh-related. Thanks again for the amazing community spirit that all of you are bringing to this forum!

[fabiosun]

 

Updated with same EFI Clover configuration

No Patches problems

[fabiosun]

@iosengineer

when you have time put in your signature your Hardware so it is more simple to understand if some problem occurs

interesting the quirks behaviour in your system

[iGPU]

6 hours ago, iosengineer said:

 

I have a potentially-interesting finding for you, @Ploddles & @fabiosun:

• First off, I was able to boot with Ploddles' config! Woo-hoo, it's undoubtedly an improvement over my minimally-guide-adapted fork of one of my Intel configs, lol (I'm glad to never use it again).
• Interestingly though, I tried from the start with these three options disabled (AvoidRuntimeDefrag, RebuildAppleMemoryMap, SyncRuntimePermissions), and it did not boot successfully.
  • It did get farther than the ultra-early failure I was seeing before [the Err(0xE) one], but fails on the trusty x86_validate_topology!
  • I have not yet isolated which of the three flags is required to solve this (running out of time for tonight).

Additionally, I've read the excellent iGPU tutorial on MMIO Whitelisting—a true community service, thank you @iGPU! I hope your recovery has been brisk and total.

 

That said, even though I like running OpenCore DEBUG and have the boot logs at hand, I've been changing the PCIe config of the machine almost every day this week. In the next couple days, I'll finalize the configuration and MMIO is my top priority before anything else TRX40-Hackintosh-related. Thanks again for the amazing community spirit that all of you are bringing to this forum!

 

At this late date, why the issue over these Quirks? None of those are necessary. (This is where newbies muddy the waters for no good reason.)

 

This works and has not changed for months:

 

[Ploddles]

@iGPU, I understand that they are required if you want to boot Linux, as well as macOS, from OC.

 

I haven't, yet, install Linux on mine so cannot confirm this but it was an issue found by @jncasey.

[Jaidy]

is there any one here who uses products that on the backend use Intel's MKL, like numpy, MATLAB etc.? If so, can you please tell me how did you manage to get them to work, as I can not install Intel's MKL library yet - and openBLAS while works perfectly fine, is there a way to configure numpy etc. to use openBLAS instead of MKL? Of course if MKL works, then no need to openBLAS..

[jncasey]

19 minutes ago, Jaidy said:

is there any one here who uses products that on the backend use Intel's MKL, like numpy, MATLAB etc.? If so, can you please tell me how did you manage to get them to work, as I can not install Intel's MKL library yet - and openBLAS while works perfectly fine, is there a way to configure numpy etc. to use openBLAS instead of MKL? Of course if MKL works, then no need to openBLAS..

 

I'm pretty out of my depth on this topic, but there are environment variables you can set to trick the Intel library to run optimized on AMD processors – it's how some of the Adobe compatibility is achieved.  

 

This article is for Linux/Windows, but it should (maybe?) point you in the right direction to get MKL working. 

[Ploddles]

51 minutes ago, Jaidy said:

is there any one here who uses products that on the backend use Intel's MKL, like numpy, MATLAB etc.? If so, can you please tell me how did you manage to get them to work, as I can not install Intel's MKL library yet - and openBLAS while works perfectly fine, is there a way to configure numpy etc. to use openBLAS instead of MKL? Of course if MKL works, then no need to openBLAS..

 

I've never used MatLab but you can try one, the other, or both of these:-

 

Apply the AMD Adobe patches to Matlab (change what folder the scripts looks for to patch, ie change Adobe* to MatLab* or whatever it is called)

If memory servers me right, which it doesn't always do 😀, PS etc use some of the same MKL libraries as MatLab.

 

Use a script, as below, to launch MatLab, try values 5 or 2.

 

 

Edited November 6, 2020 by Ploddles

[Jaidy]

45 minutes ago, jncasey said:

 

I'm pretty out of my depth on this topic, but there are environment variables you can set to trick the Intel library to run optimized on AMD processors – it's how some of the Adobe compatibility is achieved.  

 

This article is for Linux/Windows, but it should (maybe?) point you in the right direction to get MKL working. 

I looked at it before posting here. MKL won't install at all. The installer just fades away when clicked on. I think the DUBUG mode was discontinued in the previous builds (probably due to this notoriety?)..

[Ploddles]

Anybody else's Catalina system, that is not enrolled in the beta program, being offered the latest Big Sur Beta (Release Candidate / GM)? This is also happening on my real MBP that has never been in a beta program.

 

 

Edited November 6, 2020 by Ploddles

[jncasey]

3 hours ago, Jaidy said:

I looked at it before posting here. MKL won't install at all. The installer just fades away when clicked on. I think the DUBUG mode was discontinued in the previous builds (probably due to this notoriety?)..

 

I'll reiterate that I'm pretty out of my depth (and know nothing at all about MATLAB), but with numpy at least, aren't the libraries it uses determined at compile time?

 

I confess I've never looked into which libraries my numpy is using. On the Intel hackintosh I'm typing from, numpy was installed via pip as a dependency of pytorch, and it looks like it's using openBLAS by default. It's the same on my Threadripper on Ubuntu (though I haven't installed numpy on that machine's macOS side to check there). See below.

 

Spoiler

(hf) jesse@iMac ~ % python

Python 3.7.7 (default, Mar 10 2020, 15:43:33) 

[Clang 11.0.0 (clang-1100.0.33.17)] on darwin

Type "help", "copyright", "credits" or "license" for more information.

>>> import numpy

>>> numpy.show_config()

blas_mkl_info:

  NOT AVAILABLE

blis_info:

  NOT AVAILABLE

openblas_info:

    libraries = ['openblas', 'openblas']

    library_dirs = ['/usr/local/lib']

    language = c

    define_macros = [('HAVE_CBLAS', None)]

blas_opt_info:

    libraries = ['openblas', 'openblas']

    library_dirs = ['/usr/local/lib']

    language = c

    define_macros = [('HAVE_CBLAS', None)]

lapack_mkl_info:

  NOT AVAILABLE

openblas_lapack_info:

    libraries = ['openblas', 'openblas']

    library_dirs = ['/usr/local/lib']

    language = c

    define_macros = [('HAVE_CBLAS', None)]

lapack_opt_info:

    libraries = ['openblas', 'openblas']

    library_dirs = ['/usr/local/lib']

    language = c

    define_macros = [('HAVE_CBLAS', None)]

>>> 

 

 

So since all I did to get this version of numpy was "pip install torch", my guess is the default binaries are compiled with openBLAS, and for MKL support you need to compile from source. So maybe just install a new numpy binary? Or, if that's what you already did, compile numpy from source and explicitly use openBLAS?

 

It won't get MKL working on your machine, but if openBLAS suits your needs, you can at least use numpy.

 

 

[Rox67er]

6 hours ago, Ploddles said:

@iGPU, I understand that they are required if you want to boot Linux, as well as macOS, from OC.

 

I haven't, yet, install Linux on mine so cannot confirm this but it was an issue found by @jncasey.

Use the exact same list as iGPU and boot Catalina / Big Sur / Windows 10 all with same plist. Have no Linux running.

 

[Rox67er]

😀

 

Update: Smooth update.

Edited November 7, 2020 by Rox67er

[iGPU]

Earlier in this thread, there was some issues with log-in screen being changed after a Big Sur update. This post describes how to fix that issue.

 

First, a written description and then some images that try to illustrate the words. Attached is a folder containing a Big Sur log-in image; download it and then you must change the UUID to be specific for your system.

 

Description:

 

1) Open System Preferences.

2) Select "Users & Groups" --> unlock the padlock --> right click on user --> select 'Advanced' --> copy user UUID from next screen and cancel out of dialog box.

3) Place downloaded folder (Desktop Pictures/UUID/lockscreen.png) into /Library/Caches/

4) Replace folder UUID ("my-UUID") with your UUID copied in step 2.

5) Assign proper writing permission to that folder, using "Get Info", making it Read & Write.

6) After re-booting, you'll now see the BS log-in screen rather than the multi-colored screen (or you can use another image; just re-name as "lock screen.png").

 

 

Get your UUID:

 

 

Copy your UUID:

 

 

Place folder and re-name UUID:

 

 

 

Desktop Pictures.zip

Edited November 7, 2020 by iGPU

[fabiosun]

1 hour ago, iGPU said:

 

 

***

 

NOTE: the system here is not allowing me to upload the image. The image is 42.7 MB and the limit is < 47.37 MB; so some sort of glitch. I/ve tried to upload folder/image several times. I repeatedly get an error ("-200") in trying to upload...

 

Moderator help?

 

***

 

 

 

 

 

 

Sorry I do not understand well

we have that mb limit and a max resolution on embedded pictures of 1200 pixel for x axis resolution

Could you send on my email that picture to understand better?

Thank you

[iGPU]

1 hour ago, fabiosun said:

Sorry I do not understand well

we have that mb limit and a max resolution on embedded pictures of 1200 pixel for x axis resolution

Could you send on my email that picture to understand better?

Thank you

 

 

I logged out and back in again and now the folder uploaded.

 

 

 

Edited November 7, 2020 by iGPU

[fabiosun]

AMD Power Gadget on TRX40 (CPU PM management)

 

 

[Driftwood]

On 11/8/2020 at 10:12 AM, fabiosun said:

AMD Power Gadget on TRX40 (CPU PM management)

 

 

 

Nice. haven't checked this out for some time.

 

Is SMCAMDProcessor.kext working with virtualsmc correctly?

[Rox67er]

On 11/8/2020 at 11:12 AM, fabiosun said:

AMD Power Gadget on TRX40 (CPU PM management)

 

 

Wow 375 W ! That is quite aggressive. With PBO set to level 3 I get up to 345 W and 17631 points in CB20. (@2667Mhz memory, I have to retest but it looks like I have better benchmarks on 2667 than on XMP2.0 3200Mhz)

 

Just confirmed, only changed my memory to XMP2.0 profile default 3200Mhz and I get 17405 pts. (Power ramps to 375 W too by the way.) 

Very strange.... 😳

 

I'm starting to think the ASRock TRX40 creator doesn't like my G.Skill Ripjaws memory.

Edited November 10, 2020 by Rox67er

[fabiosun]

@Rox67er too much little variation in my opinion, I see in the same run 17780 or 17350..or similar

but I will check about wattage 🙂