- Joined
- Nov 3, 2023
- Messages
- 2
- Motherboard
- ASRock-Z390-Phantom-Gaming-ITX
- CPU
- i7-8086K
- Graphics
- RX 6900 XT
Hello kSilva. I bought a new 6950XT and everything is back to normal. Just because it works perfectly under Windows doesn't mean it can run normally under a hackintosh.
- Joined
- Dec 9, 2016
- Messages
- 40
- Motherboard
- ASUS Prime Z490-A
- CPU
- i9-10850K
- Graphics
- RX 6950 XT
I'm having the same problem. I'm using the "AMDRadeonNavi2xExt" kext which works perfectly (benchmarks over 215k for metal and 122k for OpenCL) except for certain graphics/video apps forcing a restart. I tried using XMP II as recommended by the OpenCore docs, tried XMP I, set v-droop to level 1, all of which seemed to extended processing time by about 1 minute before the computer rebooted. I'm using an ASUS Prime Z490-A with XFX 6950XT. Could you share what you changed in your BIOS? What method did you use for getting the graphics card to work? I've spent so much time trying to get this to work by other methods (SSDT and aml files), but as soon as I try making my own, something else breaks and I end up spending days troubleshooting the new unrelated issue only to realize I'm behind in work and revert back to the 4+ year old EFI I'm using. Any help or advice is appreciated, tutorial links or other. Thanks.
- Joined
- Nov 19, 2015
- Messages
- 137
- Motherboard
- Gigabyte Z490 Vision G
- CPU
- i9-10900K
- Graphics
- RX 6900 XT
- Mac
- Mobile Phone
Believe it or not, I resolved my issue by upgrading the PSU. I went from a 850W to 1350W PSU and I have not had any issues so far, even when performing tasks in Lightroom that would trigger the shutdown. However, I upgraded the PSU 2 days ago, so if anything changes I'll make a post.
The funny thing is that many suggest that I needed a more powerful PSU, but I didn't get one because I thought it was enough based on my computer specs. But hours of troubleshooting and nothing else to try, I reluctantly decided to spend the money and upgraded the PSU. So far no issues.
WARNING: Very, very important!!!
I you change the PSU make sure you also use the cables that come with the New PSU. I didn't do this and one of my SSDs was destroyed, multiple chips in it literally blew up when I powered on the computer. Super scary and the smell was awful. You can search about this PSU issue online. I unfortunately learned this the hard way!
- Joined
- Aug 12, 2010
- Messages
- 98
- Motherboard
- Asus Z790 TUF + Wifi DDR5
- CPU
- i9-13900KF
- Graphics
- RX 6950 XT XFX
- Mac
- Classic Mac
- Mobile Phone
I had a post of all my screenshots and BIOS settings but it got deleted (it was a reply in a thread and that thread is now gone). I had copied the text and saved that, here it is if this can help you:
Ai Tweaker
ASUS MultiCore Enhancement (MCE) is usually disabled but you will see later in the BIOS, I'm limiting the total watts the CPU pulls. Thus I am managing MCE. Putting MCE to enabled or auto will be a disaster and fail for heavy video renders and also - if possible because you have good water cooling - pull more wattage than the 13900k is rated for (which is a max of 253 watts).
Performance Core Ratios is the speed at which the CPU will run depending on how many cores are active. Syncing all to a set speed is detrimental to the technology built in the CPU, namely, the ability for each core to be controlled and run independently. I am running 1 core and 2 core ratios at the stock speed of 5.8 ghz which means if one or two cores are active, they will hit 5.8 Ghz. My water cooling is barely able to keep all 8 cores at 5.2 Ghz so I bumped 7 cores running at a time and 8 cores running at a time down to 5.1 Ghz.
Getting a good Cinebench score means trying to get all 8 cores running at 5 to 5.2 Ghz without thermal throttling. Pushing 5.2 and throttling is slower than running 5.0 or 5.1 without throttling. I doubt anyone with normal AIO cooling will be able to cool 253 watts for long without throttling!
I synced eCores because I am using CPUTopologyRebuild.kext which supposedly turns eCores into pCore hyper-threading. Because of this, I do not want eCores running independently. Also, from what I read on overclocking forums, 40-44 is the max threshold to push them.
Specific Performance Core Ratio Limit Setting the ratio limit per core is great if you have certain cores which run hot. In Windows, I tested benchmarks with HWMonitor to see each core's temperature. Cores 5,6, and 7 were the absolute worst running 11C hotter than the rest! Ridiculous. Talk about chip flexing. Cores 0 and 1 are the most cool. So these will be the ones which get to hit 5.8 Ghz.
Example: when I put Core6 and Core7 at 51 that means core 6 and core 7 will never run above 5.1 Ghz. If everything was 'auto' here, then in the previous page for Performance Core Ratio, when setting '1 Core ratio limit 58' and '2 Core ratio limit 58', that means any one or any two performance cores would hit 5.8 Ghz. I don't want that. I only want my coolest two cores hitting 5.8 Ghz - core0 and core1.
Yes, I can do variable voltages here - and I have - but I decided to do variable voltages relating to the Ghz curve (see below). However, doing variable voltages here is good to underclock hot cores. For me, offsetting cores 5,6,7 by .025 (and then ramping that up) could eek out a little better efficiency.
Specific Efficient Core again, you can find your cooler vs hotter groups of eCores in windows and then do voltage settings. Here I just slightly underclocked them all the same which offers a little more cooling for the p cores.
DIGI VRM Load-line Callibration (LLC) is intense. There's a lot to read up on it. Basically, the lower the number, the more voltage sag which can be problematic if quick and high voltage demands are given to the CPU and the higher the number, the higher the voltage.
However, lower numbers run cooler, higher numbers run hotter and have voltage spikes which can overshoot the specified limit. This is why I think restarts were occurring when running PudgetBench for Premier and Lightroom AI Denoiser because it was not just a bench like Cinebench but rather, very fast changes between CPU and GPU and various cores were ramping up and down constantly. This is why I put 140% current capability and upped the VRM switching frequency.
Changing VRM switching frequency from 'auto' to manual at 500 did immediately give me benefit in Lightroom without as many restarts. (It rendered more denoising of photos before restarting). This goes to show it is not just one change but many which get us to the desired result - which is why this is extremely tedious and frustrating.
CPU Management this is where we manage Asus' MCU. The chip can do 253 watts but by limiting it to 230 (you can experiment) I can run cooler. It won't get as high a score in Cinebench but you can fiddle around and get almost the same. CPU Core/Cache limit with MCU is like 4000 watts - crazy. I put 400, it won't hit 400. It will not hit past 230. Package time window is how long to pull that 230 watts before lowering, I read to keep it at 56s.
VR voltage limit - it will never hit this, but 1750 is slightly less than the max of the 13900k so it's a fail safe.
V/F Point Offset Whether you choose to do voltage settings here, or globally or in the specific performance core section is all up to you. I like to know that when a certain Ghz is reached, the voltage will change. These settings changed whether PudgetBench for Premiere would finish or restart - specifically, the voltage settings for the higher Ghz (above 5.1).
My chip running at 5.1 Ghz can be undervolted and stable but above 5.1 Ghz it will not be stable with an undervolt.
However, if I gave say 5700Mhz +.1v, it would restart, if I gave it +.025v it would restart. It's the goldilocks conundrum here - it has to be 'just right' - which is extremely time consuming. And this will most likely differ depending on the 'silicon lottery.'
So why not put this all in auto? I DID. I put everything in auto. And the computer ran almost perfectly, except when I needed it to do what I actually built it to do: render videos with VBR 2 pass
So why not just do a VBR one pass and move on? Because I'm stubborn. I want this thing to WORK.
Note: V/F point 10 and 11 are repeats of 9. It's been talked about on forums and no one knows why. For me, I just put the same settings in there 'just in case'.
Back to Main AI Page Ring Down Bin if disabled will allow you to put ratios in. They need to be the same, 45 is good, I read some people put 50 but no more else it will not be stable. At this point, I'm doing auto.
So why in the world would I do a Global Core SVID Voltage addition? This is putting it on every core at every speed - it's a global setting. But in the V/F page, we added or subtracted a voltage at a specific Ghz. Are they the same results? NO.
This is not a simple compound addition or subtraction to the core voltage. At 5.1 Ghz, if I did a negative .05v in V/F, Cinebench worked and ran a bit cool (remember, I have 8 cores active during Cinebench and my ratio is set for when 8 cores are active, to run them at 5.1 Ghz. Thus, the V/F curve also kicks in once they hit 5.1 Ghz and then subtracts .05v). Cinebench would drop a few degrees C when this kicked in.
Now, for the test: in V/F I did a -.025v and global SVID a -.025v, which should equal to the same -.05v, right? Well, Cinebench would then freeze. So, the reaction is not the same. Maybe because at least -.025v was occurring constantly and not only at a certain Ghz.
At the end of the day, I was able to run my Hack with a negtaive -.025v globally but would restart at times for video renders. So I am adding .025v globally which lowered my Cinebench score slightly (more heat).
Advanced
CPU Configuration nothing much to say here.
CPU Power Management Control Boot performance mode just means what mode the CPU will be in while in BIOS or booting to the bootloader. Running without turbos makes your life easier - especially if you are having voltage issues. Nothing like not being able to get into BIOS without waiting a super long time because something errored out.
CPU C states - at this point, I've tried them all and am sticking to Auto.
APM Configuration Erp Ready determined whether it would wake properly or not. S5 works for me.
Onboard Devices Configuration LED lighting is a big issue. Depending if you did your USB port mapping correctly, what you choose here will determine your luck at waking from sleep. You need to make sure the internal LED USB port is disabled in your mapping. PS: the lights always are on, no matter what I do. So 'all off' 'Aura off' does not change what I see on the mobo but it did change how it reacted to sleep.
Disable serial port configuration.
System Agent VT-d must be enabled to use native Intel network drivers for this board. Without this on, I had all types of weird issues resulting in restarts or not sleeping/waking properly. I thought it was voltage issues - which may had played a part - but things were much more stable with this on.
Platform Misc Configuration Did not change anything here.
PCI Subsystem remember to have Config settings properly set for Resize BAR support.
Thunderbolt I'm not running TB devices
Trusted Computing
USB Configuration
Boot Configuration
Make sure Setup Mode is advanced so you get right to business in BIOS.