7950x's iGPU is unstable (blackouts, freezes. Both on Linux and Windows)

Hi there!

I’ve just built a new PC and I’m observing some odd behavior/instability when using integrated GPU.

I have Ryzen 7950x running in ECO mode (105 W) and 64GB DDR5 kit from Kingston (KF560C30BBEAK2-64). Motherboard is Asus ProArt X870E-CREATOR WIFI (BIOS version 0505 dated 2024/09/27)

During the first boot into Linux Mint a few seconds after the desktop environment appeared (it defaulted to 4K resolution) the mouse cursor became unresponsive and then the screen went black. Then a few seconds later the desktop came back but a few seconds after that it went black again. The same behaviour occured again and again… (mind that I was able to boot into Linux Mint live USB to install this OS and so this part at least worked fine). Here’s a video of this behavior:

I then booted into Windows 10 to figure out if it’s OS-specific issue. By default Windows uses Microsoft drivers for 7950x’s integrated GPU (1080p (“low”) resolution but it run just fine) but as soon as it automatically installed AMD’s drivers for the integrated GPU (resolution changed to 4K) and I started seeing black screen on and off, some cursor freezes. After a few black screens OS reverted to default Microsoft drivers and “AMD Bug Report Tool” popup appear saying “AMD software detected that a driver timeout has occured”. Updating to the latest available AMD driver manually did not solve the issue (see photo and video of this behavior below)


However what seems to make things better for iGPU was changing GPU frequency setting in BIOS from Auto to 2200 which is what CPU specs declare (idea of this guy on AMD forums). Manually setting RAM frequency to 4800 or 5200 (maximum that CPU officially supports, my RAM kit is 6000 Mhz by the way). Windows seems to work better after with change. All this is when having a monitor connected to iGPU, if I connected it to a dedicated GPU I was able to run Geekbench 6 and Cinebench benchmarks multiple times and it worked without a problem. I even played Cyperpunk 2077 and Stalker 2 but I do need to have iGPU working because I plan to isolate dedicated GPU from the linux host and do a GPU pass-through to QEMU VM so iGPU has to work otherwise there will be no GPU left for the host OS.

(Also interestingly enough when I open HWInfo app on Windows 10 to go to Sensors tab there’s a warning popup (“Warning: Asus EC HWinfo has found an embedded controller (ec) sensor from which it can report data”) and a second after the entire PC shutdown hard. No BSOD, no freezes or blackscreens - full shutdown immediatelly). Here’s a video of this:

Now, back to Linux Mint. I’ve installed it again from scratch. Connected my monitor to HDMI on the motherboard (iGPU). Booted into it, running 4K resolution by default. Desktop environment seems to run fine, I can use terminal but as soon as I open Firefox the entire desktop environment becomes unresponsive, I can’t click on anything (but I can still move the cursor freely) and then it just kick me out of the user session into the Mint’s user login screen. I log-in again, open Firefox - the same thing happens.

Now (and this is a crucial part) after tinckering in BIOS I noticed that this (LinuxMint/Firefox) issue seems to be resolved if I change “Core Performance Boost” setting from Auto to Disabled. This setting turns off frequency boosting for CPU (and from what I understood RAM as well?) so that CPU frequency stays at 4.5 Ghz even during load (it runs at around 5.5 Ghz with “Core Performance Boost” Enabled or set to Auto). After this change I can use Firefox on Linux Mint and it doesn’t kick me out of the session and things look more or less stable. Now, this is far from ideal… 1Ghz lower CPU clock is not what I want, but I do want to have a stable Linux system. I haven’t even started with PCI-device isolation, QEMU virtual machines and GPU-passthrough (which is something that can be considered a more advanced stuff) and I’m already observing so many hiccups. Not the best user experience…

Now the question becomes: what’s the problem? Why Windows 10 runs fine with “Core Performance Boost” set to Auto but Linux Mint freezes and kicks me out of the user session a few seconds after starting Firefox and why Firefox? (could it be that Firefox has some kind of a hardware acceleration which trips the iGPU?) And why did I have to manually specify iGPU and RAM frequency just to be able to run iGPU reliably even on Windows? Is it a hardware fault? If so, which one: CPU, RAM or Motherboard? Is it an issue of BIOS configuration? (note that with BIOS’s Optimized defaults I can’t run iGPU reliably on any operating system). Is it a problem of CPU+RAM compatibility? (CPU officially supports 5200 but I have 6000 memory, though I run it at 5200 and with EXPO disabled. I saw people recommend and run 7950x with 6000 RAM kits just perfectly fine - and just in general I assume buying a 6000 memory but running it at 5200 should be fine or am I mistaken?)

Should I do a full memory (RAM) test? Should I return the CPU? Should I return RAM?

Now a somewhat emotional part (i.e. can be skipped): I’ve built PCs before, my previous system worked fine for 7-8 years and it still works fine but now when I’ve decided to make myself a birthday gift and assemble a powerful workstation PC all I’m doing is troubleshooting instead of enjoying it and I can’t call it stable at all. I have to keep on using my old system for my daily tasks.

What do you think about all this? What should I do? I’d appreciate any piece of advice. Thank you!

P.S. am I going to get banned for tagging @wendell ? Please don’t ban me on my birthday, I’m just seeking help/support/acknowledgment. Thank you for reading all this, friend :slightly_smiling_face:

1 Like

Is that a 1.4-Volt memory kit? If I remember correctly there were BIOS updates for the AM5 platform that capped memory Voltage to 1.3 Volts…

I would start by removing all memory overclocks.

Maybe reseating everything – cpu, cooler, memory sticks as well…

Any undication of CPU thermal throttling?

Good luck!

Not memory, but SoC. However, that is actually a reasonable item to check, especially with the combination of high-capacity sticks and the iGPU being in-use, especially with XMP not being enabled.

For what it’s worth, most or all boards tend to bump SoC voltage just a bit when XMP is enabled; iGPU power is derived from the SoC rail, so it could well be that the demands between memory and iGPU usage are intermittently exceeding what’s stable at the “default”.

Try turning EXPO/XMP on; while 6000 is above official support, it’s more likely to be stable than a manually-set RAM speed coupled with “auto” timings.

Thank you for replying, I think that BIOS update with 1.3V cap was for the CPU, not the RAM. Or am I wrong? The information I’ve found on the internet seems to only mention the SoC, not RAM.

CPU doesn’t seem to be thermal throttling, it’s well within thermal limits. HWMonitor reports around 60 dC during full load and 30~40 dC when idle (mind I’m running it in ECO mode (105W)) using Noctua D15 air cooler (which is a pretty big cooler)

Thank you for replying, I just tried to turn EXPO on at 6000 (is this a correct place in BIOS to do it? (see BIOS photo below)).

It didn’t solve the problem, in fact I think it makes things worse in Linux Mint (see video below). While previously during a ‘blackouts’ issue I was able to move the mouse cursor and it would eventually kick me out of the user session into a login-screen and allow me to shutdown/restart gracefully now it completely freezes the PC and the fans go rather loud (louder than during an all core Cinebench test, I can’t move cursor, it doesn’t kick me out of the user session, Ctrl + Alt + Del doesn’t work either) so I had to do a hard reset (as I mentioned, all this can be observed in the video below)

One thing to note, it seems that if I change the iGPU memory capacity BIOS setting from Auto to 8GB (in Advanced > NB Configuration > UMA Frame Buffer Size) it seems to significantly prolong the period of time when the system is stable and doesn’t freezes but eventually it still blacks out and freezes (I got this idea from this reddit post from someone facing very similar issues to those that I have)

Also, Windows 10 with EXPO enabled (at 6000) and as well as “Core Performance Boost” enabled seems to run fine. Here’s RAM info from HWMonitor:

You are running heavily OC memory and using board automated tooling to “optimize” settings.

Neither of these are path to stable system or guaranteed to work. Tweaking OC is wrong move , first verify stability under normal condition, then try to OC if possible at all.

  1. Disable EXPO and all auto tuning (AI) options in bios. Ideally full factory reset
  2. Run memtest for few hours, OCCT stability test in windows and linux desktop use for confirmation - if there are errors without EXPO, then memory module or hardware might be faulty
  3. Rerun the same with default expo (no AI optimizers) - if it fails, reliable operation is not possible with defaults setting/at all. Not unexpected outcome.

@greatnull thank you for replying to my thread, I’m ok with not overclocking anything. The stability is a priority for me, not the every last drop of performance (that’s why I use ECO mode at 105W)

Could you please elaborate on what do you mean by “You are running heavily OC memory”? I mentioned in my first post that I initially run my RAM (KF560C30BBEAK2-64) at 4800 (Ai Overclock Tuner set to Manual → 4800) since I saw this as one of the supported timings presets in the RAM kit specs document and had EXPO disabled but I still faced those issues. I only enabled EXPO (6000) because @RotaryJunkie suggested to try EXPO/6000 since it might be more stable than Manual 4800/5200 setting with auto timings.

And could you please explain what do you mean by “Tweaking OC is wrong move”? Are you refering to me changing the iGPU frequency to 2200 and enabling/disabling “Core Performance Boost”?

  1. By “full factory reset” you don’t mean “Load Optimized defaults”, right?
  2. With what RAM settings should I run this test? EXPO disabled, but what frequency and timings? Is it ok to set Manual → 4800 or 5200 + ‘auto’ timings?
  3. By default EXPO you mean which of these settings or none of them? (see possible settings in the screenshot below, it’s from the BIOS manual for my motherboard) This point is not very clear to me, could you please clarify?

P.S. again, thank you very much for taking the time and even looking into this. I really appreciate it, friend.

1 Like

“You are running heavily OC memory”? I mentioned in my first post that I initially run my RAM (KF560C30BBEAK2-64) at 4800 (Ai Overclock Tuner set to Manual → 4800)

Didn’t see the original quote, so ignore that gripe of mine.

But double check you actually ran stock settings uninfluenced by asus “optimizations”. Its getting harder to remove those and they are often on by default.

You should not manually set anything, just let either:

  1. inherit stock jedec memory profile ( DOCP II in asus terms)
    Your memory stock jedec profile is DDR5-4800 CL40-39-39 @1.1V
  2. inherit stock EXPO profile set by dimm manufacturer ( EXPO II option)
    your memory expo profile is either DDR5-5600 CL40-40-40 @1.25V or DDR5-6000 CL30-36-36 @1.4V

Run memtest and occp stability testing for both scenarios to see if random glitches dissapear.

I run 7950x3d system with 2x16 6000Mhz mem kit and it is was almost as unstable as what you observed. Memory training did not remain stable across multiple reboots, leaving system passing few stress tests + memtest initially then start behaving erratically days later.

I dont run expo anymore, its not worth the headache and premium to boot.

If your system throws faults even after factory reset in scenario 1), then its reasonably safe to assume hardware fault and start RMA for suspect parts.

It seems that DOCP options are not available for me, I can only select Auto, Manual, EXPO I, EXPO II and Expo Tweaked.

So it looks like my only option is running with EXPO on at 6000 which is still unstable (I can see blackouts in Linux Mint) or can I (should I?) manually specify “safer” (Default (JEDEC)) timings and frequency? Namelly DDR5-4800 CL40-39-39 @1.1V. Does it make sense to run memtest and occp on a system/configuration that I can already see is unstable? Or the purpose of memtest and occp is to check if the RAM kit is faulty? Will a failed memtest indicate a faulty RAM or perhaps a bad BIOS configuration will also make this test fail? Can/should I run Windows Memory diagnostic instead of memtest to check this?

And what does “factory reset” mean in my case since we presume that Asus’s default settings might be problematic? These questions might sound strange but I really don’t know…

And generally speaking, is the RAM kit I’ve picked up (KF560C30BBEAK2-64) a good option for my CPU/Motherboard combination? It is in the list of supported RAM modules for the motherboard but it’s a 6000 kit while 7950x officially only supports 5200 so that why I’m starting to doubt my decision whether it was a good idea to get this RAM kit for this CPU. I assumed you can buy a higher frequency RAM but then run it at lower frequency (i.e. the frequency supported by the CPU)

They are always available, since they are default configuration profile for the memory kit. You are either in wrong section , okayed auto optimization somewhere or asus bios is not logically organized.

You are still in “AI tweaker” section, which is asus proprietary profile based auto tuning system. It can and will overclock and overvolt things without you input, so do not use it until you:

  • have stable base system
  • know and accept what is it doing

Reset everything to factory default state and dont touch ai tweaker section at all if possible.

I would do factory reset first, to see wha asust defaults look like. This should be the base state that must be stable.

Okay, that definitely confirms some amount of unhappiness that should not be there. I do have to ask at this point which RAM slots you have populated; A1/B1 (so the ones physically nearest the CPU socket) are very much not the “right” slots when you have 2 sticks; if your RAM is in them, try A2/B2. If it isn’t, I would be suspecting trouble elsewhere, possibly the Eco Mode setting since it’s generally stable without the iGPU.

@greatnull ok, I’ve reverted to BIOS factory defaults and run memtest (4 passes succeed with 0 errors) as well as Windows Memory diagnostic test (1 pass with no errors). With factory defaults RAM seemed to be running at Default (JEDEC): DDR5-4800 CL40-39-39 (I’m adding screenshots of memtest results below)

However with these factory defaults the Windows 10 won’t boot at all (stuck at Asus Win-loading screen) and then shuts down while Linux Mint freezes a few seconds in and kicks me out of user session into login screen. Mind that this only happens when working with iGPU (monitor connected to the HDMI output on the motherboard)

Assuming these successful memory tests can be an indication that at least RAM is working properly and since I think it should be reasonable to expect that when an average user boots his operating system with factory BIOS settings he should be able to use it for more than a few seconds (on iGPU) so I’m leaning towards either faulty iGPU part of the CPU or some kind of motherboard quirk/issue. This post on AMD’s forum describes similar issue with iGPU and many other people in comments confirm that they also had these issues and although these workaround can make things better - replacing CPU helped them.

I would be more inclined to think it’s the first option (faulty CPU) especially since setting ‘Core Perfomance Boost’ to Disabled make things much more stable. But one thing still bugs me… why is it that when I open HWInfo on Windows 10 and go to Sensors tab there’s a warning popup (“Warning: Asus EC HWinfo has found an embedded controller (ec) sensor from which it can report data”) and a second after that the entire PC shutdown hard? No BSOD, no freezes or blackscreens - full shutdown immediatelly) - this does sound like a possible motherboard (sensor) problem or does it? I mean, no software (HWInfo) bug should be able to cause immediate hard shutdown of the entire PC, right? Even memory corruption shouldn’t ‘kill’ the PC, it kills the application or at worst causes BSOD but not an immediate shutdown (video of this can be seen in my initial post)

So… do I return the CPU to manufacturer complaining about iGPU instabilities and prey that they can reproduce it? Buy the cheapest AM5 CPU (7600x) with iGPU and see if my motherboard works fine with it? Contact Asus regarding that sensor warning/shutdown and ask how else can I check if motherboard is ok?



@RotaryJunkie RAM is in A2/B2 (2nd and 4th slot from the CPU), I’ve disabled ECO mode now (the same issues with iGPU)

Yup, its time top scratch off RAM from suspect lists. Still do not reenable any “enhancement settings”.

Time to isolate/verify what causes instability for you to initiate RMA. Do you have spare gpu handy? Do you have any friend that could let you borrow known good AM5 board or spare cpu?

  • does the stability issue go away when igpu is not in active use?
  • does the stability issue go away when igpu is hard disabled in bios? (system wide stability issues unless hard disabled even with dgpu means definitely bad silicon)
  • if alternative AM5 board is available, does switching it help? ( Asus ProArt X870E-CREATOR WIFI might be the issue )
  • if alternate cpu is available , how does it behave in original board, does the issue persist ( cpu might be the problem, second confirmation)

Warning: Asus EC HWinfo has found an embedded controller (ec) sensor from which it can report data

I remember reading explicit warning somewhere that this is possible failure scenario when trying to interface with some sensor chips on some boards.

It well outside of my capability to debug, so I would skip it as troubling but not directly relevant to issue at hand.

I would be more inclined to think it’s the first option (faulty CPU) especially since setting ‘Core Performance Boost’ to Disabled make things much more stable.

Since CPB auto OCs your cpu and igpu, there no stability guarantees whatsoever and its failure alone it does not imply anything negative about your cpu.

If your setup were stable (and it isnt) under default configuration, only further determination would be your silicon absolutely OK but unfit for OC, dont even try.

If you are going to do CPU rma in future, dont mention using it at all. Using these tools is automatic and rightful warranty void scenario if vendor feels like pursuing it.

EDIT: Haven’t found any hard info whether CPB is amd general technology or motherboard vendor specific tool, but found a lot of reported instance of instability with it on matching your problems.

Unlike PBO which is ON in stock configuration, it should be OFF in stock.

EDIT2: Well I am scratching my head now, there is no official documentation to be found within reasonable time frame and some sources say that CPB feature might be on by default (board specific?).

Double check if its your case.

Either way for production use CPB is worthless performance wise and efficiency wise even if stability is disregarded. Even disabling PBO does not have significant performance impact, while improving thermals and power consumption massively.

Yes, stability issue goes away if iGPU is not in active use. If I connect my monitor to my dedicated GPU I get no blackouts, no freezes. I can play games and run benchmarks. It’s all about iGPU which I need to work properly (since I will be eventually isolating the dedicated GPU to pass it to a VM so I need a functioning iGPU for video output from a host OS)

I just ordered another CPU (7600x) to confirm whether it’s actually iGPU (hence CPU) issue or something else (motherboard?)

I think “Core Performance Boost” is not exactly an OC, it’s more like Intel’s “Turbo” frequency. For example my i7 8700K has a Base clock of 3.70 GHz and Max Turbo Frequency of 4.70 GHz. When idle it runs at much lower than 3.70 GHz, when it’s somewhat loaded it will work at around that 3.70 GHz mark and when fully loaded (especially when only a single core is fully loaded) it will boost close to that advertised 4.70 GHz. I didn’t do or change anything manually to make my (Intel i7 8700K) go to 4.70 GHz, it’s a normal boosting behavior. To my mind OC would be pushing my CPU beyond that 4.70 GHz Turbo Boost mark.

Now with 7950x 4.5 GHz is the Base Clock while Max. Boost Clock is at “Up to 5.7 GHz” according to the specs. If I disable “Core Performance Boost” then my CPU will always run at it’s Base Clock of 4.5 GHz and never higher than that, with “Core Performance Boost” at Auto it will run at different frequencies depending on a load (from a very low to around 5.5 GHz during full load) so I’m not sure that enabling it (or leaving it at Auto, which is what the BIOS’s default setting is) is really an overclocking but I can be wrong about that. I’d be surpriced to run a modern consumer-level relatively low-count CPU in 2025 at 4.5 GHz max, as I mentioned my 7~8 year old i7-8700K boost to 4.70 GHz.

According to my motherboard’s BIOS manual:

New (temporary testing) CPU (7600x) will arrive on Sunday. I plan to test my system with that new CPU just to see if the iGPU issue is gone at default BIOS settings. It’s from the same line of CPUs and it has the same iGPU so if 7600x will work ok then my 7950x is definitelly faulty. If it also has problems then it’s probably a motherboard… at least that’s my current line of reasoning.

Perhaps a crazy idea but is it possible that AMD pushed bad drivers that messed up the iGPU in Ryzen 7000 CPUs to both Windows and Linux? Sounds like a paranoia, doesn’t it? :sweat_smile:

1 Like

That is not right, unless amd changed something again:

  • Precision Boost is amd equivalent of intel turbo and should be always on and doing its thing as ,ong there is enough power and thermal headroom (NOT OC)
  • Precision Boost Overdrive tunes power limits and perf curves, can be used to both overclock or undervolt yout cpu , should not be on by default (OC)
  • Core Boost Overdrive - auto overclocking, should never be on by default (OC)

If disabling CBO also prevents your cores from going beyond base clock, your board is doing something else that you think (possible implementation error) or something else is also problem (power delivery or cooling insufficient). Based on manual description, you should keep it disabled for sure.

Perhaps a crazy idea but is it possible that AMD pushed bad drivers that messed up the iGPU in Ryzen 7000 CPUs to both Windows and Linux

Driver lines are completely divergent, so it way more likely you see hardware error.

TLDR: double check if you are ruuning stock settings, CBO is not stock and can easily cause all the observed instability on marginal cpu. This is not cpu fault if its happening. And can happen even on new cpu.

This was my understanding as well. Might test later if I feel like rebooting my PC…

“Core Boost Overdrive” is not an option I see on my Asus AM5 board… Perhaps it’s an older term from AM4 days? But my understanding of core performance boost is the same as OP, it’s more like turbo boost on Intel and PB (no O) is part of CPB.

1 Like

I cannot find hard trustworthy explanation anywhere also, so this is pieced from marketing slides and anecdotal references.

There is also possibility that motherboard vendor messed up nomenclature.

I am on gigabyte B650 board, so my local observation are not applicable to OP.

Yeah the nomenclature is a mess, already on the AMD side. Then the motherboard manufacturers come in and add a bit more on top…

Maybe core boost overdrive is some Gigabyte thing?

I miss the old days where there would be official whitepaper explaining all the related technologies in detail , from vendor no less.

All we have now are dumbed down marketing slides and youtubers. Or worse, reddit :slight_smile:

Still I put my trust in puget systems test that I posted above, seems like PBO and CBO are in line what I described and they are not wort it from efficiency standpoint. I.e opportunistic automatic OC based on runtime thermal and power with no guarantees, so not worth the efficiency impact if its stable at all.

And those test are 7950x specific, so they are applicable.