FW Desktop randomly shuts off in "Best Performance" but seems much more stable in "Best Efficiency"

Framework Desktop
1

Hi folks,

I have a strange problem and would like to share some interesting observations.

My System:

  • Framework Desktop mainboard (AMD Ryzen AI Max+ 395, 64GB RAM), built into a mini-ATX case (not the Framework case) with 3 case fans and be quiet Pure Power 13 M 550W.
  • Windows 11 Pro, BIOS 3.06 (latest), AMD drivers up to date.

My problem:

  • Roughly once a day the system cuts power completely and instantly — no freeze, no lag, no black screen with a still-responsive system, just gone in under a second. Happens mid-session during completely mundane tasks (browsing, Office work), regardless of load or which application is in the foreground.
  • I can switch it on again up afterward with a simple press of the power button and then runs stable for many hours. Initially I couldn’t find any pattern at all.

What I’ve checked:

  • Prime95 (CPU stress) runs stable for extended periods — doesn’t look like a raw thermal/stability issue under sustained load.
  • No minidump is ever generated
  • Checked Kernel-Power Event ID 41 in detail. It’s logged every time, but with: BugcheckCode 0, all Bugcheck parameters 0x0, PowerButtonTimestamp 0, LongPowerButtonPressDetected false, SleepInProgress 0, ConnectedStandbyInProgress false, WHEABootErrorCount 0. So: no preceding Windows-visible crash, no user action, no sleep transition — and notably, no WHEA-logged hardware error either. From the OS’s perspective it’s a completely traceless power interruption.

Now I have found out how to reproduce:

  • Running the built-in Far Cry 5 benchmark (4K and Ultra) reliably triggers the shutdown, every time, somewhere between 5–15 seconds after I start it. Interestingly, the Far Cry 6 benchmark at identical (Ultra) settings runs through cleanly and repeatably with no issue at all.
  • I thought: Okay maybe it is because Far Cry 5 is pretty old, so I downloaded Monster Hunter Wilds Demo. And I didn’t make it even through the setup procedure: Tried several times, and it always crashed at another point/setup-mask. One time even while everything was idle, not even the mouse was moved (system was waiting for me to configure brightness, but I had a phone call) - and crashed after 30 secs.
  • Btw. temperature does not seem to be an issue: most times between 70-80 Celsius for CPU, observed it on second monitor.

New finding: Windows Power Mode has a big effect

  • I had set the Windows Power Mode originally to “Best Performance”.
  • Now when I set it to “Best Efficiency”, I can run the Far Cry 5 benchmark multiple times back-to-back before a shutdown occurs. With the power mode set to Best Performance, the shutdown happens immediately, within a few seconds, on the very first run. This has been consistently repeatable in my testing.
  • And Monster Hunter does not crash any more.

My questions:

  • Does anyone else have comparable problems, i.e. random or reproduceable shut-downs as if someone pulled off the power cord?
  • Did you find out any helpful patterns?
  • And do you have an idea about that Windows Power Mode setting? - Is “Best Performance” not allowed for Framework Desktop?

Thanks, Holger

1 Like
2

What do you have plugged in to the desktop apart from keyboard, mouse, display? Try unplugging everything else.
Also when nvme devices do you have. Make, model?
If that does not help, contact FW support via their web site form. I may need a replacement mainboard.

3

I’m sure I’m far from the most knowledgeable person on this topic, but it reads to me like either the CPU might not be getting the voltage it needs in some spiky load scenarios or the CPU or memory might be at the limit of stability. It might be helpful to run a memory test.

Back in 2014, I built an HEDT workstation and overclocked it. It was stable in absolutely every activity I threw at it except VectorWorks CPU renders (which were why I built the system in the first place). I had to drop the overclock by 100 MHz, and then everything was fully stable. I’ve had similar issues with memory overclocking on my Zen 3 desktop. Loosening the timings a bit or lowering the frequency a bit (depending on the situation) resolved the instability.

Of course, you shouldn’t be having these issues, since I assume you’re not overclocking, undervolting, etc. That’s sort of the point of stock configurations.

It’s probably worth reseating any and every part and connector you can think of, just in case. I’m not sure if, say, a poorly seated M.2 drive could possibly cause this.

I can definitely imagine a power supply fault causing this if for some reason it’s not able to soak the transient load spikes. If possible, try reproducing it with a different power supply. I’m glad you at least have a specific test that reliably reproduces the issue.

4

Thanks for your quick reply!

My SSD is: WD_BLACK SN7100 2TB M.2 2280 NVMe. No second drive. Nothing else plugged aside from Mouse, Keyboard, 2x Monitor via DP.

What I also tried: Remove the WD_BLACK and put in a Samsung 980 Pro 1TB. Installed Windows 11 Pro from the scratch plus AMD Driver, and nothing else. Tried Far Cry 5 - same behavior: Shut-Downs in Performance Mode, quite stable in Efficiency Mode. Far Cry 6: All good. Monster Hunter: Couldn’t make it work in Performance Mode, works in Efficiency Mode.

Any other hints? Thanks!

5

See if there are any firmware upgrades for the NVME SSDs. Especially the WD_BLACK as there was an old firmware version that caused the symptoms you are seeing.
I have not seen reports of problems with the Samsung 980, so maybe that firmware update suggestion won’t help.

Linux seems to have more diagnosis options. Specifically something called the S5_RESET_STATUS register that you can read after a reboot, that gives some hint as to what went wrong.
I think there is a way to read the S5_RESET_STATUS in windows, but you will need to google/LLM a while for that. My knowledge is mainly Linux kernel stuff.

I am just another user like you, but I have been investigating these sorts of problems for a while. Search for FTH (Freeze then Halt) and FTR (Freeze then Reboot).
Yours is a bit new, being a Instant Power off without a reboot.

An instant power off, normally happens when the CPU detects something wrong, and wishes to limit corruption or damage to itself so switches off.

Have you contacted FW support via their web page. They only work weekdays, so you won’t get a response till Monday.

6

can you try run furmark 2? and tell us the result and temperature ?

7

Thanks a lot for your hints and replies!

  • Firmware for WD_BLACK SSD: Yes there was one available. Installed it, but no effect :frowning:
  • FurMark 2.1, setting: FurMark (VK), Resolution 4K UHD, FullScreen, 60 Seconds
    • Efficiency Mode: 2692 Frames, 44 FPS. Chip Power: 100W. GPU Temp: Stable at 70-72 Celsius all the time.
    • Best Performance Mode: 2755 Frames, 45 FPS. Chip Power: 140W. GPU Temp: Stable at 85-86 Celsius all the time.
  • Additional Observations while running FurMark:
    • GPU Temperature drops quickly after finishing the test back to ca 40 Celsius. So I assume cooling is ok?
    • When Starting in Best Performance, the Chip Power “jumps” for few seconds to 160 W(!) - I thought 140 is the maximum for that Chip?

One more crazy thing: I bought DOOM The Dark Ages. Okay maybe that is not very crazy, but the following observation:

  • Started Benchmark-Test in “Performance Mode”: Guess what - crashed after ca. 10 seconds.
  • Restarted computer, started Benchmark-Test again in Performance - and it completed! - And I could repeat several times. No crashes. Self-healing mainboard? Btw. Power Consuption stable at 140W, GPU-Usage 99-100%, GPU-Temp ca. 85-88 Celsius.

Any thoughts on that?

8

A shame the SSD firmware update did not help.

The AMD chips can peak, for very short periods, a lot higher than their average watts.
For example, we (other users like you) have measured a FW16 drawing over 350W.

During these peaks, if the PSU cannot handle the peaks, a brownout occurs and commonly results in the CPU switching itself off. The FW16 uses the battery to handle the peaks and it does quite a good job of that.

This is partly why the FW desktop PSU appears so over specified, but it is to handle the peaks.
So, I guess the next question is, which PSU are you using? (You mentioned which one, but I ask for comparison reasons)

Probably a good idea to compare your PSU specifications and look for what the peak figures are.
Some details on the FW desktop PSU are here to compare with:

One thing that adjusts how the CPU behaves is the PMT P3T setting. This seems to control the peak power draw a bit. Normally the EC sets this value. You can do “ectool console” soon after changing the profile from balanced to performance to see what the EC sets it to.
Noting that the FW16 has P3T set to something like 228W but peaks still can hit over 350W.

Unfortunately, FW do now say what the FW Desktop peak power needs are, so choosing a PSU is harder.

I think the only way to prove the brownouts would be to put a voltage probe on the 12V, 5V and 3.3V lines and continuously graph them. Then look at what was happening about the time of the self-power-off event. It might also be worth putting a Power Watts probe on the 12V, 5V, 3.3V lines to see how high it is peaking too.

On the FW16, there is a chip (the charger chip) attached to the EC that can measure all this, so we just modified the EC firmware to measure it.
I doubt the FW Desktop has a charger chip, so I don’t think just changing the EC firmware would help measure it.

I did some calculations of cost of PSU vs efficiency of PSU vs Electricity prices.
In general purchasing a Titanium PSU instead of a Gold PSU, paid for itself in reduced electricity cost is less than one year here in the UK. Note: Electricity costs a lot in the UK, so the cost/benefit might not be the same where you live.

I would probably suggest trying a bigger PSU, so it can handle the peaks a bit better might be worth a try, but saying that, at a guess, a 550W PSU should have been ok.

9

Thanks for your response and hints!

It is a be quiet! Pure Power 13 M 550W, the specs are here: PURE POWER 13 M | 550W silent essential Power supplies from be quiet!

I am not a technician, but as far as I see,

  • That PSU has a 12V Rail that can handle even 45.8A while FW only requests 32.5A
  • It has a peak Power of 600 W
  • And it says: “Pure Power 13 M offers 550W of continuous power with one massive 12V rail. Even if the graphics card demands excessive power for a short time: Pure Power 13 M 550W handles power excursions up to double its rated power with ease.”

To me it sounds good, that’s why I chose that PSU.

I do not have the possibility and the knowledge to put a voltage probe on the 12V, 5V and 3.3V lines and continuously graph them :frowning: But I might try a different PSU. Are there PSU which you would recommend?

10

I think we should find out the threshold of “brown out”. Is it 10V, 11V 11.5V or 11.9V? and combine it to the PSU’s “droop” (how much the voltage drop per W)

11

I agree. Your current psu reads like it should be ok for peak power draw.
Maybe it is just a bad mainboard and needs replacement?

12

Okay so I assume that it makes sense to open a ticket?

13

what is your ambient temperature ? 25C ? (my ambient is hot, 32C :melting_face: )

14

Well, I am in the North of Germany, so not too hot… except the last weeks/days. More than 25 but less than 30 Celsius here in my office… do you think it’s a thermal problem? Cause even in Performance Mode with 140W the temperature stays stable at ca. 85-88 Celsius (FurMark2), and drops quickly to 40C when finishing a test… is that still ok? What’s the maximum temperature that will cause the CPU to switch off?

15

Sounds like the problem I have been running after for months. The problem with this APU is that it has very spiky in power consumption and the motherboard does not have any capacitors to balance this out. For me the load triggered crashes stopped when I swapped the PSU to a 650W one.
I had a discussion with framework open for this as well and their PSU is actually higher rated that what they state on the specs. Still the have some chips that are very power hungry and thus cause crashes. They are working on a beefed-up PSU with more headroom to prevent such crashes.

16

ok. here is a interesting 1. today due to systemd problems. i installed older mxlinux 25.1

then i tried some benches. funny enough, on furmark 2.1, i used tuned + ryzenadj. the temp was only 75C (Tctl), the power draw was way up 150w+, i was able to hit 171fps.

im guessing it has to do with the kernel capability as well. previously on mxlinux 25.2 , the temp was similar but it only hit 140ish fps. i will retry again. maybe i missed something. im after all a newbie to strix halo. :squinting_face_with_tongue:

17

Hey there, if you haven’t found a solution yet I’d recommend opening a ticket with support. Thanks!

18

Yes I will open a ticket - as soon as I am back home, I am on the road until Sunday.

To everyone who has contributed here: Thank you for your input and support :slight_smile:

19

ok this is my i think 6? 7 ? try furmark ? but this is after re-applying TIM. i discovered mine came with dried TIM (the paste thing on the APU to “sticker” the heat sink). my furmark 2.1 ran 150w burst /140w @ 76C, my ambient is 32C. so i think, it could be a TIM issue. could be. i went with the TIM solution because i think my furmark ran too high and remember someone else on some forum chatter saying 65C constant load ? i thought that 65C is fantastic, but 85C would be too high.

re-applying TIM should be easy but on this heatsink it has multiple floppy silicone TIM for RAM/PS region. it is very floppy and i had to redo veryyyy sloowwwly with very light fingers, very light application of tweezers. :rofl:

right now im thinking, how to get it down to 65C :rofl: . water block :melting_face: (looks at empty wallet :money_mouth_face: )

20

regarding the sudden power draw, i think the issue is not just a matter of 12v cap reservoir, there has to be enough thermal buffer n current draw capability (and heat dissipation). in 1 sec say if we have capacity to do a 220w draw, the thermals have to go somewhere. the waste heat coming out is like faeces … :sweat_smile: .. we need to clear it immediately :rofl: … otherwise :rofl:

** to be clear, i changed the TIM after 2nd try at furmark. so about 2nd day of trying the motherboard. i have also ran the thought that the heatsink was faulty :rofl: