Question Intermittent no-POST + disable iGPU doesn't stick - ASRock X670E Pro RS + 7950X3D + 4090: How to solve?

[romandesign]

I have built a new system with X670E Pro RS, Ryzen9 7950X3D and Gigabyte 4090. I have 2 very serious problems that I can't solve:

1. I can't disable iGPU. When I put iGPU to "disabled", then save and exit BIOS, the setting for iGPU doesn't stick! It works on the first boot (Windows Device Manager doesn't list the iGPU), but it resets from "Disabled" to "UMA_AUTO" after every restart! I tried with "detect discreet display" enabled and disabled - it doesn't make a difference. I have "prefer external GPU" set as well.

2. Often the PC doesn't post - the lights/fans are on but all monitors are black and there is no BIOS/POST screen or cursor, even if I wait for several minutes there's nothing. Monitor's backlight turns on but then I see "no signal" and they go to sleep. Pressing Reset 2-3 times gets me a cursor and normal POST, after which the PC is working perfectly. Also, the screens often don't wake up after a wake-up from sleep - the displays stay black. Only hard reset helps.

I have used the same 4090 for a couple of months in a previous system and it worked perfectly with no such problems. So it's not the GPU. I have updated BIOS twice and I'm now on the latest 1.24.AS02. PC is passing all stress tests (CPU and memory) running for ours, so it's not RAM or any CPU settings (I also tried running stock settings on both - same problems). When it's working, it runs perfectly stable. I think something's fishy with the display detection: when I get the POST screen, it is randomly coming on any one of my 3 monitors. I suspect that often it doesn't detect any monitors (or doesn't wait long enough) and then I have black screens and no POST, and maybe it also re-enables the iGPU settings. But again, GPU is fine, tested on the previous PC. The PSU is 1250W and also used on the previous PC and is perfectly fine. I don't get any beeps when I have no POST.

I don't know what else to try. This is extremely annoying: it takes me 10 minutes sometimes just to turn on my PC. Then if it sleeps and the monitors don't wake up, I have to hard reset which can't be good and can result in OS corruption. I also don't want iGPU using my RAM and want to disable it permanently. Having it on results in a known bug: it's Gigabyte Control Center doesn't see the 4090 GPU at all and can't do anything. If I actually enable iGPU and install Adrenaline drivers, they conflict with NVidia drivers and Windows just crash. Also a known bug. Leaving only a single monitor running doesn't solve the problem.

I built a top-of-the-line PC expecting it to work, but it's a minor disaster so far.

 

[zx128k]

Yes, I think that's may be what happens. Just setting RAM at default profile (4800) didn't help, but both forcing "restore memory context" off AND memory at default so far gave me 3 good cold starts, no beeps, no POST screen though, boot into Windows. I'll give it a couple of days and if this keeps working fine, I'll see if I can find memory settings that would work with XMP.

The memory is G.SKILL S5 rated at XMP 6000 CL32 @1.4V - not EXPO, only XMP. But it totally refused to work on XMP profile with this MB. Raising the voltage only made it worse. Curiously, lowering VDD, VDDQ, VPP from 1.4V to 1.3V AND rising the other voltages (SoC 1.29V, VOD 1.14V, VDDG CCD/IOD 1.29V, VDDP 1.19V) makes it work fine, even at CL30 without loosening timings, with all TestMem5 etc. tests passing fine, and no crashes/errors in work and gaming. But it's possible that this combination of voltages (which took me a very long time to find) doesn't play nice with the MB BIOS at POST. Maybe I can raise all voltages a bit and see if it's still stable.

I know I should have bought an EXPO memory from the list for my MB, but it's so expensive, and this one was a good deal. I was expecting the CPU within a week of ordering RAM and though if it doesn't work I'll return it. But the CPU took 1.5 months to come and the return window closed, so I'm such with this RAM.

I had the same issue with my DDR4 overclock but when I would shutdown or restart it would also hang with a black screen. Fixing the voltages fixed the issue. For me VCCIO/VCCSA(10900k) was too high. Just be careful with SoC voltage on AMD chips (<1.3volts).

 

[romandesign]

Nope, I still sometimes get 3 beeps with RAM on jedec clock and "context memory restore" disabled. Not sure what else I can try...

 

[zx128k]

Check one stick at a time and look in the manual to make sure you have the RAM in the right slots. Try one stick at a time (page 27 2.3 Installing Memory Modules (DIMM)). Try reseating the run and/or cpu.

Is your memory in the QVL? What is your memory kit? Serial number everything.

Is it 3 long beeps or 3 short ones. User manual. What does the Post Status Checker show after the 3 beeps? page 57 2.14 Post Status Checker

Do you have any USB sticks connected? Does the issue happen with no drives connected.

Also it takes some time for DDR5 to train. Sometimes my monitor only turns on for the windows login screen.

 

[romandesign]

Check one stick at a time and look in the manual to make sure you have the RAM in the right slots. Try one stick at a time (page 27 2.3 Installing Memory Modules (DIMM)). Try reseating the run and/or cpu.

Didn't try with 1 stick, but swapped them a couple of times, they are 100% in the right slots (A2, B2 I think, 2nd and 4th slots).

Is your memory in the QVL? What is your memory kit? Serial number everything.

Not in QVL. It was a good deal at the times, and over 10 PCs I built I could always make any RAM run without problems, even on modest OC, maybe on slightly bumped voltages. I was going to test and return it if it dodn't work, but then had to wait in line for the CPU for 1.5 months instead of a few days, so return window ended.

G.SKILL Ripjaws S5 Series 64GB (2 x 32GB) 288-Pin PC RAM DDR5 6000 (PC5 48000) Desktop Memory Model F5-6000J3238G32GX2-RS5K
I don't know how to check the serial number.

Is it 3 long beeps or 3 short ones. User manual. What does the Post Status Checker show after the 3 beeps? page 57 2.14 Post Status Checker

3 short beeps, the beep codes are not in the manual. I googled and 3 beeps is memory error. I have 2LEDS staying red , I think they are DRAM/GPU. I'll look it up next time. Even without beeps they always stay on the longes, then go out on POST, which I understand is normal. If 3 beeps happens, system resets automatically, or if not - those 2LEDS stay lit.

Do you have any USB sticks connected? Does the issue happen with no drives connected.

Also it takes some time for DDR5 to train. Sometimes my monitor only turns on for the windows login screen.

No USB sticks connected. I know the training time, if I get no POST - it stays black indefinitely. Past dew minutes. When it works though - it works fine.

Initially, because DRAM won't run stable on XMP profile, it took me a while to find a running stable voltages combo for the DRAM to run at 6000 CL30. It did it only on lowered first 4 voltages (1.3V instead of XMP 1.4V) and raising second 4 voltages. I was thinking that maybe my instability happens because of low voltages - though stock voltages are also giving 3 beeps, but they are low too (1.3V), so maybe raising the voltages a bit without making DRAM crash would help and make it more stable? I now upped VPP voltage to 1.8V (from 1.5V) which is XMP profile value, and upped VDDIO,VDO, VDDQ to 1.31 from 1.30, Soc to 1.295V from 1.29V, and it passed the TestMem5 stress test. So I'll run it that way for a while and see if it makes any difference.

 

[zx128k]

Try FCLK Frequency 2033. 2067 and 2100 are harder to get running. Memory Frequency to DDR5-6000. Timings are DDR5-6000 CL32-38-38-96 1.40V. Td = 32. Trcd = 38, Trp =38, Tras 96.

From the video.

DRAM Voltage Control = VDD/VDDQ/VDDIO
VDD_SOC = 1.3 volts. Comment from forum 1.250 V + 0.1 = 1.35 which is > 1.300 SoC, so you actually are within the recomendated VDDIO +0.1V > SoC. Don't go higher than 1.3 volts.
VDDIO Voltage = 1.25 volts lower can sometimes help with stability. Signals.
DRAM VDD = 1.25 volts Actually hardcore overclocking sets 1.4-1.42 volts. Internal DRAM voltage.
DRAM VDDQ voltage = 1.25 volts lower can sometimes help with stability. Signals.
DRAM VPP Voltage 1.8 volts
UCLK DIV1 MODE = UCLK=MEMCLK

VDD Misc Voltage 1.10 volts
VDDG CCD Voltage = 1.050 volts
VDD IOD Voltage = 0.950 volts
VDDP Voltage 1.100 volts

Memory Context restore = disabled.

The main issue is that voltages above 1.25 volts can cause issues with memory training which could be the issue here. VDDIO/DRAM VDDQ > 1.25volts.

Personally I would change VDDIO Voltage = 1.25 volts, DRAM VDDQ voltage = 1.25 volts and test. See if the training issues go away.

 

[romandesign]

Try FCLK Frequency 2033.

I was running 2033 already, but now I"m running 2000 so see if it's more stable.
Thankyou for the analysis of the voltages, here's exactly what I'm running now:

So you're saying that I should try to lower VDDIO and VDDQ to 1.25, leaving the rest as is, correct? Should I put 2033 back?

My timings are basically the XMP CL32 profile timings, but running at CL30. Trying to tighten the secondary timings to a popular video ones, got me errors, so I left them as they are. But just lowering CL32 to CL30 worked OK.

For some reason entering 1.3 into SoC resets to 1.295, some kind of limit is there I guess, but it's almost the same.

 

[zx128k]

1.3 SoC is the maximum.

Basically lower VDDIO and VDDQ to 1.25. Then hope that fixes the black screen on boot issue and you stay stable. Also the infinity fabric frequency should be between 2033-2100 as video states this gives good performance. If IF is bad you get restarts and errors.

Also the other voltages are way too high. This is what the video used for DDR5-6200

VDD Misc Voltage 1.10 volts
VDDG CCD Voltage = 1.050 volts
VDD IOD Voltage = 0.950 volts
VDDP Voltage 1.100 volts

You don't want the i/o die getting too hot.

You likely dont need all that voltage. Too much can make things worse. You have to have primaries stable and all issues fixed before moving onto secondary timings.

 

[romandesign]

Basically lower VDDIO and VDDQ to 1.25. Then hope that fixes the black screen on boot issue and you stay stable. Also the infinity fabric frequency should be between 2033-2100 as video states this gives good performance. If IF is bad you get restarts and errors.

Also the other voltages are way too high. This is what the video used for DDR5-6200

VDD Misc Voltage 1.10 volts
VDDG CCD Voltage = 1.050 volts
VDD IOD Voltage = 0.950 volts
VDDP Voltage 1.100 volts

I tried the settings exactly as you suggested, and it booted fine, then passed two 40m runs of MemTest5, so it looks stable to me. I'll keep it as is and see if it resolves my problems. Thanks for the advice.

Update: tried to bump VDD to 1.4V (XMP rated), leaving the rest of the voltages the same as your recommended settings. It passed the MemTest5 but then rebooted on idle within an hour. So I'm back to 1.31 and will see if it's fully stable. Then if it is, I'll try to up it a bit and see if it helps me to tighten timings etc. according to that video.

 

[zx128k]

See if your booting issue has gone away. You will need to run more tests later to make sure primaries are stable but so far so good. If we can work out the settings that stopp the no boot issue, then we have a way forward.

 

[romandesign]

See if your booting issue has gone away. You will need to run more tests later to make sure primaries are stable but so far so good. If we can work out the settings that stopp the no boot issue, then we have a way forward.

So far so good. I had 3 cold starts yesterday and one this morning and it posted and booted fine, I never got black screens. I'll see for another couple of days without changing anything, and if it holds I will see if I can up VDD a bit to get more headroom with timings. I know that 1.4V is not stable - both with my old settings, and with new ones. 1.31V is stable. Maybe I can try 1.35V or somewhere between 1.31 and 1.4. It wasn't stable with my old settings, but now all other voltages are different, so maybe I can get VDD higher. If I can, maybe I could tighten the timings a bit, or push the frequency to 6100, or both. Though I'd be happy even if it runs as it is (6033 @CL30), as long as I get no black screens on POST and it's stable.

 

[zx128k]

Once the no post issue is fixed then you can push the frequency if you want too.

 

[uKER]

Good god, I'm having the exact same scenario as you with:
- ASRock B650M Riptide
- Ryzen 7600X
- 32 GB GSkill 6000 CL30 (EXPO)
- RTX 3070 Ti
I get both the inability to disable iGPU and the occasional boot to black screen.

Here's this other guy with the same issues. Seems like it's an issue with AM5/Ryzen 7000.

Reddit - Dive into anything

www.reddit.com

So you didn't find any solution other than fiddling with the voltages?

 

[romandesign]

Here's what I have now, it works fine that way.

 

[FAhentai]

Question - Asrock X670E RS PRO nightmare ?

Hi guys, you can call me Hentai. I am an Australian-Chinese who live in Melbourne Australia. Recently, I had some really rared problem, together with Australia local warranty, plus some other parts went down, finally everything become a total nightmare. :ROFLMAO: Now the problem gets...

forums.tomshardware.com

Mate, you are not the only one who has problem with this setup.
Clearly it seems like I really get QC issue here.