Hard resets playing games

[Space Lynx]

Yes, I forgot to update, but I tried the above voltages. Same issue. I've also had the issue occur with both sets of memory at non-DOCP. New set is GSkill.

Yes, I don't think it's memory related at this point. I also did try disabling PBO and it still occured.

I guess the question is what I should try next? Mobo would be easiest to try since it's still within Amazon's return window. CPU I got direct from AMD so I have no idea how long that process will take. But their shipping was SLOW so probably a while. I have an older PSU (6 year old EVGA G2 750W that has worked flawlessly) if anyone things that could make any difference whatsoever.

I honestly don't know what your next step is. that sounds really frustrating

 

[tabascosauz]

Yes, I forgot to update, but I tried the above voltages. Same issue. I've also had the issue occur with both sets of memory at non-DOCP. New set is GSkill.

Yes, I don't think it's memory related at this point. I also did try disabling PBO and it still occured.

I guess the question is what I should try next? Mobo would be easiest to try since it's still within Amazon's return window. CPU I got direct from AMD so I have no idea how long that process will take. But their shipping was SLOW so probably a while. I have an older PSU (6 year old EVGA G2 750W that has worked flawlessly) if anyone things that could make any difference whatsoever.

The issue is not with the RAM itself. You've already done your memory stability testing. Bus/Interconnect can only be one thing - the Infinity Fabric on the CPU. Not the cores on the CPU, not the memory controller on the CPU.

Unfortunately, on Ryzen, "RAM stability" encompasses many things. The RAM itself needs to be stable. The UMC on the CPU needs to be happy. The IF on the CPU has to be happy.

The only things you can really do are disable C-states, and play with your VSOC and CLDOs. Technically, to address the IF the correct approach is to search for "DF Cstates" and disable that, but the Global C-states option also works.

There are many different schools of thought when it comes to using VSOC, VDDG and VDDP to avoid Bus/Interconnect. The only thing that connects them all is that VDDG CCD, VDDG IOD and VDDP cannot be higher than 0.04-0.05V below VSOC, as they are derived from it.

• Run all 4 as low as possible. At 3600 you could try 1.0V VSOC, 0.95V for both VDDGs, and 0.9V VDDP. That's not a limit, you can go as low as you want. There are more veteran tweakers out there running 3600 RAM on less than 1.0V VSOC, and CLDOs in the 0.8V range or even lower.
• Run VDDG CCD @ around 0.05V below VSOC, then VDDG IOD as low as possible.
• Run VDDG IOD @ around 0.05V below VSOC, then VDDG CCD as low as possible. This is what fixed my terminally ill 3700X in the last few months I owned it. Bus/Interconnect were previously an inevitable part of life.
• Run both VDDGs @ 0.05V below VSOC.
• Run VSOC higher, up to a limit of 1.2V. Usually doesn't work.

If you don't want to spend this much time to troubleshoot these WHEAs (perfectly understandable), then RMA is the way to go. Every CPU behaves differently even with the same RAM kit and motherboard. The 1804 BIOS comes with AGESA 1200 and is the most recent stable BIOS.

 

[Deleted member 205776]

Although at stock it should run with auto setting without issues.

Some boards set the VDDG voltages way too high, namely VDDG IOD. Mine was auto set to 1.05v which is just excessive for my RAM. I reduced it to 0.950v and I have yet to see a WHEA error, whereas otherwise I would see one every hour, with C-States on. Auto isn't always the best.

 

[Deeveo]

Some boards set the VDDG voltages way too high, namely VDDG IOD. Mine was auto set to 1.05v which is just excessive for my RAM. I reduced it to 0.950v and I have yet to see a WHEA error, whereas otherwise I would see one every hour, with C-States on. Auto isn't always the best.

True, have the same issue on my Asus board. Although it should run at lower voltages when the memory frequency is lower.

 

[flashfrenzy]

Thanks for all the help! I really appreciate you all helping a Ryzen newbie. Here's what I have with the new RAM. I see no WHEA issues on startup so far. Will see if stability is helped at all. I don't mind fiddling with this stuff but if I can't get a stable system on what should be a trivial memory overclock I'll probably go the RMA route.

 

[Deleted member 205776]

Thanks for all the help! I really appreciate helping a Ryzen newbie. Here's what I have with the new RAM. I see no WHEA issues on startup so far. Will see if stability is helped at all. I don't mind fiddling with this stuff but if I can't get a stable system on what should be a trivial memory overclock I'll probably go the RMA route.

Everything looks good. Do let us know if the situation improves.

 

[tabascosauz]

Thanks for all the help! I really appreciate helping a Ryzen newbie. Here's what I have with the new RAM. I see no WHEA issues on startup so far. Will see if stability is helped at all. I don't mind fiddling with this stuff but if I can't get a stable system on what should be a trivial memory overclock I'll probably go the RMA route.

Okay, so the new kit is CJR, good. Settings look good for XMP, but you should still test the memory when you have time just to make sure RAM is out of the picture. Run HCI Memtest up to at least 1000-3000% coverage when you have the time.

Thing about IF stability is that there's basically no test for it aside from the test of time. Whether that's a good or bad thing, well, that's up to you. Just gotta use your computer as you would normally and keep an eye on the HWInfo error counter occasionally.

 

[flashfrenzy]

So I'm getting quite a few of these now. But no hard resets so far.

 

[tabascosauz]

View attachment 193748

So I'm getting quite a few of these now. But no hard resets so far.

Probably don't want VSOC and VDDP to be too low either. Bump them up to 1.1V and 0.9V. The hard resets are most likely the same cause - unfortunately, fatal errors don't leave a log because WHEA doesn't have time to log them, the ones you see are all corrected non-fatal errors.

Asus has been putting out new beta BIOSes for AGESA 1201. Version 2006 for your board, I think; your current 1804 is AGESA 1200. Normally I wouldn't recommend beta BIOSes but you haven't much to lose at this point.

Have you tested your memory yet?

In the interest of curiosity, try running 3200 with those otherwise default XMP timings.

 

[flashfrenzy]

Probably don't want VSOC and VDDP to be too low either. Bump them up to 1.1V and 0.9V. The hard resets are most likely the same cause - unfortunately, fatal errors don't leave a log because WHEA doesn't have time to log them, the ones you see are all corrected non-fatal errors.

Bumping these up seemed to take care of the non-critical errors. Unfortunately I am still getting resets and have seen at least one critical "Bus/Interconnect" error.

Have you tested your memory yet?

Yes, I've tested both sets of memory pretty extensively with a few kinds of memory testing programs. I haven't seen any issues with either kit in the memory tests.

In the interest of curiosity, try running 3200 with those otherwise default XMP timings.

I tried this with the first kit when I thought the issue was with XMP. But since I get the resets without XMP enabled on all auto for both kits I didn't bother testing this case with the second set.

At this point, I feel like it's probably time to try a new CPU. I opened a warranty request with AMD.

 

[JeffF]

Hard freezes/crashes and etc can be a power issue with the PSU and demanding hardware on a Ryzen system. I recommend always going into the Bios and in your BIOS (UEFI) configuration, find the following setting:
PSU Idle Control

Change it to:
Typical Current Idle
It has solved many problems as far as freezing issues yet other hardware issues such as USB disconnects and Etc.

 

[Space Lynx]

Hard freezes/crashes and etc can be a power issue with the PSU and demanding hardware on a Ryzen system. I recommend always going into the Bios and in your BIOS (UEFI) configuration, find the following setting:
PSU Idle Control

Change it to:
Typical Current Idle

This has helped me in many other systems. With Ryzen systems it should be set this way by default yet it's not and the system not only has freezing issues yet other hardware issues such as USB disconnects and Etc.

I might try this myself... mine is set to auto.

 

[milewski1015]

I might try this myself... mine is set to auto.

Me too. Never heard of this before. Will have to see if it fixes the audio disconnect issues I was having

 

[tabascosauz]

Bumping these up seemed to take care of the non-critical errors. Unfortunately I am still getting resets and have seen at least one critical "Bus/Interconnect" error.


Yes, I've tested both sets of memory pretty extensively with a few kinds of memory testing programs. I haven't seen any issues with either kit in the memory tests.


I tried this with the first kit when I thought the issue was with XMP. But since I get the resets without XMP enabled on all auto for both kits I didn't bother testing this case with the second set.

At this point, I feel like it's probably time to try a new CPU. I opened a warranty request with AMD.

Have you tried Global C-states off yet or stepping down to a lower FCLK freq at the same memory speeds to see if the WHEAs persist? You don't have to change anything else just change the FCLK freq in the dropdown.

There's also the new 2006 beta BIOS if you haven't tried that yet, as I said earlier.

The Low Current Idle option is a good suggestion, but as a solution it usually pertains to core-related WHEAs like Cache Hierarchy, not the IF, especially since your WHEAs are exclusively Bus/Interconnect. With C-states on, the Zen 2 cores are aggressively power gated and parked; I'm pretty sure the data fabric doesn't idle nearly as low as the cores as it always draws 10-15W. Doesn't hurt to try at this point though.

The RMA is a good call. They may give you a checklist of things to try before your RMA is approved:

 

[Deleted member 205776]

Hard freezes/crashes and etc can be a power issue with the PSU and demanding hardware on a Ryzen system. I recommend always going into the Bios and in your BIOS (UEFI) configuration, find the following setting:
PSU Idle Control

Change it to:
Typical Current Idle

This has helped me in many other systems. With Ryzen systems it should be set this way by default yet it's not and the system not only has freezing issues yet other hardware issues such as USB disconnects and Etc.

That didn't help my reboots with C-states on, going through 3 different PSUs. I was really hoping it would.

What did, was lowering VDDG IOD by -100mV. No more reboots at idle or WHEAs. C-states on, Low Current Idle < these two are on right now, and before the voltage adjustment, this combo would've given me tons of reboots

 

[flashfrenzy]

Hard freezes/crashes and etc can be a power issue with the PSU and demanding hardware on a Ryzen system. I recommend always going into the Bios and in your BIOS (UEFI) configuration, find the following setting:
PSU Idle Control

Worth a shot. I just updated this. None of my reboots are at idle, all under game load.

Have you tried Global C-states off yet or stepping down to a lower FCLK freq at the same memory speeds to see if the WHEAs persist? You don't have to change anything else just change the FCLK freq in the dropdown.

Yes, Glocal C-states is off as well as PBO. So just lower FCLK but not keep the 1:1 ratio? I can try that. Even the incredibly loose timings of auto didn't work though which is at like 1033mhz (although with corresponding memory timing changes).

There's also the new 2006 beta BIOS if you haven't tried that yet, as I said earlier.

Yep, that's next on my todo list.

 

[tabascosauz]

Worth a shot. I just updated this. None of my reboots are at idle, all under game load.

Yes, Glocal C-states is off as well as PBO. So just lower FCLK but not keep the 1:1 ratio? I can try that. Even the incredibly loose timings of auto didn't work though which is at like 1033mhz (although with corresponding memory timing changes).

Yep, that's next on my todo list.

Yeah, just to see if the IF is running into trouble at 1800MHz. IIRC I'm 90% sure that if you set the IF in the dropdown to something other than what it "should" be at 1:1, the Asus BIOS should be smart enough to keep memory the way it is and just desync.

At JEDEC 2133 spec the IF is indeed running at 1066MHz, but it's also running a different set of VSOC/VDDG/VDDP because the board will command different auto voltage rules depending on your memory speed under XMP. Maybe trying 1066MHz while keeping those voltages where they are will yield some different results.

I just noticed that I missed this from the OP:

Since then, I haven't seen any restarts but I've been running prime95 tests and always run into errors within 2 hours.

If Prime95 is losing worker threads, then the cores aren't stable either. Even if Tctl is getting to 90C (5800X is a warm one) and the CPU is throttling to lower speeds due to the 90C thermal limit, P95 should never drop workers.

Between the B450I Aorus Pro Wifi and B550M TUF, I had a B550M Steel Legend that lost both CPU and DRAM stability as soon as memory freq exceeded 3200. Below 3600 or at JEDEC, it would stress test just fine all day. Nothing wrong with my 3700X (well...) and RAM both of which were carryovers, it was either the board or its firmware that just couldn't figure out what it wanted to do.

Let's see if the beta BIOS changes any of this behaviour.

 

[flashfrenzy]

If Prime95 is losing worker threads, then the cores aren't stable either. Even if Tctl is getting to 90C (5800X is a warm one) and the CPU is throttling to lower speeds due to the 90C thermal limit, P95 should never drop workers.

To be clear, Prime95 was failing at JEDEC too, not just with DOCP enabled. Want to make sure I understand, are you suggesting that possibly the board is just unstable and the CPU is fine? Or that testing at JEDEC speeds needs to be done, just with modified voltages?

I haven't experience a reset since changing the PSU Idle Control setting. When/if that happens, trying the beta BIOS is next on my list but that means a clean CMOS reset and reconfiguring everything, most likely.

-edit-
(5 min after posting, system reset )

 

[tabascosauz]

To be clear, Prime95 was failing at JEDEC too, not just with DOCP enabled. Want to make sure I understand, are you suggesting that possibly the board is just unstable and the CPU is fine? Or that testing at JEDEC speeds needs to be done, just with modified voltages?

I haven't experience a reset since changing the PSU Idle Control setting. When/if that happens, trying the beta BIOS is next on my list but that means a clean CMOS reset and reconfiguring everything, most likely.

-edit-
(5 min after posting, system reset )

P95 dropping workers at stock settings? Probably board or board firmware, I've seen CPUs throw dreaded Cache Hierarchy but not yet any that fail stock boost stability. The Bus/Interconnect WHEAs? Probably could be either board firmware or the CPU.

Again, how the AGESA 1201 BIOS behaves will probably inform what your next step should be. When you upgrade to the new BIOS you should probably start at Auto settings to establish a baseline before tweaking the CLDOs.

I hate to say this but if you want to/have to tweak Ryzen you better get used to entering your BIOS settings. AMD doesn't understand the concept of releasing definitive, polished firmware alongside the product, so we're constantly getting improvements and changes to AGESA every few months, so new BIOSes all the time.

 

[flashfrenzy]

I hate to say this but if you want to/have to tweak Ryzen you better get used to entering your BIOS settings.

Yeah, that I don't mind, even for something like DOCP. What is frustrating is that I'd expect just putting the PC together to generally work without having to tweak voltages. I really just want a stable baseline to work off which I haven't been able to get. Granted, if a part actually is faulty, stuff like this happens.

In any case, I updated to the beta BIOS. No resets so far but haven't had much chance to test. I did run P95 at stock BIOS settings and a worker failed in 30 min.

 

[tabascosauz]

Yeah, that I don't mind, even for something like DOCP. What is frustrating is that I'd expect just putting the PC together to generally work without having to tweak voltages. I really just want a stable baseline to work off which I haven't been able to get. Granted, if a part actually is faulty, stuff like this happens.

In any case, I updated to the beta BIOS. No resets so far but haven't had much chance to test. I did run P95 at stock BIOS settings and a worker failed in 30 min.

Are you able to stress test the cores with something else like OCCT or LinpackXtreme? Not disregarding the P95 crashing, because that's still certainly not good news.

For P95, are you testing Small FFT or Blend? Blend mixes the different tests together and may not be indicative of a CPU core issue, as the Large FFT test primarily stresses memory controller and IF.

It would help if you had an extra AM4 board you could use to test this CPU. That would save you a bunch of time trying to guess between the CPU and board firmware.

If you're sure this is happening on stock settings, it sounds like this CPU needs to go back to AMD. Where did you buy the 5800X? Are you able to make use of a 2-week or 30-day return period if they still have availability for it? That's going to be much easier than trying to RMA.

All I can say is, Ryzen has been quite hit-or-miss since Ryzen 3000, whether the QC on the hardware itself or firmware.

 

[flashfrenzy]

Are you able to stress test the cores with something else like OCCT or LinpackXtreme?

Sure, I can run either. Is there a particular test that will help narrow down the issue?

For P95, are you testing Small FFT or Blend?

So far I've just run blend.

It would help if you had an extra AM4 board you could use to test this CPU.

Yeah, I wish I did. This is the first AM4 system I've put together so I have no useful spare pieces.

Where did you buy the 5800X? Are you able to make use of a 2-week or 30-day return period if they still have availability for it? That's going to be much easier than trying to RMA.

I bought directly from the AMD store. It seems like their return policy is effectively the warranty return process. I fully expect it will take a while.

Thanks so much for all the help, btw!

 

[tabascosauz]

Sure, I can run either. Is there a particular test that will help narrow down the issue?

So far I've just run blend.

Yeah, I wish I did. This is the first AM4 system I've put together so I have no useful spare pieces.

I bought directly from the AMD store. It seems like their return policy is effectively the warranty return process. I fully expect it will take a while.

Thanks so much for all the help, btw!

It's up to you but I'm really not a fan of Blend. Run Smallest or Small FFT for cores, run Large FFT for the UMC. Buildzoid talks about the settings he runs for P95, you can copy some of those, I think he settled on a particular custom FFT length that's particularly demanding

Can't tell you much regarding OCCT because there's been a big recent overhaul of the program, but the UI is pretty intuitive. As for LinXtreme, it opens in command line, use the Stress Test option, 10GB, as for # of iterations run 1 or 2 to begin to get a feel for how long each is, then scale to how long you want the test to be. I'd just run it about half an hour, LinX gets pretty hot.

 

[flashfrenzy]

OK! So this beta BIOS is promising! No resets at default settings and not a single WHEA error yet! It's a little disconcerting that P95 still fails (both small FFT and large FFT) so only time will tell if that really is an indicator of an unstable system. Every other stress test I've run has come out clean (but that was true with the old BIOS version as well). I'm going to try getting IF and memory speeds up over the weekend and see how that goes.

 

[Deleted member 205776]

OK! So this beta BIOS is promising! No resets at default settings and not a single WHEA error yet! It's a little disconcerting that P95 still fails (both small FFT and large FFT) so only time will tell if that really is an indicator of an unstable system. Every other stress test I've run has come out clean (but that was true with the old BIOS version as well). I'm going to try getting IF and memory speeds up over the weekend and see how that goes.

Small FTT straight up reboots my system if I don't disable AVX. I wouldn't be too concerned if everything else is stable and you're not getting bluescreens.