Bad Mainboard or Bad CPU?

[silkstone]

So I'm having issues when enabling boost on my Ryzen R5 3600 + Gigabyte AX370M-DS3H (aka AB370M-DS3H).

When boost is enabled, I'm getting crashes each day with the ntoskrnl.exe being the culprit (various error types: IRQ_NOT_LESS_OR_EQUAL, ATTEMPTED_EXECUTE OF NOEXECUTE_MEMORY, etc.)

I originally thought it to be my RAM, so switched it out and ran a full mem-test (no issues)

When I disable boost, everything runs fine and I feel that it is likely just that the mainboard doesn't want to work with the new CPUs. However, I have no idea how to go about testing whether it is my motherboard causing the issues or the CPU itself apart from going out and buying a better motherboard and testing with that.

Due to the randomness of the crashes, it's quite hard to duplicate, but it will crash 3-4 times minimum per day.

Does anyone have any ideas of how I can confirm which would be causing the issues?

 

[Basard]

BIOS is up to date? Is voltage set to auto for everything? Have you run optimized defaults for everything?

I'm not too familiar at all with Ryzen boards.....

 

[HD64G]

If normal settings are working properly, stay with those. Ryzen 3000 are made to work at their 99% of performance without OC or any boost from the motherboard settings. Maybe the VRM cooling on your motherboard isn't enough since you have wc on your CPU. Try to put a fan to send cool arit on the VRMs and chack stability again.

 

[Ware]

We can't tell by examining the CPU so testing with other parts is the most practical way to know.
I fix lots of really old PC's and a damaged CPU is very uncommon, it's way more likely to be the board.

 

[Splinterdog]

Dumb question: are you running the latest BIOS?

 

[theoneandonlymrk]

BIOS is up to date? Is voltage set to auto for everything? Have you run optimized defaults for everything?

I'm not too familiar at all with Ryzen boards.....

You want default voltages if you're having issues ,most boards overvolt on auto but then again it might be needing a bit more volts on boost , you could try running core volts offset + on auto , and possibly up load line control to level 3-4 too.
And try similar with Soc volts and LLc as well if the above didn't help.

 

[silkstone]

If normal settings are working properly, stay with them. Ryzen 3000 are made to work at their 99% of performance without OC or any boost from the motherboard settings. Maybe the VRM cooling on your motherboard isn't enough since you have wc on your CPU. Try to put a fan to send cool arit on the VRMs and chack stability again.

It crashes set to stock. I forget what AMD boost is called now, but it's default.

Everything else is up to data including the most current BIOS and everything at stock. The cooler is mounted properly and I have good temps.
I guess I'll just have to buy a new board and see if that fixes the problem

You want default voltages if you're having issues ,most boards overvolt on auto but then again it might be needing a bit more volts on boost , you could try running core volts offset + on auto , and possibly up load line control to level 3-4 too.

Thanks. I'll give that a go. One problem I have with the board is that it idles at 1.48v which seems a tad high. I'm a little reluctant to up to offset, but I can try the LLC setting, if the board has it.

I really wish I had read the reviews of this board before buying. I only didn't bother because I was only planning on using it with an Athlon 200GE and it had all the connectors I needed.
I'm thinking that it just can't deliver enough power on auto or defaults with the shitty 4-phases it has.

[Edit] No LLC seetings with this board However, I've switched the voltage control to Normal from Auto to see if that has any effect on stability. It also appears Precision Overdrive may have been enabled so I've turned that off in Ryzen Master.

 

[27MaD]

IRQ_NOT_LESS_OR_EQUAL

I've built a PC with an 8700K like 3 month ago and got this BSOD on the first run , all i had to do to fix it was to raise RAM voltage a little bit,

 

[tabascosauz]

If you think volts are high on light or heavy load, use a light offset (e.g. -0.05V). It doesn't really matter what the number is, because a lot of boards are proving to need a token offset to "kick" them into scaling voltage properly on boost. Without an offset, mine was pulling way too many volts and creating too much heat.

For some time, I had resorted to running 4.0 and 4.1 on lowest possible voltage as a manual, permanent clock setting, because leaving CPB on (boost on and subject to Ryzen algorithm, as stock) would cause all sorts of unwanted behaviour such as completely unresponsive and unopenable Windows Explorer, a whole load of unavoidable WHEA errors soon after boot (usually 3 of them, and usually the same culprits), and BSODs once in a while of the IRQ_NOT_LESS_OR_EQUAL variety.

Yesterday, I started testing with boost on again, but this time with PBO off, -0.1V offset, Standard LLC (my board goes from Auto-Standard-Normal-Low-High-Turbo-Extreme). Temps and voltages are much more under control, and so far no instability in IBT and P95 Smallest.

I would definitely give offset a shot, start small and see if it makes a difference, maybe it'll take the volts down.

Mine is the B450I Aorus, and is also 4-phase, albeit with much more respectable IR3556s instead of D-PAK 4C10s. I'd be more worried about the tiny heatsink over said 4C10s on that board - does your board report MOSFET temperatures? Even with IR3556s, mine can get really toasty if Vcore is too high.

From what I can tell in the last month, LLC doesn't really matter if you let boost do its thing. Say, if you were running a set clock like I was, you'd want more aggressive LLC to minimize the droop; for example, you'd have Vcore set at 1.3V, at idle it would be 1.29V, and at load it would droop down to 1.23-1.27V depending on what your LLC setting is.

But with Ryzen's boost algorithm, it's hard to tell what Vcore, in a traditional sense, is "set" at; at idle it's goes up to 1.4-1.5V, then at load it'll progressively drop with the amount of load applied, all the way down to about 1.206V for full load. But at the same time, most of idle is spent at 0.8-0.9V like you would expect, with occasional drops to 0.2-0.3V on deep idle; but also there can be less volts applied for partial load, I've seen values like 1.175V for load that's not completely stressing the CPU's capabilities. So the way Ryzen is boosting, it really throws a wrench into our traditional understanding of Vcore.

Use the CPU's reporting sensor for Vcore (SVI2 TFN) as it's much more accurate than the motherboard, and don't forget your Windows Power Plan. I'm running Windows Balanced (not Ryzen Balanced) with 0% minimum usage and 100% maximum.

 

[Frick]

Apart from the answers above, if actaul hardware failure the most likely culprits would be motherboard or PSU. I'm actually having slightly similar problems (not as frequent though) with the new build (in specs), so I'll definitely follow this.

 

[tabascosauz]

@silkstone actually, looking up Frick's board reminded me - the DS3H actually runs a 3+3 mode on its ISL95712. Not 4 Vcore phases, not 3 doubled Vcore phases, but just 3 Vcore phases. Combined with the typically mediocre 4C10s, I feel like this might actually be cause for concern in pushing a 3600 to 4GHz+ boost speeds at stock voltages. With 65W SKUs like the 3600 and 3700X, the CPU likes to gradually pull back on clocks and power consumption as a demanding application or stress test drags on for longer periods of time, but in short-term boost behaviour, you're still pulling up to 80-85W at max boost or close to max boost speeds. That's still quite a bit of power.

I think it might be in your best interest to find a B450 board with better VRMs. This is why I prefer mini-ITX to micro-ATX; on mini-ITX, especially for AMD, the lack of space forces vendors to opt for PowIRstages and Smart Power Stages, driving up the price but also providing much more robust power delivery. On micro-ATX, vendors are free to reach into their bin and pick out whatever ancient D-PAKs suit them and put them under a skimpy heatsink or none at all.

 

[Final_Fighter]

raise the ram voltage to 1.4v and see if it stops.

 

[silkstone]

I've built a PC with an 8700K like 3 month ago and got this BSOD on the first run , all i had to do to fix it was to raise RAM voltage a little bit,

Unfortunately it's not just that error. There are more:

This doesn't happen when Performance boost overdrive is disabled. Only when enabled

I might up the offset a little to see whether this helps. Strange thing is, it doesn't crash during AIDA stress testing. This happens at random times, when there should be little load on the CPU.
I also see that the vCore reported in ryzen master is pretty stable when boosting (3900 max!)

I might just go and buy an B450 AsRock board today and call it quits. I do enjoy tinkering and problem solving, but I won't have the time to do so for another month.

I can at least sell this board together with the original Athlon 200GE that I still have. Either that, or build a pfSense router or come up with another project for low-power parts.

 

[Cybrnook2002]

This is on your 400w inwin PSU?

 

[Mussels]

those errors scream memory problems

Why not list the RAM you have?

 

[tabascosauz]

@silkstone I've been seeing the WHEA errors pile up today from ntoskrnl.exe as well. I used to get them on bone stock settings, but don't get them if I have the manual clock and voltage settings. So I've deen dialing back the offset to see if it fixes the problem.

For the record, every single one of my BSODs during voltage testing was due to ntoskrnl.exe. Just wasn't enough voltage. But those were actually pretty rare, because the volts had to be grossly insufficient to trigger a BSOD; otherwise, P95 worker threads would individually stop, OCCT would return errors, or I'd have an extra WHEA pop up in HWInfo. Usually didn't get to the point of BSOD unless I was setting Vcore so low that even running Windows was a hard time for the CPU.

From what I can tell, it's not full load Vcore that is the issue, because that's just a simple BSOD. P95 runs just fine on all 16 threads, because the maximum droop is relatively static and is still sufficient to sustain the load (~1.2-1.3V depending on your offset). I suspect that the problem is when one or two cores are responding to activity and boosting to max speed (4.4 for me, 4.2 for you), then Vcore could be a problem because Ryzen handles voltage differently on light load. I've seen high (1.45V) voltages for those short bursts, and also low (1.17V) for similar bursts. Something tells me that -0.1V was too low for one of those instances.

Do you still get boost with PBO off, or is it like CPB off for me and just locks the CPU to base speed?

Regarding memtest, didn't you say that you ran a pass of memtest already? Why not try 4 complete passes, just to make sure? To be certain, boot memtest86 from a USB stick, not those "memtest.exe" methods from within Windows. You should always exhaustively test memory outside of Windows. P95 Large can help you double check afterwards.

 

[Bones]

So I'm having issues when enabling boost on my Ryzen R5 3600 + Gigabyte AX370M-DS3H (aka AB370M-DS3H).

When boost is enabled, I'm getting crashes each day with the ntoskrnl.exe being the culprit (various error types: IRQ_NOT_LESS_OR_EQUAL, ATTEMPTED_EXECUTE OF NOEXECUTE_MEMORY, etc.)

I originally thought it to be my RAM, so switched it out and ran a full mem-test (no issues)

When I disable boost, everything runs fine and I feel that it is likely just that the mainboard doesn't want to work with the new CPUs. However, I have no idea how to go about testing whether it is my motherboard causing the issues or the CPU itself apart from going out and buying a better motherboard and testing with that.

Due to the randomness of the crashes, it's quite hard to duplicate, but it will crash 3-4 times minimum per day.

Does anyone have any ideas of how I can confirm which would be causing the issues?

Normally those error messages are indication of not enough voltage to RAM or the CPU, possibly both.
Note I said normally.

Woudn't hurt to bump voltage to each a little and see if it starts working correctly, you'd have to set voltages manually for them to take effect. I see those, esp "irq not less or equal" all the time when benching and doing that usually takes care of it.
You've nothing to lose by trying anyway.

Normally 1.25v's to the CPU at stock speed will work. For RAM it depends on the sticks but typical B die sticks ran at 3200 only takes the same 1.25v's to be good to go. If that's what you have (B die) it should work for the RAM.

Was pointed out above it's possible you may just have a bad set of sticks but don't assume that just yet, try that and see what happens.

 

[micropage7]

i agree to check the ram first, single or dual ram? and try to setting the ram manually

 

[silkstone]

i agree to check the ram first, single or dual ram? and try to setting the ram manually

I thought it was RAM too.

I ran 4 passes on my Samsung B-die after putting it in my intel build - No issues.
I switched the 32gb hynix to my Ryzen build. The hynix ram has also been mem-tested on my Intel build - No issues. I'd previously been running this ram for 2 years with no issue.

Both were running at 1.35 V on the Ryzen. I even tried downclocking the ram and running it at 1.25 V, but was still getting the crashes

I can't manually set the Vcore on the AMD system, it has to be done via offset, and when raising it, I start getting up to 1.5 V idle which seems pretty high (I know AMD works differently, but still!)

I tried boosting just via core performance boost and disabling PBO last before bed last night and this is when these errors are from.

I am using a 400 W PSU, but measured from the wall, it draws only 90W idle and maybe twice that on load (not counting GPU). I will have a new PSU coming soon from RMA (650W) that I can throw in there, but the inwin should be sufficient, the reviews say that it is underrated at 400W and max is closer to 470 W . . . The same PSU is running a 980X system which is much more power hungry (with the same GPU) No issues whatsoever and they are both only 1 month old.

I'll give the offset a go now anyway and report back. The problem is, it happens at random times, not associated with times there is any load on the system. Stress testing for 2+ hrs everything indicates it is stable. It's when the CPU is idle, I get the crashes.

Edit - Testing with +0.03v now. I notice that it is still only boosting to 3.9Ghz which is way below where it should be The temp has also jumped by 4-8 degrees which is to be expected, i guess (I'm approaching thermal limits at times). Let's see whether it still crashes. The VRM temps are reported as only 65C on load by AIDA which seems low considering the quality of cooling I have in the case

 

[lexluthermiester]

Thanks. I'll give that a go. One problem I have with the board is that it idles at 1.48v which seems a tad high.

It is high. Drop it to 1.3v and give that a try.

 

[tabascosauz]

It is high. Drop it to 1.3v and give that a try.

This is not Intel; that's not even close to how Zen 2 works. 1.48 idle is normal behaviour; it doesn't actually put any current through the CPU when volts are that high because it's mostly idle. As load starts being applied, those volts start come down. He could set a specific manual clockspeed and a Vcore to match, but that's missing the point because full load is not what's in need of more voltage here. And 1.3V is pretty much what the CPU will be pulling at all core full load anyways, stock.

@silkstone If your CPU is under water and the VRMs aren't burning, it's a little strange as to why it's only reaching 3.9. Is this due to PBO being off, or is Ryzen smart enough to tell when the power delivery is of low quality?

One more thing; ever since I put CPB back on and let boost do its thing, I've found it hard to stress test just a few of the cores at a time. Obviously, it's easy to load either one core or go balls to the wall on all of them, but Ryzen behaves differently on Vcore with varying loads, so those stress tests don't always show you the instability you want to test. What seems to be a bit better is some game time in games that are modern but short of absolutely hardware-busting in their demands. That way, you can load a core or two for a good while and see if you get any crashes or WHEA errors.

About the RAM, the RAM may have been fine running memtest on your Intel system, but that's no guarantee that it'll work properly in the AM4; Ryzen's memory controllers have always been poor in quality compared to Intel's, even with this generation. That's why I suggested doing 4 passes in memtest on the AM4, and a stint in Prime95 Large to stress the RAM and UMC together. Furthermore, it seems that even though Ryzen 3000 UMC has made great strides in terms of compatibility and stability, the boards' topology and design does not seem to reflect that all of that improvement if you opt for an older board that isn't X570.

 

[silkstone]

This is not Intel; that's not even close to how Zen 2 works. 1.48 idle is normal behaviour; it doesn't actually put any current through the CPU when volts are that high because it's mostly idle. As load starts being applied, those volts start come down. He could set a specific manual clockspeed and a Vcore to match, but that's missing the point because full load is not what's in need of more voltage here. And 1.3V is pretty much what the CPU will be pulling at all core full load anyways, stock.

@silkstone If your CPU is under water and the VRMs aren't burning, it's a little strange as to why it's only reaching 3.9. Is this due to PBO being off, or is Ryzen smart enough to tell when the power delivery is of low quality?

One more thing; ever since I put CPB back on and let boost do its thing, I've found it hard to stress test just a few of the cores at a time. Obviously, it's easy to load either one core or go balls to the wall on all of them, but Ryzen behaves differently on Vcore with varying loads, so those stress tests don't always show you the instability you want to test. What seems to be a bit better is some game time in games that are modern but short of absolutely hardware-busting in their demands. That way, you can load a core or two for a good while and see if you get any crashes or WHEA errors.

About the RAM, the RAM may have been fine running memtest on your Intel system, but that's no guarantee that it'll work properly in the AM4; Ryzen's memory controllers have always been poor in quality compared to Intel's, even with this generation. That's why I suggested doing 4 passes in memtest on the AM4, and a stint in Prime95 Large to stress the RAM and UMC together. Furthermore, it seems that even though Ryzen 3000 UMC has made great strides in terms of compatibility and stability, the boards' topology and design does not seem to reflect that all of that improvement if you opt for an older board that isn't X570.

Well, no crashes so far with the increased offset. I can't manually set core on this board so I am seeing 0.996 - 1.500 v on Vcore. How high is too high on the Zen2's? or is there not really a limit?

 

[tabascosauz]

Well, no crashes so far with the increased offset. I can't manually set core on this board so I am seeing 0.996 - 1.500 v on Vcore. How high is too high on the Zen2's? or is there not really a limit?

The "limit" would be 1.5V when the system isn't doing anything, but I haven't seen anyone go over that yet because most people don't have to overvolt stock boost. But that doesn't really mean anything. On full load, you can still read SVI2 TFN and see your actual Vcore under load, but when not all the cores are being loaded fully, the voltages fluctuate so much and so quickly, it's pointless to try and make anything from it.

Your board is really not helping you either in not having even basic UEFI controls. If you're 100% confident it's not the RAM or the PSU, it might be time for a new board.

Spend a couple of days with the new settings, leave HWInfo in the background to log WHEA errors, and see if it's more stable. Don't forget to test that RAM to be sure.

 

[Mussels]

oh and if you have any corsair hardware be aware that HWinfo utterly breaks the corsair service, and you have to restart the service manually every time you open it (it freezes on last known readings/settings)
So many people treat that as an old resolved bug, but its back again with the latest icue software

 

[silkstone]

The "limit" would be 1.5V when the system isn't doing anything, but I haven't seen anyone go over that yet because most people don't have to overvolt stock boost. But that doesn't really mean anything. On full load, you can still read SVI2 TFN and see your actual Vcore under load, but when not all the cores are being loaded fully, the voltages fluctuate so much and so quickly, it's pointless to try and make anything from it.

Your board is really not helping you either in not having even basic UEFI controls. If you're 100% confident it's not the RAM or the PSU, it might be time for a new board.

Spend a couple of days with the new settings, leave HWInfo in the background to log WHEA errors, and see if it's more stable. Don't forget to test that RAM to be sure.

Thanks. If a 0.03 V offset is stable, I'll work on gradually reducing it.

I think I will get a new motherboard eventually anyway as a 3.9ghz boost is pretty disappointing.