Ryzen 1800x BSODs & Prime95

Can anyoner give me some guidance on my Ryzen7 1800x?
Running on -
Asus PRIME B350M-A (BIOS 3803)
with
Team Group Dark Pro "8Pack Editon" 16GB (2x8GB) DDR4 PC4-25600C14 3200MHz Dual Channel Kit (Running @ 14-14-14-31 3200mhz);
&
ID-Cooling Auraflow 240mm

The mobo auto OCs to 3850 but I have tried 3600 & 3700 with no real changes in temps or in crash frequency. Win10 updates have made this build really erratic - one week I was OK & thought I had it licked, then win updates & not anymore...

I can pass Memtest86 in DOS but I cannot get past 20 min with Prime95 torture test without a restart/BSOD.
Similarly the AIDA64 test won't run past 20mins without halting for an error

Temps are OK being reported as <70C when under 100% load by Ryzen Master.
Mobo temps never seem to get above 22-23C.

Handbrake encoder will run OK for 30-50 mins, but any longer & I get a prog crash or PC restart. It is drawing 98-100% CPU so it is 'obviously' related to load.

I have been watching the values of HWMonitor & AIDA64 & I've noticed the Powers > CPU Package gets to a peak just before the halt/error occurs.
e.g. I can see a CPU temp value of 65-70C & a Package value of 116-133w when running a 100% stress test.
Anyone know how this value relates to TDP / CPU tolerance?

Should I be looking at reducing the Package wattage - & if so how do I do that?

My main problem is I want to be able to run Handbrake without crashes. The rest of the stuff I'm doing (games or regular work) seems fine now I have the RAM where I want it.
I was still getting BSODs with 2133 RAM in Win10 x64 - all sorts of things from DX12 to kernel to bad RAM addressing. Now when I look at dmps it's almost always kernel & core win sys files.

I attach memtest & HWMon screen shots & my RAM timings.
I have my core cpu v @ 1.35 with offset/+/auto which peaks <1.42

A quick guess, your VRMs are most likely overheating. Most B350 boards, while support overclocking, are not really overclocking friendly especially for 8-cores because the VRMs are rubbish. And AFAIK yours do not even have a heatsink that makes the matters worse.

After 20 minutes of Prime your VRMs are probably going way over 100 degrees resulting in instability and crash.

^What he said, that VRM is not going to run an 1700 at stock without a fan sat directly on it, nevermind any of those overclock settings on an 1800x. RAM is also very dependent on the board quality for AM4 and 3200 CL14 is pushing that thing very hard.

Considering you could have gotten a very stout ASRock B350M board with a decent doubled VRM that has a heatsink for the same price I'm not sure what the point of buying that board was, especially if you're going to use voltage heavy presets.

You can blow up the VRM on that board with a quad core:

Thanks for the replies.
Had I realized what I was getting into I probably would have looked for VRM cooling pre-fitted onboard. My previous mobo was asus & that was good with my 8350, so whilst I like ASR I decided to stay with them.
I had assumed that Asus would make something fit for purpose....

So I need to get some heatsinks for the MOSFETS in any case.
I attach a hi-lited pic of where I think I need to put some coolers - please let me know if I got it wrong or need some elsewhere.

I have that same board and ran my 1700 at 3.75 without issue. I'm now working on modifying the bios to enable p state oc so it only runs that when fully loaded (my pc is always on, didn't make sense to have it sit there at 3.75 when idle most of the time), but it was perfectly stable with the stock wraith spire. I do have very good airflow in the case.

He may have needed a matx board, that's why I bought it as it's one of the better ones out there and mine is in an htpc. I plan to upgrade to a 470 board when they come out and throw a 2400g in this board/case, depends on performance of zen+ when it comes out. Before making a big investment in a new board or in cooling the vrms, I'd check the temp of them.

B series boards are not a good choice for overclocking the 8 core chips on, most of them the VRM is far too weak and will have issues. You can put some heatsinks on them and make sure a fan is blowing on them, that will likely cure your issue.

zubr375 said:

Thanks for the replies.
Had I realized what I was getting into I probably would have looked for VRM cooling pre-fitted onboard. My previous mobo was asus & that was good with my 8350, so whilst I like ASR I decided to stay with them.
I had assumed that Asus would make something fit for purpose....

So I need to get some heatsinks for the MOSFETS in any case.
I attach a hi-lited pic of where I think I need to put some coolers - please let me know if I got it wrong or need some elsewhere.

View attachment 55607

Click to expand...

Thats right. The mostfets you desperately need to cool is left vertical row from the CPU socket. The top horizontal ones are for memory. I'm not sure if cooling them is so necessary but if you are at it why not. But not only heatsinks you also want air circulation around them. These boards were pretty much intended to be used with stock Wraith or such downwards blowing coolers which move the air around the VRM's and not just blow it over them like tower coolers do. Even a little bit of air circulation works wonders on cooling VRMs. For example, I chose Noctua C14S cooler which is a massive tower cooler but tilted horizontally, it is floating on top of the VRMs and fan is blowing downwards directly at them through the cooler, keeps them nice and cool. Mine does have a decent VRM and a big heatsink but keeping them cool is good for longevity anyway. Since you already have a AIO that is obviously not a solution but if your case allows it you may try fit the radiator in a way that the fans also blow directly towards the VRM heatsinks. If not, just go ghetto and ziptie a fan on top of the VRM heatsinks! Hell, it may even be the best solution since you are not cooling them down with already warm air.

MaZa said:

Thats right. The mostfets you desperately need to cool is left vertical row from the CPU socket. The top horizontal ones are for memory.

Click to expand...

The top ones are for SOC/iGPU, not memory. The RAM has a single phase off to the right of the slots. DDR4 is very low power.

LigTasm said:

The top ones are for SOC/iGPU, not memory. The RAM has a single phase off to the right of the slots. DDR4 is very low power.

Click to expand...

Oh! My bad. I always thought they were for memory.

MaZa said:

Oh! My bad. I always thought they were for memory.

Click to expand...

Most boards they are up there, but AM4 is a little different.

Thanks for the feedback guys. Now I have an idea of what's next. The vid was helpful LigTasm, & thanks MaZa for the confirmation of MOSFETS location

Originally I hadn't intended any 'serious' OC, I'd figured I'd let the mobo do it's thing & give it hi-spec ram, a scheme which was fine on the 8350. Otherwise I would have opted for the higher model chipset.
I'm relieved that bobzdar is using the same mobo tho. Gives me some hope

I don't think it will hurt to add extra cooling regardless, so I have decided to try some of these
http://enzotechnology.com/performance/mos_c1.pdf
& I am ordering higher CFM fans so I can swap the existing ones which do not seem to have a rating. That will do for initial tests I think.

zubr375 said:

Thanks for the feedback guys. Now I have an idea of what's next. The vid was helpful LigTasm, & thanks MaZa for the confirmation of MOSFETS location

Originally I hadn't intended any 'serious' OC, I'd figured I'd let the mobo do it's thing & give it hi-spec ram, a scheme which was fine on the 8350. Otherwise I would have opted for the higher model chipset.
I'm relieved that bobzdar is using the same mobo tho. Gives me some hope

I don't think it will hurt to add extra cooling regardless, so I have decided to try some of these
http://enzotechnology.com/performance/mos_c1.pdf
& I am ordering higher CFM fans so I can swap the existing ones which do not seem to have a rating. That will do for initial tests I think.

Click to expand...

B350 is a fine platform, but the 8-core chips do push the VRM's hard. Remember, the X models have a higher base voltage and TDP than the non-X so they're more stressful. With some heatsinks and airflow you should be fine, although I would highly suspect your bluescreens to be related to the memory setup you're running. 3200 CAS 14 is not an easy setup to get stable on the current Ryzen IMC - run a few passes of google stressapp and you'll get your answer right away. Almost all of the RAM that will do those speeds are Samsung B-dies and for some reason there are batches of Ryzen that don't like the B-dies at all. I think buildzoid did a good video on this topic.

Thanks LigTasm.
I tried that google stress app via ubuntu terminal in w10. I guess being in win must give a different baseline to memtest in DOS. Couldn't find a linux boot cd with the app embedded - but maybe there isn't one as it needs to show an impact on the win environment?
Interesting to see that my ram fails after about 15 mins of an hour long test which memtest ran fine for for 3+ hours last week in DOS..
I have yet to get my cooling components so will have to see where that takes me as well.

I saw this comment in a ASUS guide for C6H -
Anyone have a good resource that gives more info on this, I can see the logic but that is not enuf info to get me moving ahead.

I think you have to activate developer status in Windows 10 to access the linux terminal. Kubuntu and Kali linux (not to mention Ubuntu itself) have LiveCD/DVD modes so you can run the distro without installing it.

Thanks Algrim. I will have to dig around some more to see if there is a pre-made stressapptest ISO. Unless I can figure out how to build one... I have no experience with Linux of any flavor before todays stress test.

I just ran the test in Win10 again with RAM set to 3000mhz (same timings) - all OK. See pics, the voltages were pretty good as were the temps.


HWInfo - columns = Current | Min | Max | Avg



What I've noticed about this RAM is that the timings seem almost fixed. It runs best & with least errors @14-14-14-31 it also doesn't seem to want a lot of extra voltage. The CPU wants extra voltage to keep it all stable tho - I presume that is how the on-chip memory controller comes into the equation.

I guess when I have more cooling & a better grasp of DRAM ratio / TWCL I can try 3200 in Win10 again.

zubr375 said:

Thanks Algrim. I will have to dig around some more to see if there is a pre-made stressapptest ISO. Unless I can figure out how to build one... I have no experience with Linux of any flavor before todays stress test.

I just ran the test in Win10 again with RAM set to 3000mhz (same timings) - all OK. See pics, the voltages were pretty good as were the temps.



What I've noticed about this RAM is that the timings seem almost fixed. It runs best & with least errors @14-14-14-31 it also doesn't seem to want a lot of extra voltage. The CPU wants extra voltage to keep it all stable tho - I presume that is how the on-chip memory controller comes into the equation.

I guess when I have more cooling & a better grasp of DRAM ratio / TWCL I can try 3200 in Win10 again.

Click to expand...

Ryzen has a lot of "memory holes" which vary based on your IMC. 2933-3000mhz is very common to run well (you may even get it to CAS12 without much issue). I've had about a dozen Ryzen chips now and my current 1800X absolutely will NOT do 3200mhz but it will run the same RAM at the same timings at 3000 or 3400mhz with zero issues.

I find that the GSAT takes about an hour to find errors that P95 or memtest would take days to do, which makes sense because Google built it specifically to test datacenter machines, since they have zillions of them they have to do it fast.

Thanks for the ref of the stressapptest LigTasm.

That stress test was very interesting for its' outcomes as well as the dramatic reduction in time req'd.

It bears mentioning that I'd had no problem just sticking the RAM into the mobo & booting into Win10 3200 on DOCP. I was impressed by that.
Problems came along when DOCP wasn't really playing ball with the exact spec of the ram when the PC was under load (It had changed some values to JEDEC 3200 numbers). I then manually altered it & went thru memtest 3200 with no errors.
But Prime95 showed a problem & here we are...

This is my RAM
https://www.teamgroupinc.com/en/product/dark-pro-ddr4
You can see it gives a 3333mhz rating - once I have my coolers I want to try this speed, hopefully 3200 will run in Win properly with the extra cooling but if I can get to 3333 I'd like to see the impact in terms of voltage & temps. I suspect I might need 1.4+ & I really want to stay below 1.4 if possible.
What impact does 3400 have on Ryzen volts etc in your experience?

Depends on your chip, pushing it past 3000-3200 isn't really going to net you anything besides some extra aggravation. I can run my current kit at DOCP but it has errors sometimes as well, but manually inputting the speeds and timings and letting the board handle the subtimings works fine.

I'm guessing by aggravation you mean stability & temps vs time LigTasm. I have found the way this mobo reacts to JEDEC & DOCP vs actual timings interesting. This suggests there is a lot more going on inside a Ryzen with its MCs than is immediately obvious.

My cooling stuff has finally arrived. Reading the package I discovered I need to 'clean' the mosfets. The package does not however give me any idea what kind of cleaning is req'd.

I had wondered about this as having a copper lump flying about on a mobo is not going to be component friendly.
Anyone have any recommendations on cleaning pre tape application?

zubr375 said:

I can pass Memtest86 in DOS but I cannot get past 20 min with Prime95 torture test without a restart/BSOD.

Click to expand...

Jokes asside
I am having the same issues with my ryzen 1700x even at stock speed it not stable in prime95 v29.b7 running with FMA. but with 29.b5 it runs perfect with avx
any kind of oc the 29.b5 with AVX becomes unstable.

This has ber the most disappointing experience I have either have (intel or AMD) in regards to OC'ing.

I'm not sure how much of this is on Ryzen & how much is on the software that is pre/not yet optimized for Ryzen.
Passing tests under DOS should be enough to show basic stability, & Win10 has been a major PITA since I upgraded to 17xx, takes me back to Win95!!!

zubr375 said:

I'm not sure how much of this is on Ryzen & how much is on the software that is pre/not yet optimized for Ryzen.
Passing tests under DOS should be enough to show basic stability, & Win10 has been a major PITA since I upgraded to 17xx, takes me back to Win95!!!

Click to expand...

Again where does this DOS thing come from.

Also i would disagree that running just a memtest is alone a sign of basic stability. it does not push cpu at all or gpu.

SvenBent said:

Again where does this DOS thing come from.

Also i would disagree that running just a memtest is alone a sign of basic stability. it does not push cpu at all or gpu.

Click to expand...

So are you just trolling?? i.e.

Guess I wasted my time commenting on the engineering differences being made obvious by Ryzen & the different environments...

zubr375 said:

So are you just trolling?? i.e.

Arguing semantics on what is DOS (non-multitasking environment) vs what is OS (multitasking environment), or do you really not understand what DOS is/was?
Never said memtest definitively meant everything was stable. Why imply I did?​

Guess I wasted my time commenting on the engineering differences being made obvious by Ryzen & the different environments...

Click to expand...

Clearly you did not read my entire post after getting offendended about the "DOS" part.
I did in fact comment to your believing that a simple metest86 test in "DOS" was enough to certify stability.

zubr375 said:

So are you just trolling?? i.e.

Arguing semantics on what is DOS (non-multitasking environment) vs what is OS (multitasking environment), or do you really not understand what DOS is/was?
Never said memtest definitively meant everything was stable. Why imply I did?​

Guess I wasted my time commenting on the engineering differences being made obvious by Ryzen & the different environments...

Click to expand...

If you want to see some good investigation into Ryzen IMC check out the stilts work on OCN. He has some spreadsheets for different speeds and "safe" vs "aggressive" timings that go very in depth into the subtimings. These chips are a little more finicky than the Intel stuff because the memory speed and latency has a big effect on the infinity fabric that ties the cores together.

I've been having issues with a "new" build. My 1700 keeps blue screening in Windows. I started doing memory tests, and my FlareX memory is throwing errors at D.O.C.P. I bumped the voltage to 1.37V and it's gone 10 minutes without an error where it wouldn't go 15 seconds without one before. Definitely a hassle. Running stock speed on a Prime X370-Pro board, so I can't imagine that the VRM's are at fault.

kirbyrj said:

I've been having issues with a "new" build. My 1700 keeps blue screening in Windows. I started doing memory tests, and my FlareX memory is throwing errors at D.O.C.P. I bumped the voltage to 1.37V and it's gone 10 minutes without an error where it wouldn't go 15 seconds without one before. Definitely a hassle. Running stock speed on a Prime X370-Pro board, so I can't imagine that the VRM's are at fault.

Click to expand...

You're probably going to need to set your timings manually. The DOCP/XMP settings even on the "AMD Ryzen" specific RAM don't seem to work all that well.

get a fan and just lay it on the their man, help it out some. blew a vrm...

yup. if youre OCing and using water you HAVE to cool the vrm.

pendragon1 said:

yup. if youre OCing and using water you HAVE to cool the vrm.

Click to expand...

I had a fan laying in there always done that...but hell idk...probably board...still doubting something wrong with the chip...find out in a day or two...