Ryzen 3950X Build Help

[A Computer Guy]

run just one stick only in B2? i did previously swap both sticks at the same time, it was the same results

same memory location, and same CPU reported, it was 7641MB again

Ok so just to clarify where we are...of the two sticks from the gskill kit, each one was tested individually in slot B2 and reported the same, or similar error, regarding the CPU # and memory addresses?

 

[erek]

Ok so just to clarify where we are...of the two sticks from the gskill kit, each one was tested individually in slot B2 and reported the same, or similar error, regarding the CPU # and memory addresses?

previously i only tested one of the sticks in slot B2 and was unable to reproduce the error, but have to restart that testing over again cause i lost track of which stick that was


right now i'm just running both sticks still, i'm just concerned that basically it could be the mobo, a ram stick, or the cpu memory controller


sorry i'm just getting extremely depressed about this... spent an entire week of vacation to do this build, have to return to work, and just can't imagine now facing having to RMA one component at a time to troubleshoot this and possibly months of individual processes between the manufacturers

I went through this ordeal on a Ryzen 2700x build when the x470 platfrom still had some teething issues. Long story short, a bad board killed two sets of memory sticks before we figured out what was going on.

See: https://hardforum.com/threads/3950x-overclocking-guide-needed.1989231/page-3#post-1044409708

 

[Deleted member 185158]

previously i only tested one of the sticks in slot B2 and was unable to reproduce the error, but have to restart that testing over again cause i lost track of which stick that was


right now i'm just running both sticks still, i'm just concerned that basically it could be the mobo, a ram stick, or the cpu memory controller


sorry i'm just getting extremely depressed about this... spent an entire week of vacation to do this build, have to return to work, and just can't imagine now facing having to RMA one component at a time to troubleshoot this and possibly months of individual processes between the manufacturers

Ah don't be depressed. A single error from memtest after several passes doesn't really mean much to me while it happens the same over and over. I'd like to think if there was a real issue, it would have occured during the first pass.

Again. Try OCCT in windows. It's pretty heavy. WIll give you a nice visual display of what's going on, log report in the way of a graph and so forth. https://www.ocbase.com/

I test my memory configurations with this. I don't use Memtest86 quite too much any more. I remember when it was built into the bios.... I always got errors from memtest lol. Stupid overclocker I am though....

 

[A Computer Guy]

previously i only tested one of the sticks in slot B2 and was unable to reproduce the error, but have to restart that testing over again cause i lost track of which stick that was


right now i'm just running both sticks still, i'm just concerned that basically it could be the mobo, a ram stick, or the cpu memory controller


sorry i'm just getting extremely depressed about this... spent an entire week of vacation to do this build, have to return to work, and just can't imagine now facing having to RMA one component at a time to troubleshoot this and possibly months of individual processes between the manufacturers



See: https://hardforum.com/threads/3950x-overclocking-guide-needed.1989231/page-3#post-1044409708

Yea a feel your pain. I'm just trying to help you quickly determine what is working and what is not. If one stick works but not the other then it's a clear path to investigate how far the working stick will go and you might be able to make a case for RMA on the ram kit and validate if the CPU and mobo seem to be ok at the same time.

 

[erek]

Ah don't be depressed. A single error from memtest after several passes doesn't really mean much to me while it happens the same over and over. I'd like to think if there was a real issue, it would have occured during the first pass.

Again. Try OCCT in windows. It's pretty heavy. WIll give you a nice visual display of what's going on, log report in the way of a graph and so forth. https://www.ocbase.com/

I test my memory configurations with this. I don't use Memtest86 quite too much any more. I remember when it was built into the bios.... I always got errors from memtest lol. Stupid overclocker I am though....

but the error is reproducible, it's the same memory address starting always: 7461MB and usually CPU 24, one time i did see CPU 23

 

[A Computer Guy]

Yea a feel your pain. I'm just trying to help you quickly determine what is working and what is not. If one stick works but not the other then it's a clear path to investigate how far the working stick will go and you might be able to make a case for RMA on the ram kit and validate if the CPU and mobo seem to be ok at the same time.

ShrimpBrime's suggestion I think is a good idea too. Every time I have a go at this I'm always finding something new and people with new info and approaches.

 

[Deleted member 185158]

but the error is reproducible, it's the same memory address starting always: 7461MB and usually CPU 24, one time i did see CPU 23

Yea I know. Gotta read a little bit. Just because you get an error, doesn't mean it's actually the memory itself. I quoted what you need to read to understand that.

Gotta try more than one stress test. You could get an error from memtest and never from anything else.

There's more to testing memory than just memtest86. OCCT has bee around a long time. You can choose AVX AVX2 and no AVX instructions for testing.
You can set a thermal limit for the program to shut down if gets to hot.

I tested my 3200mhz memroy at CL 14-14-14 on a Zen+ 2700x and it passes. I'd be willing to bet that Memtest would find an error for me though

 

[A Computer Guy]

but the error is reproducible, it's the same memory address starting always: 7461MB and usually CPU 24, one time i did see CPU 23

What is not clear to me is if it is happening with both sticks when tested individually. I lost track. Does one stick have the issue or both when tested individually? That is the question.

 

[Deleted member 185158]

What is not clear to me is if it is happening with both sticks when tested individually. I lost track. Does one stick have the issue or both when tested individually? That is the question.

He tested both sticks, but swapped the slots. The memory error happens in the same exact address, so it's not defined to one stick if I gathered it all correctly.

 

[A Computer Guy]

He tested both sticks, but swapped the slots. The memory error happens in the same exact address, so it's not defined to one stick if I gathered it all correctly.

Yea that is what I am confused about. I'm just trying to determine if in the same slot B2 - both ram sticks behave the same or not?

He reported "issues with chrome and black screens temporarily" so it makes me think the problem is real and maybe can't be ignored per memtest86 (realizing that software is not perfect either and only one stone in a series of stepping stones for testing stability)

 

[Deleted member 185158]

Yea that is what I am confused about. I'm just trying to determine if in the same slot B2 - both ram sticks behave the same or not?

He reported "issues with chrome and black screens temporarily" so it makes me think the problem is real and maybe can't be ignored per memtest86 (realizing that software is not perfect either and only one stone in a series of stepping stones for testing stability)

Generally a black screen is far from a memory error. "issues" with chrome. Who doesn't have those lol. (undefined issue?>)

You might have been onto something about the PSU right from the get go.

Luckily, OCCT has a PSU test . It'll put such a strain on his rig.... oh man I don't even like using it myself.

 

[erek]

I’m doing the memtest against a single stick in slot A2, could take a few hours to test both. Just going with straight XMP . Default occt test ran successfully for over 15 minutes. The issues with chrome and black screen was while running prime95 blend test.

To clarify I previously tried both sticks swapped positions from a2 to b2 but at the same time and the error is at the same address

 

[Deleted member 185158]

I’m doing the memtest against a single stick in slot A2, could take a few hours to test both. Just going with straight XMP . Default occt test ran successfully for over 15 minutes. The issues with chrome and black screen was while running prime95 blend test.

To clarify I previously tried both sticks swapped positions from a2 to b2 but at the same time and the error is at the same address

Typically a bad memory error results in a blue screen.

A black screen.... Give this more definition. Was it a shut down? Just hung for a moment? How did that balck screen occur from start to end. Did the black screen go away and the system was still usable? Hung on black screen and you had to reset the PC?

 

[erek]

Typically a bad memory error results in a blue screen.

A black screen.... Give this more definition. Was it a shut down? Just hung for a moment? How did that balck screen occur from start to end. Did the black screen go away and the system was still usable? Hung on black screen and you had to reset the PC?

the chrome window went black for a moment, I’m thinking related to running out of memory because prime95 was using as much as possible during the blend testing

 

[Deleted member 185158]

the chrome window went black for a moment, I’m thinking related to running out of memory because prime95 was using as much as possible during the blend testing

Oh just a flicker. Yea, that's the system on a heavy load. Pretty normal.

Yea, I think you'll be ok.

Next step, testing timings at 16-16-16!

 

[erek]

Oh just a flicker. Yea, that's the system on a heavy load. Pretty normal.

Yea, I think you'll be ok.

Next step, testing timings at 16-16-16!

16-16-16? That’s tighter than XMP 16-19-19

i tried and failed jedec 6 at 16-20-20 and 1.25v with both sticks, remember?

 

[A Computer Guy]

I’m doing the memtest against a single stick in slot A2, could take a few hours to test both. Just going with straight XMP . Default occt test ran successfully for over 15 minutes. The issues with chrome and black screen was while running prime95 blend test.

To clarify I previously tried both sticks swapped positions from a2 to b2 but at the same time and the error is at the same address

LOL I did the same thing when I built my Ryzen last year. I figured if I could run prime95, browse the web, and watch netflix video for an hour it was likely ok. I didn't have the black screen problem though but when I did ACE Combat 7 it did me in and I had to go back to the table and readjust my ram timings. It just goes to show that testing different workloads is important to determining how stable the ram is which is why OC'ing your ram can be risky business.

The reason I stress B2 - sorry I meant to say A2 (it's been a long day) - is because it's the first slot in the primary channel and it likely to behave the best when testing as far as I am aware. Some boards don't like it when you don't populate the first channel before the second. So by populating the A2 the rationale is it will give you the most consistent results when testing different ram sticks individually. In any case determining if you have the same, or different behavior, with a single stick at a time will make it easier to decide what to do next once you know for sure what the behavior is in that slot.

We don't know for sure if memtest86 is saying if your ram is bad, it is possible but not certain, it may be incompatible, but for starters lets take it with a grain of salt that it's it somewhere to start. You do have another ram kit so technically you have a basis for trying to eliminate some possibilities to get closer to an answer. The next test may be to take your other set of ram and test one stick in the same slot as you did with the gskill. The outcome from that may point you in the direction you simply need a different ram kit... maybe it's bad maybe it's not but regardless you just need something that works.

If the other ram module (testing in the same slot) gives you the same results then maybe it's the CPU, or the board, or even the testing software. Again maybe not bad, but maybe just incompatible in some way, but then you have some options how to choose to proceed. Most people don't have enough hardware on hand to work though all the possibilities for certain.

My concern regarding the PSU is I had received some input many years ago that some dust can have conductive qualities. You certainly had a PSU packed with a lot and perhaps you're getting some bad power delivery because of it but I certainly don't know for sure.

It's your choice how to proceed. Just do what you can to solve your issue. I can't say I'm any expert at ram diagnosis.

LOL I did the same thing when I built my Ryzen last year. I figured if I could run prime95, browse the web, and watch netflix video for an hour it was likely ok. I didn't have the black screen problem though but when I did ACE Combat 7 it did me in and I had to go back to the table and readjust my ram timings. It just goes to show that testing different workloads is important to determining how stable the ram is which is why OC'ing your ram can be risky business.

The reason I stress B2 - sorry I meant to say A2 (it's been a long day) - is because it's the first slot in the primary channel and it likely to behave the best when testing as far as I am aware. Some boards don't like it when you don't populate the first channel before the second. So by populating the A2 the rationale is it will give you the most consistent results when testing different ram sticks individually. In any case determining if you have the same, or different behavior, with a single stick at a time will make it easier to decide what to do next once you know for sure what the behavior is in that slot.

We don't know for sure if memtest86 is saying if your ram is bad, it is possible but not certain, it may be incompatible, but for starters lets take it with a grain of salt that it's it somewhere to start. You do have another ram kit so technically you have a basis for trying to eliminate some possibilities to get closer to an answer. The next test may be to take your other set of ram and test one stick in the same slot as you did with the gskill. The outcome from that may point you in the direction you simply need a different ram kit... maybe it's bad maybe it's not but regardless you just need something that works.

If the other ram module (testing in the same slot) gives you the same results then maybe it's the CPU, or the board, or even the testing software. Again maybe not bad, but maybe just incompatible in some way, but then you have some options how to choose to proceed. Most people don't have enough hardware on hand to work though all the possibilities for certain.

My concern regarding the PSU is I had received some input many years ago that some dust can have conductive qualities. You certainly had a PSU packed with a lot and perhaps you're getting some bad power delivery because of it but I certainly don't know for sure.

It's your choice how to proceed. Just do what you can to solve your issue. I can't say I'm any expert at ram diagnosis.

I've got to pack it in for the night, later and good luck for now.

 

[Deleted member 185158]

16-16-16? That’s tighter than XMP 16-19-19

i tried and failed jedec 6 at 16-20-20 and 1.25v with both sticks, remember?

Was kind of kidding there.... BUT--->

1.) P95 throw errors? no?
2.) OCCT throw errors? no?
3.) Blue Screen any? no?
4.) Memtest 1 error 4th pass? so?

I think the top three would define you to have a stable running system.

 

[oxrufiioxo]

Man what bummer having to go through all that....... I would probably try a couple different kits of ram on the qvl. Order from amazon that way you can return them easily if you have the same issues. My second step would be trying a different motherboard if you can return yours for a replacement still. I would do the CPU last considering it'll probably be ages before newegg gets more stock to replace yours with... They would likely just give you store credit anyway.

 

[erek]

One of the sticks individually gave an error, but i haven't been able to reproduce it yet after many retries:

 

[A Computer Guy]

One of the sticks individually gave an error, but i haven't been able to reproduce it yet after many retries:

View attachment 138064

So do I understand correctly that in your tests on A2 one ram stick throws the error but the other does not?

 

[erek]

So do I understand correctly that in your tests on A2 one ram stick throws the error but the other does not?

Yes, but trying to reproduce

 

[A Computer Guy]

Ok interesting. So at this point you have some options depending on your level of concern and dedication to troubleshooting.

You can choose to ignore the error and use the ram.
- maybe this comes back to bite you later *if* the problem was real

You can replace the ram kit.
- a pretty easy solution

You could continue some diagnostics.
- takes more of your time but we might learn if something else is wrong that might come in handy if you want to try to make a case for RMA or warranty claim.

A thought that occurred to me is maybe you have a problem with a core. It would really suck if you paid for 16 cores but only got 15.5 but manufacturing defects do happen. Maybe it's broken or maybe it's "weak". If it were possible to rule out or confirm a problematic core it might help. Since we have a consistent idea that one of the ram sticks is certainly problematic we can change the conditions of the test to try to learn if maybe core #12 (cpu #24) exhibits an issue by first eliminating from the test then later isolating it for a test if possible.

I had an idea of turning off core #12 (since the the problem was fairly consistent stopping on CPU23/24) to see if the misbehaving ram stick passes or continues to have an error. By turning off core #12 the core #13 should take it's place in the test and we have no indication that core #13 is reporting an issue in memtest86.
- So first starting with the good ram stick turn off core #12 and retest to establish a baseline. (we expect this to pass)
- Next put in the suspect ram stick and re-run the test with core #12 off.

Conversely from the test above by repeating the procedure with as many cores turned off as possible except core #12 we might learn something about the consistency of core #12 with respect to the good and suspect ram module.

 

[Deleted member 185158]

Yea, time to move on.
Glad you got the memory issue sorted out.

Keep up the good work Erek. Memory can be frustrating sometimes, but I think you have a good grasp on the situation.

Sometimes the cpu IMC can make memtest throw errors and is not a direct indication of a bad memory stick.

Put your system back together and enjoy some gaming for a change.
Some of the best testing is done just by sheer use. Some benchmarking and gaming is pretty stressful and if you encounter a memory issue, youll find it just through use.

I like your Cinebench scores by the way. Keep up the good work!!!
Jon

 

[erek]

Ok interesting. So at this point you have some options depending on your level of concern and dedication to troubleshooting.

You can choose to ignore the error and use the ram.
- maybe this comes back to bite you later *if* the problem was real

You can replace the ram kit.
- a pretty easy solution

You could continue some diagnostics.
- takes more of your time but we might learn if something else is wrong that might come in handy if you want to try to make a case for RMA or warranty claim.

A thought that occurred to me is maybe you have a problem with a core. It would really suck if you paid for 16 cores but only got 15.5 but manufacturing defects do happen. Maybe it's broken or maybe it's "weak". If it were possible to rule out or confirm a problematic core it might help. Since we have a consistent idea that one of the ram sticks is certainly problematic we can change the conditions of the test to try to learn if maybe core #12 (cpu #24) exhibits an issue by first eliminating from the test then later isolating it for a test if possible.

I had an idea of turning off core #12 (since the the problem was fairly consistent stopping on CPU23/24) to see if the misbehaving ram stick passes or continues to have an error. By turning off core #12 the core #13 should take it's place in the test and we have no indication that core #13 is reporting an issue in memtest86.
- So first starting with the good ram stick turn off core #12 and retest to establish a baseline. (we expect this to pass)
- Next put in the suspect ram stick and re-run the test with core #12 off.

Conversely from the test above by repeating the procedure with as many cores turned off as possible except core #12 we might learn something about the consistency of core #12 with respect to the good and suspect ram module.

Ok interesting. So at this point you have some options depending on your level of concern and dedication to troubleshooting.

You can choose to ignore the error and use the ram.
- maybe this comes back to bite you later *if* the problem was real

You can replace the ram kit.
- a pretty easy solution

You could continue some diagnostics.
- takes more of your time but we might learn if something else is wrong that might come in handy if you want to try to make a case for RMA or warranty claim.

A thought that occurred to me is maybe you have a problem with a core. It would really suck if you paid for 16 cores but only got 15.5 but manufacturing defects do happen. Maybe it's broken or maybe it's "weak". If it were possible to rule out or confirm a problematic core it might help. Since we have a consistent idea that one of the ram sticks is certainly problematic we can change the conditions of the test to try to learn if maybe core #12 (cpu #24) exhibits an issue by first eliminating from the test then later isolating it for a test if possible.

I had an idea of turning off core #12 (since the the problem was fairly consistent stopping on CPU23/24) to see if the misbehaving ram stick passes or continues to have an error. By turning off core #12 the core #13 should take it's place in the test and we have no indication that core #13 is reporting an issue in memtest86.
- So first starting with the good ram stick turn off core #12 and retest to establish a baseline. (we expect this to pass)
- Next put in the suspect ram stick and re-run the test with core #12 off.

Conversely from the test above by repeating the procedure with as many cores turned off as possible except core #12 we might learn something about the consistency of core #12 with respect to the good and suspect ram module.

Using both sticks of ram slots B2, and A2 @ XMP Profile 3600 MHz

had to use Ryzen Master to disable core 12, but due to symmetry had to disable more, so C4, C8, C12, and C16 are also disabled

saw the error at the same memory address as before 7641MB, but now on CPU 3: