Noob here to the posting crowd in Level 1, but have been following Wendell’s channel for a while :D.
So, I have built a new Ryzen 3950X computer, with the following specs and “quirks”:
Ryzen 9, 3950X
Gigabyte Aorus X570 Master (came with F4 UEFI but upgraded to latest F11, running failsafe defaults, except I deactivated onboard audio, since I have a separate Creative card)
Corsair HX1000i PSU
your random Samsung 840 EVO SSD I salvaged from my previous computer (will get upgraded as soon as finances are back on track :D)
BeQuiet Dark Base Pro 900 rev 2 case (although probably not relevant to the topic)
Right off the bat: I have no intention of actually doing any OC, but I do have the basic requirement that my system is rock solid stable at rated speeds (CPU, MEM, …). Therefore, here starts my problem:
When I run the RAM at default JEDEC settings (XMP disabled), the system looks just like it should: rock solid
I tested AIDA 64 system stress test overnight - no problem,
I ran Prime 95 stress test overnight, no problem,
I ran several passes of Memtest64 (from techpowerup) - no problem.
But:
When I enable XMP, system becomes really unstable:
Prime 95 crashes (BSOD) within minutes; I also notice that before crashing altogether various cores start to lose load (no longer at 100%, although they maintain constant frequency), until it crashes.
Aida 64 aborts the test almost instantly with a message of “HW error detected”
Memtest64 detects on every pass several errors at random memory locations on the Bitpattern tests.
if just simply using the system for light IE browsing, it will crash within 10 minutes or so.
the BSODS have various bugchecks, with no apparent pattern.
So, my kind request for help is:
I did not try to change the RAm slots on the MB; do you think it’s worth a try? Or to try to run with just one RAM stick? see what happens?
is there maybe some known quirks of the MB BIOS that I need to address to get the XMP profile running smoothly?
is there some simple things I can fine tune in the UEFI that could solve the XMP issues (remember, my intention is not to OC, and mess as little as possible with the MB settings).
or, in your opinion, I should just try to RMA the RAM, because the system behavior is too wrong to be explained away by some sort of: “But running XMP is OC-ing” explanation. - the memory is rated at 3200Mhz; the CPU is rated to work with that mem speed, and the particular RAM kit is not on the Gigabyte MB QVL list, but it’s 32GB variant is.
Many thanks in advance for any tip and advice,
Andrei
Yes, you are right, the kit is on the QVL (I missed it the first time :D);
Yes, the 1.35V is correctly set by the XMP profile (it’s in the SPD)
So, I’ll try memtest86 by running booted off a stick, and post back.
@aveamurechi I had some issues with the same RAM model and Ryzen CPUs; my other hardware was different. Initially, I used the Corsair Vengeance RAM with a 2600 CPU; memtest86 reported around a dozen errors in 4 passes. The errors persisted when I switched to a 3800X, though at a lower rate of 3-4 errors. I did a number of experiments with different memory voltages and frequencies, but still had errors at the rated frequency.
After I switched to another make of RAM, which worked well, I reinstalled the Corsair Vengeance DIMMs, intending to investigate the problem. To my surprise, they worked perfectly. My tentative explanation is that there was a RAM training error early on which set up not-quite-right timings. Changing out the memory, and perhaps clearing CMOS sometime along the way, allowed the board to make a fresh start on training the second time around - and this time the results were better. Or… who knows?
I tried three sets of RAM, and only one set “just worked” - TridentZ Neo 3600 Mhz 16-19-19-39 2x8GB. That gives me a favorable impression of the whole TridentZ Neo line (in spite of the RGB), but isn’t enough experience to prove anything.
Based on my experience, I suspect that Ryzen 3000 remains a little finicky about RAM, though less so than previous generations. I invite you to consider:
Do a thorough power down, including unplugging the system. Clear the CMOS (read the instructions - the button may have to be pressed for a certain amount of time while the system is unplugged). Restart and hope the RAM training goes well.
Try lowering the memory frequency and see at what speed the RAM works flawlessly (likely 3000 Mhz or higher). Maybe that is “good enough”?
Replace the memory with another make & model that has good reviews from Ryzen users (maybe TridentZ Neo).
Good luck; I hope this or some other advice works out well for you.
@pcwolf: I checked the manual and I see no recommendation. All configurations are equally supported. But for what it’s worth, the sticks were initially in the slots you mention…
@Caped_Kibitzer: your point about clearing CMOS is well taken, so I slightly changed my test methodology to incorporate that. So, test steps are:
Turn OFF computer (including hard switch in PSU back)
Several pushes of power button to deplete any charge in any capacitors
change RAM slot
Clear CMOS by button on the back or the MB backplate (no specific instructions in the MB manual, but I keep it pressed for 10s, then several shorter presses for good measure).
5). Boot computer
Check via CPU-Z that the memory profile is set to default (means CMOS clear was successful)
Do a round memtest64 (from TechPowerup)
reboot to memtest86 from USB
run a full pass of memtest86
go to UEFI and change to XMP
reboot and check that XMP is enabled via CPU-Z
do a round of memtest 64
reboot to memtest 86 booted from USB stick
14 run a round of memtest86
go to step 1.
I ran this procedure for both RAM sticks individually, and using them in every RAM slot.
Result is:
one RAM stick is good, works in all conditions (XMP/non-XMP), in any RAM slot
the other RAM stick is bad; throws errors both in memtest 64 and memtest86, in any RAM slot, but only when XMP profile is activated; it works just fine with default settings. The errors are obvious: memtest64 throws 15-20 errors per run; memtest 86 throws errors by the thousands, on random mem locations, on random cores, on random tests.
My conclusion is: one of the RAM sticks is bad and needs to be RMA’d. RAM slot, MB, CPU, mem controller are eliminated as culprits since they work fine with the other stick (identical - they came in a kit).
Congratulations on diagnosing the problem. Since the RAM sticks came together in a kit, you will probably want to RMA the kit. Then you get another kit whose sticks have been tested together. (Though something went wrong with that process for the first kit.)
So, I went and RMA’d the kit of RAM, and the new one came in. Problem is: makes the same behavior. I did not test it as thoroughly as the first one, but it definitely looks like it behaves the same: one stick works well by itself also with XMP enabled, the other one gives even more errors in memtest86 than the bad stick in the first kit.
I am not sure what to do now. I’m pretty sure the seller will give me a funny look if I RMA this kit as well. Besides: what are the odds that I get 2 consecutive kits with the exact same problem???
Actually, I am dealing with a very large (and customer friendly) retailer in Romania (eMag). They gave me the choice:
either RMA (in which case I have to give them the kit, they send it over for diagnostics… takes 15 days)
or get a refund and order the same model kit again; (in which case I get to keep the faulty kit and use it until the new one arrives)
Of course I chose the second option… So you say that you guys are confident there is not something else that is wrong with my system? And that I should keep getting new RAM kits until one works?
@aveamurechi If you add the “helpdesk” tag to your post, you may attract some additional helpful forum members. I am going through a similar process to yours, and find that with my CPU and motherboard, most RAM kits have to be underclocked by 66 - 133 MHz to work reliably. I realize that you would prefer not to do that, but it might be worth a try.
I suggest you disable Fast Boot in the UEFI. With Fast Boot disabled, the system will perform memory training on every cold boot. This may help resolve your issues, and will make the system more responsive to any other changes in memory settings. Cold boot may take 30 sec. or more. If you get to a good setup, you can then enable Fast Boot.
I am pretty confident that your system lies within the “normal” range of variation and your results are not way out of line. A different motherboard (of the same model) or CPU (of the same type) might give better or worse results. In particular, a different model of RAM might do better (or worse). My closest approach to a “simple answer” is to try Trident Z Neo RAM, which is well regarded for Ryzen 3000 - but it is not guaranteed to be problem-free.
Welcome to the cutting edge, and try not to bleed on the components!
First of all, thanks for trying to cheer me up. Much appreciated. I had forgotten what it’s like to be on the bleeding edge. My previous system was Haswell based, and served me very well after approx. 6 months of messing with heavy, deep, investigations on my side and several not so friendly RMA processes of various components… (I eventually had to send the CPU directly to Intel for RMA, because the retailer wouldn’t be bothered to run a thorough check on it, and see it failing…)
I guess I’ll go ahead with another exchange of the RAM kit… Wish me luck!!
Hi @FurryJackman. Since I’m new to any kind of manual tweaking in general, I just want to make sure I get what you are trying to say. So I took a snapshot of my current Ryzen Master Home screen and highlighted what I think you mean needs to be modified.
I recently got an Aorus x570 Ultra board along with a Ryzen 9 3900X and 32GB of 3200mhz HyperX memory. To begin with I had issues with the ram in both the non XMP mode and XMP modes where it was causing corrupt process contexts, caused by memory errors.
Luckily for me, there were two XMP profiles with the RAM, one that uncerclocked it to 3000MHZ which for me completly solved the issue.
I think it comes down to the infinity fabric and memory timings. Some people I found that had issues fixed it by overclocking the infinity fabric, others by tweaking the memory timings, voltages, clocks etc.
I don’t have the exact same setup as you, but it sounds like you might have some of the same problems.
Thanks so much for all the help and hints. But I feel we are missing where we started from: I have two kits in a row where one of the sticks works by itself great (and out of the box) with XMP enabled, and the other not. Obviously when I put them together, they also do not work.
So, I’m all for digging deeper into the whole manually tweaking the RAM (opportunity for me to learn), but I can’t help but feel that we are ignoring this piece of info. In other words: is there any reason to believe that one particular voltage/frequency/others setting will work just fine with one stick but be completely wrong with the other from the same kit? other than the reason that one stick is just DOA? Twice?
@FurryJackman: if you really think that this SoC Voltage might be the culprit, then I will try this, what you suggest.
But I still need some help to figure out exactly which setting to change in the UEFI. It is confusing since I have a strong feeling there are several SoC voltages to mess with, and I am not clear which one you mean.
Eg: my HW info reports at least 2 SoC voltages. Which one of them do you mean? Sorry for noobish questions, but I have never in my life had an interest in messing with these things, and this computer cost a lot of money ; don’t want to break something…
