Tweet
Can someone please explain to me why the XMP voltage (1.35v) is not set correctly? It's set in the BIOS, but for some reason, MemTest86 does not seem to register 1.35v and instead is reporting 1.2v. This is about to be my third RMA with this G.Skill RAM, and I just wanna know if maybe this is incorrect RAM BIOS settings? Then again, even at stock settings I got 200+ errors with this stick, while a single stick of Kingston RAM passed all tests, 4 times no less, in each of my 4 DIMMs, so can't be bad BIOS RAM settings, right?
-
-
The 1.35V figure is a recommendation, made by the memory vendor. The BIOS doesn't need to follow this. A user can manually set another voltage level, or the BIOS might decide to itself if in auto mode.
There is also a small chance MemTest86 is reporting the an incorrect value. You cropped the screen shot and didn't mention the version number. This is more likely if you are using older software. We aren't aware of any issues like this in the current V11.6 release.
V11.6, at least, we get the voltage value from SMBIOS, which is a data structure populated by BIOS at boot time. Which means that it is possible that BIOS has a bug in filling out the data structures.
If you are getting 200 errors at stock (slow JEDEC) settings, then clearly something is pretty wrong. Getting 2 bad sticks are 2 RMAs would also be strange.
You might want to check if newer BIOS firmware is available for this motherboard. -
Below I uploaded the full test. BIOS is the latest version available on AsRock's website for Taichi. I manually set the voltage as well, still got the same 1.2v reported in MemTest86. It says it's 11.6, so guess that's the newest one. I'm currently re-testing everything one more time (both G.Skill RAM and the kingston stick), just to be 100% sure. My first kit was bought second hand, 2022. Then at the end of 2023, I got first BSOD, tested my memory, and saw errors. Sent my RAM for an RMA, got it on Jan 2024, it worked for a week or something and then BSOD again. I opted to just power through and not go through a month long RMA process again, plus G.Skill telling me they couldn't replicate my issues made me think that something else could be an issue. Then near the end of 2024 it was getting worse, more BSOD than before. And someone pointed out that the screenshot I got from G.Skill showed they tested about 80% of my RAM's capacity because they tested within Windows. Then I decided to do a very thorough test and document everything (also told G.Skill to use your test and report back with that this time, not their own weird MemTest version which was inside of Windows). This is what I did:Originally posted by David (PassMark) View Post
1) 4 tests via Kingston stick that has no errors, testing each DIMM (default, 4 passes)
2) G.Skill stick A in each DIMM, stock and XMP, only 1 pass
3) G.Skill stick B in each DIMM, stock and XMP, only 1 pass
4) G.Skill kit together, different combinations for the A2/B2 and A1/B1, only 1 pass
So, I decided to do that again. I kinda decided to only do 1 pass on the G.Skill because the Kingston 4 passes took almost 3 hours, and that stick is 8GB only, so having two sticks of 16GB, it would take a long time for each of them to get 4 full passes. And if G.Skill RAM has errors, 1 pass should catch most of em anyway, so why waste time even on a supposed good stick if it's not needed. If both G.Skill sticks have no errors on my first pass, I'll up to 2, 3 and so on, until the # of passes are equal to that of what the Kingston RAM (which I'd also up the number of passes if G.Skill RAM would end up needing more passes, just to have something to compare to) managed to do and show no errors.
Anyway, I've also decided to not do A1 and B1 DIMMs tests with G.Skill RAM this time, and test only A2 and B2 DIMMs, while using the Kingston RAM to test all DIMMs first. I only use those two DIMMs, and if the Kingston RAM already passed 4 in those, I'd assume they are not the culprit. But maybe I end up doing one pass in those on stock just to be safe, we will see. Anyway, once again, the Kingston tests, in each DIMM, no errors.
Stick A showed a single error, in B2 DIMM, when i used XMP and used auto for infinity fabric. If I set it manually, at 1800MHz, then it had no errors. However, on A2 DIMM, it showed no error even if the infinity fabric was set to auto for the XMP profile. That was kinda weird. And it was only Test 8. I re-did that test 4 times in one run (4 passes basically of just that one test) with both auto infinity fabric and manually set. And each time, auto had an error or two, manual didn't. Yet, both had same XMP stuff except that infinity fabric setting. This whole test threw me off a bit tbh.
Stick B failed all tests. At stock it only showed a few errors in the last few tests and I also got the NOTE This RAM may be vulnerable to high frequency row hammer bit flips. This only shows on the B stick, and it shows on stock and XMP, doesn't matter.
So, now I'm in the process of redoing the tests, i guess for my own sanity, since this is my third RMA in less than 3 years of when my issues began, and I'd like to measure twice, cut once. And I'm just finishing up my B2 DIMM testing with stick B, with it failing stock and XMP on A2, and currently failed stock on B2. And this time there are many more errors, on both stock and XMP. Before I maybe got a few on stock, and a bit more on XMP, but now, it's easily close to 200 errors on both settings on this stick. Need to see if stick A in this retesting shows different results.
I really don't know what else could be causing these issues. The Kingston RAM passing in all DIMMs kinda makes it clear that this is in fact a problem with the G.Skill memory, no? I mean, the stock speed of that RAM is still faster than the stock speed of the G.Skill one and it passed 4 full cycles, while the G.Skill one failed in the first.Comment
-
I can't with this troubleshooting lol
Ran My Stick A this morning in slot B2.
Stock, pass.
XMP, fail.
It had a bunch of errors and even had that note "RAM may be vulnerable to high frequency row hammer bit flips."
Decided to go to sleep and continue tests when I wake up. Fired up the test for the stick again, in B2, XMP, but I decided to manually enter the power to 1.35v. Mind you, when I select the XMP profile, power is changed from auto to 1.35v. I just clicked on the box, deleted that and manually entered 1.35v, that's it. The stick passed the test.
I then went to BIOS, and allowed the XMP profile to set the power by itself to 1.35v when I select it. It failed.
I then re-did the same test, and again, manually, deleted and re-entered 1.35v for power. The stick passed.
I'll continue my tests, will move on to A2 now, and then go back to Kingston to run the 4 passes on both A2 and B2 to see if that stick passes the tests.
But I'd rather consistent errors (like setting up my RAM in a certain way would always result in the same results), because this, just feels so random and I have no idea how to pinpoint what is the issue. As I said, I already tested all my DIMMs with that Kingston RAM more than I tested G.Skill sticks (passes) and it passed all of them.
I guess I'll just continue to test and get in contact with G.Skill, RMA this and see if they can match my issues like last time. But I don't OC my system (aside from using XMP on RAM), so I don't know if the issue is somewhere else. My PC is air cooled, with my 5800x using that Noctua with double towers, and the case is a full tower with 6 fans in total, 4 intakes (two front, one bottom, one top), 2 exhaust (one top, one back).
I don't know if it's something else, my mobo, CPU, PSU? But the only issues I've had is from Nvidia drivers post 577, causing random black screens, which is a known issue, and the "memory_managment" BSOD I got recently. The system was stable since March 2025 when I first got back my RMA.
And just that SMIBOS setting, always showing 1.2v even when I use XMP, and even manually enter the 1.35v... it just boggles my mind. I've uploaded all the four tests I did, two failed, two passed. I dunno what to make of this now tbh, just all over the place it seems.Comment
-
My guess is that it is a combination of marginal motherboard/CPU and marginal RAM.
Both are probably just on the edge of working, but put them together and they sometimes don't work. i.e. random errors.
Kingston is probably less marginal.
This is an eye diagram showing the voltage levels (Y axis) and time (X-Axis). The grey shading is many many signals (bits of data) overlayed on top of each other.
To get the correct value for a bit of date the signal must never cross the blue box. Crossing on the edge of the blue box can give a random result (a bit flip). So even the tiniest of timing or voltage differences can screw it up. That blue box is probably around 60 picoseconds wide. Needs to be incredibly precise on both sides to work. This is also why the 'training" process is important.
Comment
-
So, how do i test my motherboard? Or CPU, so I am sure they are not the problem. And why am I only getting errors with one stick, on both XMP and stock settings, while the other passes stock settings but is a bit unreliable with XMP?Originally posted by David (PassMark) View PostMy guess is that it is a combination of marginal motherboard/CPU and marginal RAM.
Both are probably just on the edge of working, but put them together and they sometimes don't work. i.e. random errors.
Kingston is probably less marginal.
This is an eye diagram showing the voltage levels (Y axis) and time (X-Axis). The grey shading is many many signals (bits of data) overlayed on top of each other.
To get the correct value for a bit of date the signal must never cross the blue box. Crossing on the edge of the blue box can give a random result (a bit flip). So even the tiniest of timing or voltage differences can screw it up. That blue box is probably around 60 picoseconds wide. Needs to be incredibly precise on both sides to work.
Even in Dual Channel, the errors I am getting are just from the slot with the stick that cant pass any test, xmp or not. No matter how i put them, the errors are consistent in from where they show up.
So, is it worth then RMA-ing my RAM? And why isnt then the Kingston ram affected, even if its not on the edge of working, if my CPU and mobo were, wouldn't they affect the RAM as well, which also runs at higher speeds that the G.Skill one does, both on stock?
Comment
-
As mentioned above, it is just a guess, that both sides are marginal. As this would explain a lot of the behavior.
Some of the randomness may also be due to training differences.Training minutely adjusts the timings when new RAM is inserted in a slot and it can be slightly different each time. We see the same thing in our our hardware designs.
To really know for sure you need some very high end test equipment, like the R&S RTP-K94 Oscilloscope (around $60,000 I believe) and probably an electrical engineering degree with a focus on high speed signalling and several days of spare time. That is to say, nearly no one can properly debug this stuff with any accuracy and it is extremely expensive to do so.
So you are left with trial and error. Swapping parts in and out until you get a combination that works. Of course this should not be necessary as the RAM, CPU & motherboard vendors should do all this testing for you, but corners get cut, the permutations of parts is enormous & testing on the production line is often rather superficial. Manufacturers also often move on to new models of motherboards and RAM before they ever get around to getting the old models stable (i.e. poor support)
Also there are lots of other people online having problems with the Taichi motherboards and RAM compatibility.
https://forum.asrock.com/forum_posts...n-x870e-taichi
https://www.reddit.com/r/ASRock/comm...th_gskill_ram/
https://forum.asrock.com/forum_posts...ichi-ram-issue
Comment
-
Only $60,000? Time to properly test my RAM!Originally posted by David (PassMark) View Post
Originally posted by David (PassMark) View Post
Yeah... I had to tell the G.Skill RMA people to test with MemTest86 because the first RMA they tested my RAM inside of Windows, and I wasn't aware of this meaning they didn't test all the RAM, because duh, Windows is using RAM. That's why I didn't even send the kit back when it failed (showed errors) a few weeks after I got it, because they said they couldn't replicate my issues and so I just kinda gave up. But when at the end of the year the BSOD became more frequent, I sought help, and someone on Reddit pointed out to me that they didn't fully test my RAM. So, then I decided to do a in-depth test of my own, meaning, one stick all slots for both sticks, both sticks A2/B2 and B2/A2, and of course, using my spare 8GB Kingston kit to see if it was actually my slots or not. Sent those tests to them and told them to test my RAM and the replacements with MemTest86. I did tell em to only do the default 4 passes. And it passed, i then also tested that RAM myeslf, 4 passes too, it passed. Should I maybe tell em to do more passes?
Also, those issues with Taichi seem to be for DDR5. But could be it also has problems with G.Skill RAM for some reason, which I can't find anywhere. And it's not like OC my system or anything. Aside from the RAM, nothing really is OCd (i think the CPU OCs itself with Ryzen 5000 series?). I run pretty much everything at stock.
So, yeah, I'll just RMA it. Kinda I guess, this error, happened at a "good" time, since i'll be having surgery and I'll be unable to use my PC for a while, so kinda good timing to fix this now, rather than at some point later when I could be using my PC but am unable to due to waiting for that RMA to be done.
Thanks for your help. The only thing tho, that kinda confused me was why Stick A passed XMP a few times. It also failed those tests too. I'll try to put manually in the voltage and infinity fabric one more time and see if Stick A passes, but either way, time to contact G.Skill.Comment
-
Also, I just installed this software (Zentimings) and I'm interested to know if anything stands out as an issue. Also, why is my tWRRD on my A2 stick 2 while it's 3 on my B2 stick?
Comment
-
There are people who know in detail what all these values mean and their interaction with each other and with various CPU memory controllers. It's complex and it's their full time job to know this stuff. I would never assume I know more than these people, so I trust them to set sensible values. But having said that, I value stability much more highly than getting the last few percent of performance.
I don't touch any of these settings. So I'm not going to spot anything wrong with one of these numbers.
Comment
Comment