Test: 8 Addr: F335D53F0 Expected: C6AD2417 Actual: C7AD2417 CPU: 0

[Sebarious]

In January I built a new PC with an Asus ROG STRIX Z790-E GAMING WIFI II motherboard, and the only RAM kit in the QVL for 192 GB, the Corsair CMH192GX5M4B5200C38. It shows two versions of it. Mine is the Version 5.53.13, manufactured by SK Hynix. CPU is the i9 14900KF.

Back when I built the machine, even before installing Windows, I ran Memtest86 directly from the motherboard. It came back fine after the four passes. And the machine has been very stable since I built it, running games at their maximum settings, doing MIDI mockups in Cubase Pro 14 with over a hundred VSTi tracks and FXs, and rendering from Blender to multilayer EXRs and compositing in Fusion and Resolve. I didn't overclock anything, I read the BIOS manual and set it up to have the best possible settings without going over the line.

But yesterday something really weird happened. While running a program called FreeFileSync, which compares and syncs contents of folders, the machine rebooted by itself and went straight into the BIOS. I rebooted it manually and it went into the BIOS again. That's when I realized that the BIOS was showing all the SSDs except the one that is set to be the C:\ drive in Windows, and obviously that was the reason it was booting into the BIOS, not finding any OS to boot up to.

I assumed that the SSD was kaput, which I couldn't understand because it's a Samsung 990 Pro 4 TB that I bought in January. And what I was doing with FreeFileSync was an update to a backup in an external mechanical drive, so if anything the SSD was being seriously underused.

So I turned off the machine, disconnected all the USB cables and other stuff and left only the bare minimum, the USB receiver for the mouse and keyboard, and the DisplayPort cable to the monitor. I turned it on and it booted into Windows just fine. I opened Samsung Magician to see if there was anything wrong with the drive, but everything was fine. I set the temperature history to 1 hour and it showed me that around the time it had crashed, the drive's temp was 55º, far below the 100º limit. Everything seemed to be fine, but before going to bed I left Memtest86 from the motherboard and this morning I found this:



Obviously I'm concerned about this, and while I'm good with computers and I have built several machines before, I don't have the knowledge of a professional computer technician or engineer to know what this means.

Is this telling me for sure that the RAM is faulty and I should start the warranty process, or is there anything that I could change in the BIOS to make this error disappear? I don't have extra sticks, so if I have to send these to Corsair, basically I can't use the machine for as long as the turnaround process is.

The BIOS in this motherboard is quite impressive. There are many settings I was familiar with, and many I wasn't, so I made sure to read the manual thoroughly to make sure I was making the right choices.

[David (PassMark)]

Is this telling me for sure that the RAM is faulty

It is sure there is a fault somewhere. Usually the RAM, but sometimes the CPU or motherboard. And sometimes external factors (voltages, power supply, EMI, temperature, clock speeds & timings)
Single errors (that can't be reproduced) can also be "Soft errors"

See this page to start with
https://www.memtest86.com/troubleshooting.htm

The BIOS in this motherboard is quite impressive

In my opinion, modern BIOSs are too complex. With too many options that can easily make the machine unstable. What other electronics device comes with settings that when set slightly wrong the entire device doesn't work?

Also, you are using an old release of MemTest86, but I don't think it would change anything in this case.

basically I can't use the machine for as long as the turnaround process is

You were probably never using the 192GB. You might find you can buy a 32GB kit and limp along with that for a couple of weeks.

Or test 2 sticks at a time and you might just be able to return a pair (rather than all 4). Not sure about this, depends on Corsair's warranty process.

[Sebarious]

It is sure there is a fault somewhere. Usually the RAM, but sometimes the CPU or motherboard. And sometimes external factors (voltages, power supply, EMI, temperature, clock speeds & timings)
Single errors (that can't be reproduced) can also be "Soft errors"

See this page to start with
https://www.memtest86.com/troubleshooting.htm

Thanks, I'll take a look

In my opinion, modern BIOSs are too complex. With too many options that can easily make the machine unstable. What other electronics device comes with settings that when set slightly wrong the entire device doesn't work?

Yeah, seems a little too much. I'm guessing they do it because of the gamers (it is, after all, a gaming motherboard, but that's not why I bought it) so they can overclock with plenty of options. But it seems excessive for the common person. Thing is, there are many things about this motherboard that make it exceptional for creative purposes, like the insane number of USB ports.

You were probably never using the 192GB. You might find you can buy a 32GB kit and limp along with that for a couple of weeks.

Or test 2 sticks at a time and you might just be able to return a pair (rather than all 4). Not sure about this, depends on Corsair's warranty process.

Corsair would probably tell me to send the four sticks because I bought them as a kit. And I didn't buy 192 GB to brag about how much RAM I have. In fact, I have a Mac Studio M1 Ultra with 64 GB of RAM that I love and would be happy to keep using as my main machine, but the problem with Apple Silicon is that once you bought a Mac with a specific amount of RAM, that's all you're ever going to have. They have no RAM sticks anymore, so if I wanted 192 GB I was going to have to spend about $8,000 on a new Mac Studio.

But even if Corsair would allow me to send them two sticks, no chance I would do that. Every single PC technician will tell you that mixing RAM is a terrible idea, even if it's from the same manufacturer and model. So I'm only going to install this kit from one package, not mixing two sticks from the one I have and two from another one.

[David (PassMark)]

mixing RAM is a terrible idea, even if it's from the same manufacturer and model.

Yes, mixing RAM is sub-optimal. But it isn't mixing if you are using the (exact) same model RAM. If this was an actual requirement to only use a single kit, doing a RAM upgrade would be very difficult or very wasteful. (For example it is perfectly fine to use two identical kits of 2 sticks to fully populate 4 slots)

[Sebarious]

Yes, mixing RAM is sub-optimal. But it isn't mixing if you are using the (exact) same model RAM. If this was an actual requirement to only use a single kit, doing a RAM upgrade would be very difficult or very wasteful. (For example it is perfectly fine to use two identical kits of 2 sticks to fully populate 4 slots)

Oh I've been reading for years that the best way to ensure stability is to have all four sticks as part of the same pack, not even mixing 2 sticks from one pack and 2 from another, even if they are both the same exact brand and model.

The good thing is that it seems to have been a fluke. I actually forced myself to wait until all four passes were done, which took over a day, and these were the results:



Note that it says "Consider running in Parallel Mode to exert a heavier load and uncover more potential errors". So I shut down the machine, and when I booted into the BIOS and then into Memtest86, I configured it to run in parallel mode, but also I added all the tests available. There was only one test, the 14th, which it didn't let me add, I can't remember why. But it was more than the 8 tests that it has selected by default.

That was yesterday. So I let it run, and judging by the fans ramping up I can tell it's using all the CPUs (although it seems that it's only using the 16 cores without HT), and not too long it finished the 3rd pass with no errors:


So it would seem to me that I don't have anything to worry about, and the 2 errors were most likely "soft errors". Do you think that's a good take, or should I keep testing?

[David (PassMark)]

I've been reading for years that the best way to ensure stability is to have all four sticks as part of the same pack, not even mixing 2 sticks from one pack and 2 from another, even if they are both the same exact brand and model.

Isn't really true. It should all work fine. All the sticks will follow the same JEDEC standards.

There is a video here testing completely mismatched sticks together. Different models, from different vendors at different speeds & timings.
https://www.youtube.com/watch?v=bTS0ybQ3lCI
It mostly just works fine. So it should be even more fine when using the same model RAM at same speeds.

So it would seem to me that I don't have anything to worry about,

Maybe.
Or the fault could be intermittent. Or dependent on temperature, or EMI, or small voltage variations.
Definitely not systemically faulty however.

[Sebarious]

Isn't really true. It should all work fine. All the sticks will follow the same JEDEC standards.

There is a video here testing completely mismatched sticks together. Different models, from different vendors at different speeds & timings.
https://www.youtube.com/watch?v=bTS0ybQ3lCI
It mostly just works fine. So it should be even more fine when using the same model RAM at same speeds.

Thanks, that's good to know.

Maybe.
Or the fault could be intermittent. Or dependent on temperature, or EMI, or small voltage variations.
Definitely not systemically faulty however.

I wonder if it's because I activated parallel processing. Maybe this RAM is designed to work with several CPU cores and that makes it fail when it's only 1? I mean, you would think that enabling parallel processing would put more strain on the system, but I stopped the new test after it had gone through 3 passes, and it was on test 13 of pass 4. Unfortunately I had to use the machine so I couldn't wait another hour or two for it to finish, but it seemed pointless anyway. It was more than 50% on test 13, so the chances of an error happening after 3.9 passes were not that great.

But I will run another test tonight with 1 CPU and just test 8 selected, which is the one that gave the errors. And I will put the current version of Memtest86 on a USB stick just to be certain.