Originally Posted by Ichirou
The last time I built a PC was almost a decade ago, so I'm just wondering whether the process to testing RAM has changed since then.
The RAM sticks I'll have now will be DDR4-3000, four of them for a total of 64 GB. Do I still use MemTest86, or is there a better alternative now? Could somebody link me?
I've personally run RAMTest (small purchase required), PassMark's MemTest86 (there are multiple versions of this code, so my comments apply only to the free version from PassMark) and Google's stressapptest (something Google uses to test their server hardware).
For just verifying memory stability to rule out an issue or after setting an XMP profile, you can use any of these (and I've heard HCI Memtest is good too). You will have to make absolutely sure you run them all long enough to get a thorough test. Memory errors are not as simple as a given memory cell works or it doesn't work. A memory cell can work if accessed in one way, but not work if accessed with a different pattern a longer test is needed to increase the likelihood that all the bases have been covered with the test.
For example with RAMTest, I run it overnight to 25,000-30,000%. If it passes 4000%, it will probably pass 25,000-30,000% too, but if you're trying to verify stability, you may as well run it overnight. I set it to stop on first error so I can see how far it got if there was an error.
With PassMark's MemTest86 (booted from a USB drive), I like to run it sequentially (not parallel) so it will cycle through the different CPU cores and run all tests and set it for 99 passes. I then run it overnight and let it see how many passes it can do. If it makes it through four passes of all tests, it will probably make it overnight, but again I run it overnight to verify stability.
For Google's stressapptest, people seem to say to run it for an hour. I've run it just to try something different and to verify stability with it, but don't have a feel for how long you have to run it to use it to independently verify stability of your system without running any other tests. I find it the least convenient to run if you're not already Linux bootable. One can apparently also run it under BASH on Windows, but I never found a good set of instructions for how to set it up. For non-Linux people, RAMTest and PassMark's MemTest86 are very turnkey and easy to set up.
For memory overclocking and optimizing of secondary and tertiary timings, where you're likely to have to run hundreds of tests, I find RAMTest to be the most useful tool. It generally finds instabilities fairly quickly and it's really convenient to run (it runs right in Windows) and you can even set it to allow a small amount of remaining RAM for Windows so you can still do something while it's testing (like browse the web, work on a Word document, etc...).
Originally Posted by Ichirou
Also, I've heard rumours that there is the potential for bad sticks of RAM to corrupt your (system) files, so I was wondering if this is true, and if so, is there a way to test the RAM without any storage drives connected, to make sure that they're not faulty before I fully install the PC?
A memory error can absolutely cause data corruption. It all depends upon where the error occurs.
The absolute lowest risk way of testing memory is to test it without booting into Windows. Then, you don't run any risk of corruption your Windows installation. You can do that by running one of the test programs that can be run from a bootable USB or bootable DVD. PassMark's MemTest86 (free version) has a program that will make a bootable USB drive. You reboot your computer, press F11 to select the boot drive, then pick the USB drive and it will boot right into MemTest86 and you can run your tests there. This also allows you to stress test nearly all of your memory because there's no Windows OS already running and occupying memory that can't be stressed.
With more work, you can also put together a bootable *nix disk or USB and run stressapptest from there.
That said, I spent over a month working on overclocking and optimizing my memory configuration and probably had several hundred failed memory tests including dozens of times where it booted through the BIOS, but Windows failed to start and it never once corrupted my Windows installation. I'm not saying it can't happen, just that it didn't happen to me. To be safe, you would boot from a USB or DVD (not Windows) and test from there.
Another safeguard you can take it to do a disk image backup of your Windows boot drive (that's something I did) and then if it does get corrupted and can't recover from its own backup partition, then you can restore the backed up disk image. You will just need an external storage device (I used an older hard drive that I was no longer using connected through a USB dock) and some software. There is free software (Macrium Reflect) that can do that backup.