mixing 2400Mhz with 2666Mhz stability

[rido.bahar02]

Hi guys, I'm running a Ryzen 5 1600 - Asrock AB350 Itx - gtx 1050ti - corsair veangence LPX 8gb 2400Mhz, Corsair v450 PSU, Cougar QBX, Win 10 machine, few days ago I decided to upgrade my RAM, so I bought another 8gb of the same brand and model, except I chose the wrong speed, I bought the 2666Mhz.
After a few tries I have the machine running stable, I set the Ram on xmp profile by OC-ing the speed to 2666 without any errors whatsoever, although I did not touch the timing as I don't know anything about it and didn't wanna screw up. Both Ram are running on dual channel recognized to be 2666 on windows taskmanager and Cpu Z, tried a few memory benchmark also getting a good result, exept for the latency which is pretty high touching 65 to 80, but then I read few articles that are saying this latency is generally amd problem and cause no noticable impact on the performance (please correct me if I'm wrong).

So at this point I am very happy with the Machine, though its not a big thing but feels like a free upgrade for the old ram, but my question is, is it really stable for long term? did I do enough test or did I miss anything?
I did request for a 2400Mhz replacement but I have to wait for another week, so should I cancel the replacement and go with this?

By the way, I use my machine for my professional work daily driver, I do simulation and rendering a lot on this machine.

Thank you.

 

[Darkbreeze]

You don't mention what you used to test?

Memory testing is best done in two steps or three steps.

Step one, download Memtest86 from Passmark. Just get the free version. Create bootable Memtest media and boot to the USB drive that contains Memtest. Run it for four full passes. If there are no errors, great. If there are, the memory is not configured to be stable.

Next step, if you passed step one, run Windows memory diagnostic tool on the Extended setting for 3 passes. If there are no errors, go to step three.

Step three, download and run Prime95 version 26.6 OR any newer version, but you must edit the local.txt file contained in the Prime95 program folder. Change the setting in the local.txt folder as follows, CpuSupportsAVX=0 (instead of CpuSupportsAVX=1) .

Open Prime95 and choose the Custom option. Change the min FFT to 512k. Change the memory to test field to 75% of your current free memory. You can use HWinfo or Windows resource monitor to establish how much of your memory is currently not in use, and then enter approximately 75% of that amount. Run the Custom test for 8 hours. Alternatively you can simply leave the minimum FFT at it's default setting and can also simply enter 50% of your total installed memory capacity, so if you have 16GB installed, enter 8GB, and run that for 8 hours.

If all these pass, then your memory configuration is as stable as you're likely to be able to validate, and you'll be very unlikely to experience any instability or cumulative micro-errors.

 

[rido.bahar02]

You don't mention what you used to test?

Memory testing is best done in two steps or three steps.

Step one, download Memtest86 from Passmark. Just get the free version. Create bootable Memtest media and boot to the USB drive that contains Memtest. Run it for four full passes. If there are no errors, great. If there are, the memory is not configured to be stable.

Next step, if you passed step one, run Windows memory diagnostic tool on the Extended setting for 3 passes. If there are no errors, go to step three.

Step three, download and run Prime95 version 26.6 OR any newer version, but you must edit the local.txt file contained in the Prime95 program folder. Change the setting in the local.txt folder as follows, CpuSupportsAVX=0 (instead of CpuSupportsAVX=1) .

Open Prime95 and choose the Custom option. Change the min FFT to 512k. Change the memory to test field to 75% of your current free memory. You can use HWinfo or Windows resource monitor to establish how much of your memory is currently not in use, and then enter approximately 75% of that amount. Run the Custom test for 8 hours. Alternatively you can simply leave the minimum FFT at it's default setting and can also simply enter 50% of your total installed memory capacity, so if you have 16GB installed, enter 8GB, and run that for 8 hours.

If all these pass, then your memory configuration is as stable as you're likely to be able to validate, and you'll be very unlikely to experience any instability or cumulative micro-errors.

I used passmark performance test which shows decent result in my understanding,
https://ibb.co/h45sre

and, userbenchmark which shows good result
https://ibb.co/cVs5Be

I also did a couple multiframes scene renders succesfully.

I didn't feel sure about using Memtest86 coz it sounds too technical for me as you need to boot on it, but if you say I still need to do it I might gonna take a look at how to do it safely, big thanks mate.

 

[Darkbreeze]

Neither of those tests are for testing stability, at all. They are simply benchmarking utilities to weigh synthetic performance.

Running Memtest86 is a MANDATORY step anytime you make changes to your memory configuration, at the absolute minimum, unless you are ok with running an unstable system that is probably going to become unusable and unrecoverable at some point in the next few months depending on the level of instability. Even if you don't get pure blue screen errors or freezing, it's is not only likely, but probable, that with an automatic configuration (Or even a manual one for that matter) of 2400mhz memory overclocked to 2666mhz, and two modules that are not even the same specifications, that you are at best getting micro-errors that you are not aware of.

Micro-errors from unstable memory is an unseen problem, getting a one where a zero should be, a couple of zeros where there should be a one, and before you know it there is serious cumulative corruption of data. This will not be something you can correct. ALL corrupted data will basically be unusable and unrecoverable. This could be the OS, personal files, game data, anything that has been processed and written or edited. Memory data corruption from unstable configurations is no joke and this is not a case of being overly cautious. It's a serious and well documented fact. It is not worth the risk. Do the testing. Be sure, or else run them both at the default speed of 2133mhz, OR at the very least, set both modules up to run at the speed of the lowest speed module, including using the same specs on your 2666mhz stick as what is specified for the 2400mhz module.

There is a lot less chance, but still a chance ANY time you set memory to the XMP profile or manually configure them higher than that, that you'll have instability problems running them both at the lower speed than running them both at the higher speed. Consider, even running ONLY the higher speed stick, or either stick alone for that matter, at the XMP values, they should STILL be tested because those values actually constitute an overclock and the values used for the XMP profile while tested by the manufacturer and found to be pretty stable, are not guaranteed. You still need to test memory anytime you make automatic or manual adjustments beyond what the motherboard specifications state is supported without an OC next to the memory speed specification.