How to stability test memory?

[kiicki]

I have been overclocking my memory, but had some mixed test results. At 1.38v, Realbench had no issues running for an hour. Aida64 detected issues after a few seconds.

 

Changing to 1.39v, Aida64 was fine running for an hour before I ran Memtest86. I used default settings where I disabled "hammer test" and it took me around 2 hours to complete the test. No errors which is good. Then I decided to use Aida64 again but for a longer time. After 2 hours or so, I noticed that my PC had rebooted. I increase to 1.4v and tried Aida64 again, but it crashed after 19 min. On 2nd attempt it failed again where my monitor stopped displaying, and my on/off button on my case didn't work. I don't think this is related to overclock or stress testing since this has happened before. I also noticed that Windows update tried to find updates around the time Aida64 crashed for the 2nd time.

 

I'm a bit confused as I get these weird results. Sometimes there's no issues, and sometimes there are. How do you guys test for memory?

 

• Memory frequency: DDR4-3200MHz
• DRAM CAS# latency: 16
• Trcdrd 18
• Trcdwr: 18
• DRAM RAS# PRE time: 18
• DRAM RAS# ACT time: 36
• DRAM voltage: 1.4
• ProcODT: 60 Ohm
• SOC Voltage: 1.1v

•  
• BIOS Version: 3803 (Latest)
• CPU: AMD Ryzen 5 1600 3,2GHz Socket AM4 Box
• CPU Cooler: AMD Wraith Spire cooler (Stock)
• Motherboard: Asus ROG Strix B350-F Gaming
• Memory: Corsair Vengeance LPX Black DDR4 PC25600/3200MHz CL16 2x8GB (CMK16GX4M2B3200C16)
• Video Card: Asus GeForce GTX 1060 Dual OC 2xHDMI 2xDP 6GB
• Power Supply: EVGA SuperNOVA G3 550W
• SSD: Samsung 840 EVO 500GB

[porina]

Ram OC isn't easy as you just found out. It takes a lot of testing for confidence. I found I needed a range, any of which may fail under different scenarios. memtest86+, aida64 mem, prime95 blend find different problems at different times. P95 in particular I think is a particularly heavy stress on the IMC (on Intel at least), at the right FFT sizes.

 

I'm not experienced with doing this on Ryzen, but on Intel, it was more the supporting voltages that were more limiting than the main ram voltage.

 

Maybe use free thaiphoon burner to find out what chips are in the ram, and see what else others are getting with what settings on it? Dunno if aida64 might also be able to detect the chip type.

[kiicki]

16 minutes ago, porina said:

Ram OC isn't easy as you just found out. It takes a lot of testing for confidence. I found I needed a range, any of which may fail under different scenarios. memtest86+, aida64 mem, prime95 blend find different problems at different times. P95 in particular I think is a particularly heavy stress on the IMC (on Intel at least), at the right FFT sizes.

 

I'm not experienced with doing this on Ryzen, but on Intel, it was more the supporting voltages that were more limiting than the main ram voltage.

 

Maybe use free thaiphoon burner to find out what chips are in the ram, and see what else others are getting with what settings on it? Dunno if aida64 might also be able to detect the chip type.

I have disconnected the internet cable, and applied 1.39v for the DRAM. I kinda just want to start over because I'm not 100% sure what it failed before. The fact that my monitor went black and on/off button on the case didn't work, doesn't mean it was because of overclocking or stress testing. It has happened before, and I have no idea if Windows update had anything to do with it. I will run these tests for 12-24 hours. Some might find that overkill for memory, but I will most likely use Aida64 and/or RealBench for CPU overclocking later which will also run 12-24 hours. Makes no sense to run memory for a couple of hours, and CPU for 12-24 hours if you are going to use the same application for memory and CPU overclocking. If it would fail, I wouldn't really know what failed.

 

I guess I will do long tests for memory with Aida64, Realbench, and Memtest86. I don't want to run these for a few hours in case my CPU overclock fails at the 6 hour and I have no idea if it was my memory or CPU. I will probably do 8 passes or so with Memtest86

[Leonard]

@kiicki, OCing RAM is tedious and can take a long time to get really stable. The crash could be due to just general heat build up in the case and not instability. Did you check for any thermal throttling during the tests? AIDA64 provides a better test IMO and with the latest build they have DDR4 support, which is not to say you can't use any other stress tester of your choosing.

 

Method:

1. start doing your tests at the RAM's XMP setting and once you do not get any crashes then do a slight bump in frequency, timings and carry the voltage to the maximum allowed and test again, if stable incrementally increase the frequency until you get a crash then reduce the frequency until you don't get any crashes then for that new frequency increase timings and then test the frequency that crashed. Repeat that method until you get as high as you can go or wherever you find is good enough

Tips:

1. when you start to get crashes you should firstly adjust the second to last number in the timings firstly, then to fine tune your OC you can adjust the command rate, in the timings <------hope i made this clear......i doubt it though
2. some RAM chips just don't like being OCed past the XMP so this might be what you are experiencing
3. best RAM chips to OC are Samsung which is not to say that other vendors won't OC just that Samsung's chips are very forgiving

Remember doing this on your own manually can take weeks to get stable and then there is the last and final test in which you use all the programs and games you use normally and check for issues.

[kiicki]

1 hour ago, Leonard said:

@kiicki, OCing RAM is tedious and can take a long time to get really stable. The crash could be due to just general heat build up in the case and not instability. Did you check for any thermal throttling during the tests? AIDA64 provides a better test IMO and with the latest build they have DDR4 support, which is not to say you can't use any other stress tester of your choosing.

 

Method:

1. start doing your tests at the RAM's XMP setting and once you do not get any crashes then do a slight bump in frequency, timings and carry the voltage to the maximum allowed and test again, if stable incrementally increase the frequency until you get a crash then reduce the frequency until you don't get any crashes then for that new frequency increase timings and then test the frequency that crashed. Repeat that method until you get as high as you can go or wherever you find is good enough

Tips:

1. when you start to get crashes you should firstly adjust the second to last number in the timings firstly, then to fine tune your OC you can adjust the command rate, in the timings <------hope i made this clear......i doubt it though
2. some RAM chips just don't like being OCed past the XMP so this might be what you are experiencing
3. best RAM chips to OC are Samsung which is not to say that other vendors won't OC just that Samsung's chips are very forgiving

Remember doing this on your own manually can take weeks to get stable and then there is the last and final test in which you use all the programs and games you use normally and check for issues.

Thanks for your answer. I won't be doing overclocking above the memory specs. All I'm trying to accomplish is to run them stable at 3200MHz. When I applied 1.39v, it passed Memtest86 without any errors. This was with default settings where I removed the hammer test. With Aida64, it crashed after 2 hours. When I do memory testing, should I run Aida64 with stressing CPU, FPU, cashe, and system memory enabled? This is what I have been doing, or should I only enable system memory? I assume I should just enable all 4 even for just memory stability right?

 

Also after 1.39v passed memtest86, but failed after 2 hours with Aida64, I increased to 1.4v. This led to my system crashing after 19 min which was weird. I think that was just some bad luck as I saw that Windows update was trying to look for updates around that time. Makes no sense why it would crash that early when max temp was around 50c

[Leonard]

2 minutes ago, kiicki said:

Thanks for your answer. I won't be doing overclocking above the memory specs. All I'm trying to accomplish is to run them stable at 3200MHz. When I applied 1.39v, it passed Memtest86 without any errors. This was with default settings where I removed the hammer test. With Aida64, it crashed after 2 hours. When I do memory testing, should I run Aida64 with stressing CPU, FPU, cashe, and system memory enabled? This is what I have been doing, or should I only enable system memory? I assume I should just enable all 4 even for just memory stability right?

 

Also after 1.39v passed memtest86, but failed after 2 hours with Aida64, I increased to 1.4v. This led to my system crashing after 19 min which was weird. I think that was just some bad luck as I saw that Windows update was trying to look for updates around that time. Makes no sense why it would crash that early when max temp was around 50c

With stressing the memory you should just run the memory testing at first to know if the memory is stable at whatever frequency, timings, and voltage you have it set at. 

 

When next you test with AIDA64 use the sensor page to monitor the temperatures and keep an eye on the MOBO temperature and not say CPU only because the heat generated can make the MOBO temperature go up as well and then a crash will happen but it would be the general temperature causing the crash and not memory instability.

 

keep the voltage to the 1.39v or see if you can go like 1.395v and try again.

 

[kiicki]

19 hours ago, Leonard said:

With stressing the memory you should just run the memory testing at first to know if the memory is stable at whatever frequency, timings, and voltage you have it set at. 

 

When next you test with AIDA64 use the sensor page to monitor the temperatures and keep an eye on the MOBO temperature and not say CPU only because the heat generated can make the MOBO temperature go up as well and then a crash will happen but it would be the general temperature causing the crash and not memory instability.

 

keep the voltage to the 1.39v or see if you can go like 1.395v and try again.

 

So I have been fiddling some more. I ran memtest86 for 8 passes, disabled "hammer test" and it took me around 5 hours without any issues. I ran Real bench for 4.5 hours without any issues. Then I ran Aida64 and I get an error after 10 seconds. I try again and it fails after 20-30 second. I then try Aida64 for the third time, but without opening the prebuilt "CPUID" and no crashes for 5 min. In the stress test I opened "CPUID" and it crashed immediately. Forth time with CPUID opened and it crashes again.

 

Something tells me that there's a bug if CPUID is opened while stress testing. I get errors.

[Leonard]

4 minutes ago, kiicki said:

So I have been fiddling some more. I ran memtest86 for 8 passes, disabled "hammer test" and it took me around 5 hours without any issues. I ran Real bench for 4.5 hours without any issues. Then I ran Aida64 and I get an error after 10 seconds. I try again and it fails after 20-30 second. I then try Aida64 for the third time, but without opening the prebuilt "CPUID" and no crashes for 5 min. In the stress test I opened "CPUID" and it crashed immediately. Forth time with CPUID opened and it crashes again.

 

Something tells me that there's a bug if CPUID is opened while stress testing. I get errors.

Sounds so to me too and good find but i have never heard of this one, anyway do the AIDA64 test without opening the CPUID and see how long it runs and if it just runs for hours then you know it is a bug that is causing the crashes.