RAM not 100% stable even after 4 passes?

[h3r3t1k]

I OCed both my CPU and RAM. I tested the RAM and it did 4 passes without errors. But when I ran Prime95 Blend it came up with errors pretty much right away which persisted even after removing the CPU OC. In the end I had to increase CPU voltage to make the CPU stable and go from 3200 MHz to 3066 MHz on the RAM to make it stable. Thing is how could it have passed a full Memtest and still not be 100% stable?

[David (PassMark)]

Maybe the fault was in the CPU, or maybe it was temperature related. Prime95 (in some modes) will heat the CPU more than MemTest86

While MemTest uses the CPU, it isn't really a CPU test. Especially if you are only running it in single threaded mode.

If you think it might be CPU or temperature have a look at BurnInTest Temperature testing.

[h3r3t1k]

I watched the thermals. They can't have been the issue. Also the errors persisted after removing the OC on the CPU. Any way Prime95 Blend can come up with errors while Memtest can't?

[David (PassMark)]

Were you running MemTest86 in single threaded or multi-threaded mode?
Prime95 also runs in Windows. Which means there is a lot more software and hardware being potentially active. Video card, networking, maybe multiple hard drives, a huge number of background services and device drivers, etc.. It isn't impossible there is an interaction with other hardware. Especially is the stability was linked to voltage levels.

[h3r3t1k]

You're saying a "freak error" or even multiple in Prime95 can't be ruled out? I ran Techpowerup Memtest64 for 8.5h and had no errors. So I'm back at 3200 MHz running Prime95 Blend. Againg getting a faltal rounding error. There's no OC on the CPU. I don't know what to do here but I'm inclined to go to 3066 MHz even tho Memtest says 3200 MHz is rock solid.

[David (PassMark)]

Err no, I didn't mention the work freak. Just saying that there are potential explanations for different test scenarios giving different results.

[h3r3t1k]

A redditor suggested upping the voltage a tad and right now it seems that going from 1.35v to 1.4v did the trick. No more errors in Prime95.