[SOLVED] RAM Timing - How stable is stable enough?

[kwangle]

I've recently built a Ryzen 7 3700x system with 32GB of Corsair Vengeance 3600 rated RAM (4 x 8GB).

I've entered the arcane world of RAM timing adjustment in the motherboard BIOS (an ASUS Prime 470x PRO). This was very difficult to get into but with tools like Thaiphoon Burner and Ryzen DRAM Calculator I eventually got faster timings to work, seemingly with good stability.

I also tweaked all the sub timings which took weeks of painstaking logging of settings and testing. I have now got complete timings and the system has been running perfectly with no crashes at any time including when doing heavy gaming.

I followed up with RAM stress testing with MemTest86, booting off a USB stick. Now this is where the issues lie. Despite the computer working reliably and stably, I still get sporadic errors when doing a complete 6+ hour RAM stress test. This usually consists of a single bit error in that time, but sometimes it passes with none.

I assume that my settings are on the knife edge between perfect stability and the very start of unstable operation, albeit at extremely unlikely frequency. Are errors of this type likely to occur in normal operation or is a heavy and prolongued RAM stress test the only situation where this is likely to happen?

Many thanks for any advice.

 

[fredfinks]

Dont know, but are you willing to take that chance?
How hardcore are you?

I'd like my RAM to be as good as non-ecc gets. Especially if the effects of oc aren't noticeable.

 

[boju]

Memtest is thorough and tests ram in ways games might not, you'd tend to trust it because the last thing you want is ram randomly failing on you while doing something important. Another test you can try is Prime95 and set a large chunk of ram and test over several hrs.

If this is a hobby and interest learning how memory works and nothing you do on the computer is of risk then that's fine, good on you. But if you're trying to improve performance, the negligible differences might not be worth the hassle. If the latter, i'd just return memory settings to defaults and only enable XMP for it to adjust the primary timings.

 

[kwangle]

No, you are right that random errors aren't acceptable. I am continuing to test the settings to isolate what is causing the instability. There is a decent performance increase with the settings tightened as they are considerably faster than the DOHP defaults but I've probably over-egged the pudding. If I can't get consistent stability I will start loosening the settings a little.

Thanks for all your help.