How much RAM?

[ScottieMckinley]

Hello. I ran my first PRP a few days ago and the PC crashed. It took several reboots to get it working again. The ram was set to default. I now have it set to 25gb. System total is 32gb. I have now turned off all overclocks. I hope this fixed the problem. New PRP attempt is coming up in a few days. I only have 2 PC's and I could take all the ram and put it into 1 PC for a total of 64gb. Setting are "what makes sense" and "get smallest exponents" Windows 11.

Thanks
Scottie

[Prime95]

For PRP tests extra RAM cannot be put to good use. Prime95 will only use a few hundred MB.

[Denial140]

The RAM settings are (as far as I know) not used for PRP testing. They only determine how much RAM is allowed for use in stage 2 of P-1, P+1 and ECM factoring methods.

Turning off the overclocks may well have helped though - they potentially could've made the machine unstable, causing the crash. If you want to overclock, I would recommend stress testing first to see if it is stable, and can increase overclocks by small increments to see if it remains stable. I think the usual recommendation is the default torture test, but if your overclocks are more specific there are settings to tune towards particular aspects.

[kriesel]

Welcome Scottie to the hunt and the forum.

Originally Posted by ScottieMckinley

The ram was set to default. I now have it set to 25gb.

OP appears to be writing about prime95, not the other GIMPS applications capable of PRP. It's better to specify software, and include version, than leave readers to deduce one and wonder about the other. Also gb normally means gigabits, while GB would be interpreted gigabytes.

Originally Posted by Prime95

Prime95 will only use a few hundred MB.

For normal usage PRP DC, PRP first time tests, LLDC, a few hundred MB or less per worker, yes. Even a dual-worker session on 500M+ exponent PRPs uses less than 3GB ram. Rough guide, and chart, added here in the reference info. Total required can become large on many-core systems if running many workers as prime95 will default to doing (# workers = actual cores / 4).

[ScottieMckinley]

Thanks for the info. I'm running Prime95 30.8. I also noticed the crashed system had mixed ram chips. I have changed the ram to a dual channel 16x2 kit with the bios set to default 2133MHz. I hope its stable enough for PRP, if not D-LL has been running fine.

[Prime95]

Originally Posted by ScottieMckinley

I hope its stable enough for PRP, if not D-LL has been running fine.

It is stable enough for PRP which has fabulous error detection and recovery. In fact an unstable system is better suited to PRP than DC-LL with LL's vastly inferior error detection.

[kriesel]

Originally Posted by ScottieMckinley

I hope its stable enough for PRP, if not D-LL has been running fine.

LL DC can APPEAR to be running fine, while errors sneak through. The Jacobi symbol check applicable to LL has a 50% chance of detecting an error that occurs. The Gerbicz error check (GEC) applicable to PRP has almost zero chance (<<1ppm) of failing to detect an error. In either case, upon detection of an error, the computation backs up to the last thought good save file and tries again. A sufficiently unreliable system can fail to make any progress in PRP, but can produce bad results in LL DC. PRP with GEC is the gold standard for checking hardware reliability.