P-1 Ram Vs ????

[crash893]

Hi all

Time for some more noob questions.


I was under the impression that more ram was a good thing when doing p-1 testing HOWEVER i recently heard that after a certain point that it actually was less efficient than just running the regular testing?


does anyone have any sort of formula that would tell me where the "sweet" spot is for allocating ram?

[cheesehead]

Originally Posted by crash893

I was under the impression that more ram was a good thing when doing p-1 testing

It is, all else being equal (but see below).

HOWEVER i recently heard that after a certain point that it actually was less efficient than just running the regular testing?

No, that's a misunderstanding.

I answered a similar inquiry in "optimal memory settings for the P-1 stage" at http://mersenneforum.org/showthread.php?t=6454. Read that, then come back here.

Allocating more RAM (notice that I write "RAM", not "memory") to P-1 stage 2 never hurts. (That is, it never hurts Prime95 directly. Others trying to share the system may complain, and if that allocation causes thrashing it'll hurt Prime95, too.)

There's a misunderstanding among some past postings, perhaps for two different reasons:

(A) Someone noticed that when Prime95 chooses B1/B2 and you increase the amount of "Available Memory", Prime95's new choices take a longer running time, but did not understand that doing so also increased the chances of finding a factor by a greater ratio than it increased the running time.

(B) Someone noticed that after increasing "Available Memory", and Prime95 chose a higher B2, it ran slower, but didn't realize that the slowdown was caused by thrashing because the actual RAM wasn't large enough to accomodate both "Available Memory" and all the other tasks on the system at that time.

Thus there came to be posted a mistaken conclusion that more memory after a certain point was less optimal.

BTW, please spread this information back to whoever you heard "after a certain point" from, and tell him/her to please spread the same information back to whoever s/he heard the misunderstanding from ...

does anyone have any sort of formula that would tell me where the "sweet" spot is for allocating ram?

The easy way is to first let Prime95 choose B1 & B2 (which will be the "sweet spot" for your particular combination of several different parameters), then stop it and specify B1/B2 yourself based on that knowledge plus other parameters you know that Prime95 doesn't know. (E.g., Prime95 doesn't know when its memory allocation passes the point where system thrashing starts.)

The hard way is to study the source code where Prime95 chooses B1 & B2.