FFT-Size

[cheesehead]

Originally Posted by jinydu

I'm using version 24.13, and judging by an increase in the per iteration time of well over 10%, there is a FFT length change somewhere between 17440543 and 17776357.

In source module mult.asm you can find in table "xjmptable" that the change is at 17550000.

After playing around with the Time function in the Advanced menu, I have found that the jump occurs at precisely 17.55M.

In the long run, it might be faster to download source2413.zip (3.5Mb) and unzip module mult.asm from the gwnum folder once, then browse the source whenever.

Oddly enough, the per iteration times I got during the Time test were far better than my usual times. Then, when I resumed the usual test, my per iteration time improved dramatically!

Were you running Prime95 under a Windows OS?

[jinydu]

Originally Posted by cheesehead

Were you running Prime95 under a Windows OS?

Yes, Windows XP.

I was getting per iteration times of about 0.100s, whereas my best times now are about 0.125s. Unfortunately, the speed-up quickly wore off, and when I tried the same tactic again, no speed-up occured.

[jinydu]

Anyway, it seems that the Time feature in the Advanced menu of Prime95 will not allow me to play with exponents larger than 99,999,999.

I wonder though: If I play around with Test, will it disrupt my current (serious) LL test? More specifically, will I have to start all over again?

[jinydu]

Originally Posted by jinydu

Unfortunately, I suspect that Primenet may not recognize this jump and could underestimate my credit by over 10% when M17,776,357 finally finishes (it just started less than an hour ago) . Time will tell...

If my memory serves me correctly (I didn't keep a record of what my CPU years credit was before the test finished), my concern was well founded. I missed out on about 0.2 CPU years...

I'll upgrade to version 24.14 to make sure this doesn't happen again.