ECM Takes far longer than estimated time

[Prime95]

Originally Posted by Batalov

The PRP estimated times are about a half of what it really takes; for example, try:
PRP=2,10,249448,1,0,0,"3"

Bug in the code. Estimate is off by log2(10).

[Batalov]

Right, sound about right.
I re-observed the running times and they are about 9-10minutes (the next FFT size kicked in), so 1:0.3 is right.

[petrw1]

Originally Posted by Prime95

The server is now handing out the enhanced ECM cpu credit.
Past ECM credit has been bumped about 20%.

Do you plan to also adjust the code that estimates the completion times? For me it is still out by about a factor of 2. I am completing the assignments in 11.5 hours but the estimate is 5.5 hours

[Rhyled]

Originally Posted by petrw1

Do you plan to also adjust the code that estimates the completion times? For me it is still out by about a factor of 2. I am completing the assignments in 11.5 hours but the estimate is 5.5 hours

Under 26.3b3 (64-bit) the estimates for ECM work in the 5.4M range are about 3:05 apart in my assignments. Actual time is ~ 4:00. That's closer than I used to see at least.

The GHz-day credits now match roughly what I see from LL or TF work (~ 4.25 GHz-days per 3.8GHz i7 core), so that change was well done.

[petrw1]

Just for fun I decided to complete the ECM of 355717; the lowest exponent that was not "DONE" ECM at B1=50,000, B2=5,000,000.

I ran 114 curves (took my PIV only 18 hours).
Sorry, no factor found.

I checked the ECM Progress page about an hour after the test finished and sure enough it shows "DONE" in the first column

BUT it also shows 56 curves in column 2 though no such activity shows up in the Exponent Status report.

[Prime95]

Originally Posted by petrw1

BUT it also shows 56 curves in column 2 though no such activity shows up in the Exponent Status report.

This is normal (and confusing). The column header shows the number of curves that need to be run at that B1 level if you are starting from scratch. However, we aren't starting from scratch. We've already run 300ish curves at B1=50000, which is equivalent to having run 56 curves at B1=250000.

[petrw1]

I took the remaining exponents in the 35x,xxx range and completed the first level of ECM for them (B1/B1=50,000/5,000,000)

In most cases I took the count from about 180 to 280.
80 exponents
~7,700 curves in total
4 factors found.

[KingKurly]

Originally Posted by petrw1

I took the remaining exponents in the 35x,xxx range and completed the first level of ECM for them (B1/B1=50,000/5,000,000)

In most cases I took the count from about 180 to 280.
80 exponents
~7,700 curves in total
4 factors found.

Roughly how many GHz-days, and how long did it take you?

[petrw1]

Originally Posted by KingKurly

Roughly how many GHz-days, and how long did it take you?

80.6 Ghz days total (0.5239 for each 50 curves)
PIV 2.8 GHZ completed 50 curves in just under 10 hours
PIV 3.4 GHZ completed 50 curves in just under 8 hours

[James Heinrich]

Originally Posted by Prime95

For exponents below say 100,000 add a 10% overhead to the time estimate and Primenet credit. Between 100,000 and say 5,000,000 I'll linearly increase the overhead from 10% to 20%

For what it's worth, I've now included these modifications in my credit calculator.