ECM users - version 30.9/30.10 (see post#168)

[congsz]

Originally Posted by axn

Apparently this is the unix time

On Windows, it looks like this:
PolyR built. Time: 188831.706 sec.

[Prime95]

Originally Posted by axn

A cosmetic bug in b2
nQx complete. Time: 5.552 sec.
PolyR built. Time: 1664077385.142 sec.
Poly compress. Time: 0.510 sec.

Adding PolyVerbose=1 to prime.txt will fix this (and get even more useless timings :)

[axn]

Originally Posted by Prime95

Adding PolyVerbose=1 to prime.txt will fix this (and get even more useless timings :)

No thanks, I'm good :)

[Stef42]

Originally Posted by Prime95

30.9 build 2

If you aren't doing ECM, do not download! PRE-BETA software!

What's new over build 1? Should be a little more stable and a little faster. Fixed nasty bug that sometimes caused subsequent curves to incorrectly calculate curve starting point.

Caveats? Same as build 1:
Save files during ECM stage 2 are still broken (probably a crash)
Stage 2 time estimates and optimal B2 bounds could be off
Accurate estimates of stage 2 memory consumed may be off
Further stage 2 multithreading improvements are needed.
Stage 2 is pretty verbose, there's lots of code cleanup in my future.

Windows 64-bit: https://mersenne.org/ftp_root/gimps/p95v309b2.win64.zip
Linux 64-bit: https://mersenne.org/ftp_root/gimps/...linux64.tar.gz

I was running build 1 for about two months and found 1 ECM factor. Ran one machine for ECM-CF tasks since then. Was a bit surprised with no factors, but as you mentioned the bug I updated to build 2 today.
Found 2 factors already since this morning. A bit worrying or not? (I might want to re-do my previous tasks with build 1 for example)

[LaurV]

Is this affecting ECMF too? I am currently running few curves on F21 right now, and would like to finish them before going to "more serious" things. Should I stay with 30.8, or upgrade? Thanks in advance.

[Prime95]

Originally Posted by LaurV

Is this affecting ECMF too? I am currently running few curves on F21 right now, and would like to finish them before going to "more serious" things. Should I stay with 30.8, or upgrade? Thanks in advance.

Yes, ECMF is affected. I'd give 30.9 try if you can live with its shortcomings (mainly no save files in stage 2).

[LaurV]

Ok, no issue, it runs. Verbose messages don't bother me, maybe I have a chance to understand better what's goin on .

This is much faster too, the other one said ~11+ days to go, this one says ~8+ days to go.

Thanks.

[Prime95]

30.9 build 3

If you aren't doing ECM, do not download! PRE-BETA software!

What's new over build 2? Completed fix for nasty bug that sometimes caused subsequent curves to incorrectly calculate curve starting point. Previous fix did not solve the issue.

Caveats? Same as build 1:
Save files during ECM stage 2 are still broken (probably a crash)
Stage 2 time estimates and optimal B2 bounds could be off
Accurate estimates of stage 2 memory consumed may be off
Further stage 2 multithreading improvements are needed.
Stage 2 is pretty verbose, there's lots of code cleanup in my future.

Windows 64-bit: https://mersenne.org/ftp_root/gimps/p95v309b3.win64.zip
Linux 64-bit: https://mersenne.org/ftp_root/gimps/...linux64.tar.gz

[joejoefla]

Just ran the 30.9 b3 build. Running 11600k with 32GB of ram on Win10.

Tried running a ECM test. Forgot to change the memory resources usage for ECM testing and left it at 0.3. It kept crashing because as soon as Curve 1 Stage 2 started it complain about insufficient memory, 16MB needed. As soon as I changed the memory resources from 0.3 to 16 it worked fine.

I can't seem to replicate the bug

UPDATE:

Ok so as far as I understand, if I start with a fresh folder of 309b3, run a ECM test, it'll crash once it starts curve1 stage2. The memory resources were left at 0.3. If I go back to making a fresh 309b3 folder, rerun the same ecm test, then quickly change the memory resources from 0.3, to 16, click ok, then change memory resources back to 0.3, then click ok, curve 1 stage 2 seems to start fine.

Is it because when stage 2 starts there's a bug where memory resources isn't being read correctly?

[James Heinrich]

Originally Posted by joejoefla

Just ran the 30.9 b3 build. Running 11600k with 32GB of ram on Win10.
... insufficient memory, 16MB needed. As soon as I changed the memory resources from 0.3 to 16 it worked fine.

I hope you're giving it more than 16MB to work with on a 32GB system.

[kriesel]

Or 300MB ~0.3GB. Downloaded defaults are not enough for polynomial-paired P-1, for wavefront DC or first test exponents. joejoefla doesn't say what exponent is being attempted or how much free ram his system has, so it's not practical to give guidance for ECM memory requirements, beyond what is in the program's readme.txt, which appears to me to not consider the additional ram needed for polynomial pairing to reach its full potential. It takes lots of Mersenne-number-size or rather IBDWT-fft-length-size buffers to do so, and that takes a lot of ram for stage 2. If there's plenty of free available ram (at least several GB) on the system, it's probably safe to increase prime95 allowed ram by a few GBs.

George has stated that the program ships with allowed stage 2 ram set to a safely low value, so it does not risk sending new users' systems into page faulting and slowing down their other usage, alienating them early.
Look in the readme.txt section about stage 2 ram, and consider that in the thread about DC wavefront remedial P-1 using polynomial pairing introduced in v30.8, George set a minimum of 24GiB allowed stage 2 ram when it started at 63M exponent and is now around 74M exponent. That's ~50 times higher ratio ram/exponent than in the readme.