Strange run time failure (build 2749)

Gordon · 2016-07-29, 15:39

Hi,

running w7pro, 32gig and the above build.

Running some curves on 4007 with B1=800M

Set B2=1E13

Estimated memory use = 21G.

Init = 1888ms
Compute roots of F = 484757ms
Build F from roots = 307416ms

then it blows up with GNU MP : Cannot allocate memory (size=4292934560)

which is only 4 gig.

Taskmanager shows gmp-ecm is only using 16 gig and there is over 8 gig free....

Yes, I do know I could limit memory or increase the number of chunks it does it in, but that doesn't answer the fact that this should work.

screen caps attached

Gordon · 2016-07-29, 17:44

Quote:

Originally Posted by Gordon

Hi,

running w7pro, 32gig and the above build.

Running some curves on 4007 with B1=800M

Set B2=1E13

Estimated memory use = 21G.

Init = 1888ms
Compute roots of F = 484757ms
Build F from roots = 307416ms

then it blows up with GNU MP : Cannot allocate memory (size=4292934560)

which is only 4 gig.

Taskmanager shows gmp-ecm is only using 16 gig and there is over 8 gig free....

Yes, I do know I could limit memory or increase the number of chunks it does it in, but that doesn't answer the fact that this should work.

screen caps attached

Update, using -maxmem 16000 gives exactly the same problem, at the time the system was showing 13 gig free ram.

Gordon · 2016-07-29, 18:43

Quote:

Originally Posted by Gordon

Update, using -maxmem 16000 gives exactly the same problem, at the time the system was showing 13 gig free ram.

and again at -maxmem 14

system shows peak usage at only 11.5 gig

ATH · 2016-07-30, 16:33

Try this svn 2908 I compiled a while ago, maybe you are missing parameters when compiling. If it works I can compile the newest svn:
gmpecm2908.zip

WraithX · 2016-07-31, 18:27

Quote:

Originally Posted by Gordon

Hi,

running w7pro, 32gig and the above build.

Running some curves on 4007 with B1=800M

Set B2=1E13

Estimated memory use = 21G.

screen caps attached

You say 800M, but the command line in your screen shot shows 800e8. I think if you change your command line to 800e6 it may work for you.

Gordon · 2016-08-01, 20:38

Quote:

Originally Posted by ATH

Try this svn 2908 I compiled a while ago, maybe you are missing parameters when compiling. If it works I can compile the newest svn:
gmpecm2908.zip

I am using a pre-compiled version, I tried your one and got the following

with no maxmem it wanted 21gig, more than that free, it failed in "Building F from its roots"

-maxmem 16000 (it wanted 14.75g) failed in "Building F from its roots"

-maxmem 15000 (it wanted 13.11g) failed 8 stages in "Computing G * H Mod F"

-maxmem 14000 (it wanted 13.11g, same as with 15000 param), same fail "Computing G * H Mod F"

-maxmem 13000 (it wanted 11.8g), worked.

Every time it failed, the exact same error

GNU MP: Cannot allocate memory (size=4294909984)

Perhaps it can't allocate memory past the 16 gig boundary?

ATH · 2016-08-01, 21:00

Did you fix your typo making the B1=800e6 instead of 800e8 as WraithX pointed out? Or did you actually run B1 to 800e8 = 8e10 in Prime95?

Can you post your results file from Prime95 that you are trying to resume from? I want to try if I can get it to work on my computer.

Gordon · 2016-08-01, 21:32

Quote:

Originally Posted by ATH

Did you fix your typo making the B1=800e6 instead of 800e8 as WraithX pointed out? Or did you actually run B1 to 800e8 = 8e10 in Prime95?

Can you post your results file from Prime95 that you are trying to resume from? I want to try if I can get it to work on my computer.

WraithX is correct, the first batch I posted screencaps from does show command line as 800E8 which is incorrect, it should be 8e8 or 800e6.

However the runs done today the command line was this

ecm-2908 -v -resume results02.txt 8e8-8e8 1e13

I have attached the p95 stage 1 output file, it contains 10 curves.

MatWur-S530113 · 2016-08-02, 12:13

you could try another directive:

ecm-2908 -v -resume results02.txt 8e8-8e8 8e8-1e13

starting the stage2 at 8e8 and not at 1, but not sure if this helps...

edit: the first curve startet at my computer without problems, result later.
I used the option -no-ntt

edit2: hm strange, I got an error, too.
It seems the 16 GB forces the error, I restart the curve with k=8, result later...

ATH · 2016-08-02, 13:16

I got the same error message on my computer. I have done P-1 and P+1 using 15 GB RAM before but never used this much RAM on ECM. I have about 27 GB free out of my 32 GB RAM.

Building F from its roots took 371906ms
GNU MP: Cannot allocate memory (size=4294934560)

ecm.exe -v -resume results02.txt 8e8-8e8 1e13
k=2, Estimated memory usage: 21.01GB, peak usage 16707M FAILED

ecm.exe -v -resume results02.txt -k 3 8e8-8e8 1e13
k=3, Estimated memory usage: 17.12GB, peak usage 13622M FAILED

ecm.exe -v -resume results02.txt -k 4 8e8-8e8 1e13
k=4, Estimated memory usage: 14.75GB, peak usage 11669M FAILED

ecm.exe -v -resume results02.txt -k 5 8e8-8e8 1e13
k=5, Estimated memory usage: 11.80GB, peak usage 10797M SUCCESS

So for now I guess you have to use the -k 5 option (or the -maxmem 13000 which should give k=5 as well).

These tests was in svn 2908. I compiled the newest svn 2968 and did the k=2 run again and it failed again.

MatWur-S530113 · 2016-08-02, 14:45

same result as ATH at my computer, with k=8 it works correct (peak 9345MB).
I think there is a problem of GMP (not gmp-ecm) allocating more than 16GB of RAM for a single thread.

2016-07-29, 15:39	#1
Gordon Nov 2008 3×167 Posts	Strange run time failure (build 2749) Hi, running w7pro, 32gig and the above build. Running some curves on 4007 with B1=800M Set B2=1E13 Estimated memory use = 21G. Init = 1888ms Compute roots of F = 484757ms Build F from roots = 307416ms then it blows up with GNU MP : Cannot allocate memory (size=4292934560) which is only 4 gig. Taskmanager shows gmp-ecm is only using 16 gig and there is over 8 gig free.... Yes, I do know I could limit memory or increase the number of chunks it does it in, but that doesn't answer the fact that this should work. screen caps attached Attached Thumbnails

2016-08-01, 21:00	#7
ATH Einyen Dec 2003 Denmark 3,037 Posts	Did you fix your typo making the B1=800e6 instead of 800e8 as WraithX pointed out? Or did you actually run B1 to 800e8 = 8e10 in Prime95? Can you post your results file from Prime95 that you are trying to resume from? I want to try if I can get it to work on my computer. Last fiddled with by ATH on 2016-08-01 at 21:02

2016-08-02, 12:13	#9
MatWur-S530113 Apr 2007 Spessart/Germany 2×3⁴ Posts	you could try another directive: ecm-2908 -v -resume results02.txt 8e8-8e8 8e8-1e13 starting the stage2 at 8e8 and not at 1, but not sure if this helps... edit: the first curve startet at my computer without problems, result later. I used the option -no-ntt edit2: hm strange, I got an error, too. It seems the 16 GB forces the error, I restart the curve with k=8, result later... Last fiddled with by MatWur-S530113 on 2016-08-02 at 13:12

2016-08-02, 13:16	#10
ATH Einyen Dec 2003 Denmark 101111011101₂ Posts	I got the same error message on my computer. I have done P-1 and P+1 using 15 GB RAM before but never used this much RAM on ECM. I have about 27 GB free out of my 32 GB RAM. Building F from its roots took 371906ms GNU MP: Cannot allocate memory (size=4294934560) ecm.exe -v -resume results02.txt 8e8-8e8 1e13 k=2, Estimated memory usage: 21.01GB, peak usage 16707M FAILED ecm.exe -v -resume results02.txt -k 3 8e8-8e8 1e13 k=3, Estimated memory usage: 17.12GB, peak usage 13622M FAILED ecm.exe -v -resume results02.txt -k 4 8e8-8e8 1e13 k=4, Estimated memory usage: 14.75GB, peak usage 11669M FAILED ecm.exe -v -resume results02.txt -k 5 8e8-8e8 1e13 k=5, Estimated memory usage: 11.80GB, peak usage 10797M SUCCESS So for now I guess you have to use the -k 5 option (or the -maxmem 13000 which should give k=5 as well). These tests was in svn 2908. I compiled the newest svn 2968 and did the k=2 run again and it failed again. Last fiddled with by ATH on 2016-08-02 at 13:19

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Assertion failure in 6.4.2	bsquared	GMP-ECM	4	2013-03-01 15:52
HELP!!! Strange error on new build...	petrw1	Hardware	2	2009-11-27 05:53
LA Failure	R.D. Silverman	NFSNET Discussion	10	2007-05-23 21:53
What does this failure indication mean?	krunsj	Hardware	5	2004-07-17 16:09
Failure Functioins	Unregistered	Miscellaneous Math	0	2004-02-12 11:51

2016-07-30, 16:33	#4
ATH Einyen Dec 2003 Denmark 3,037 Posts	Try this svn 2908 I compiled a while ago, maybe you are missing parameters when compiling. If it works I can compile the newest svn: gmpecm2908.zip

2016-08-02, 14:45	#11
MatWur-S530113 Apr 2007 Spessart/Germany A2₁₆ Posts	same result as ATH at my computer, with k=8 it works correct (peak 9345MB). I think there is a problem of GMP (not gmp-ecm) allocating more than 16GB of RAM for a single thread.