GPU to 72 status... - Page 270

chalsall · 2014-05-30, 18:51

Quote:

Originally Posted by TheMawn

What happened? Did we just fall behind?

The issue is that many requests for Cat 4 never complete. Further, many of the requests are by way of "Manual Assignments", most of which are made by "Anonymous", and yet are valid for 180 days.

I earlier sent a PM to George suggesting he consider lowering the Cat 4 cut-off back down to 50,000 to deal with this. But the decision is his.

James Heinrich · 2014-05-30, 22:23

Quote:

Originally Posted by James Heinrich

I'll try making a list as kracker suggested for benchmarking.

Here's my first attempt. Contains the first-found factor (excluding factors found in class-0) for an exponent in (10M < exponent < 100M) with factor sizes between 2⁶⁴ and 2⁸⁰ (where available). There are 1289 entries. You can run the benchmark as

Code:

mfaktc-win-32.exe >> benchmark.txt

Note with the double >> it will append to the output file so you can Ctrl-C to abort the benchmark and later resume it.
Benchmark should be run with no other GPU load present, obviously.
mfaktc.ini must have PrintMode=0 please.
Having mfaktc0.20-standard header config would be good for consistency:

Code:

# mfaktc 0.20 (default)
ProgressHeader=Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
ProgressFormat=%d %T | %C %p%% | %t  %e |   %g  %s  %W%%

As of yet I don't have anything useful to do with the data, but if anyone wants to run through the benchmark set and email me the results I'll see if I can massage the output into something resembling useful data.

TheMawn · 2014-05-30, 23:25

You have dozens of lines like

Factor=10409843,79,80

That benchmark will take weeks to complete...

kracker · 2014-05-30, 23:45

Quote:

Originally Posted by TheMawn

You have dozens of lines like

Factor=10409843,79,80

That benchmark will take weeks to complete...

Have you actually run it?

James Heinrich · 2014-05-31, 01:11

Quote:

Originally Posted by TheMawn

That benchmark will take weeks to complete...

I haven't actually run it through myself, but the ideas is that each exponent list does have a factor, and where possible that factor will be found within the first few classes and so runtime per exponent should be mostly within the order of seconds-to-minutes. There are undoubtedly some worse-case instances where, due to having a poor selection of known factors to choose from, the factor won't be found until later in the run and each iteration taking considerable time. I haven't (yet) checked for expected runtime for each case. Perhaps that is something I should build into my benchmark-generator.

James Heinrich · 2014-05-31, 02:33

Quote:

Originally Posted by James Heinrich

Here's my first attempt.

Here's my second version. Same basic file, but omitting any assignment that would take more than 1 GHz-day. That takes the count down from 1289 to 868 entries.

LaurV · 2014-05-31, 03:17

Pure stupid question, I just woke up and didn't have my coffee yet, but I don't get it why the class-0 has to be excluded (?!) Forgive my brain fart if any.

kracker · 2014-05-31, 03:19

Quote:

Originally Posted by LaurV

Pure stupid question, I just woke up and didn't have my coffee yet, but I don't get it why the class-0 has to be excluded (?!) Forgive my brain fart if any.

see:

At least for my GPU, it is fastest after ~10 sec(warmup?), not the very beginning.

EDIT: even with PrintMode=1...

Code:

mfakto 0.14-MGW (64bit build)


Runtime options
  Inifile                   mfakto.ini
  Verbosity                 1
  SieveOnGPU                yes
  MoreClasses               yes
  GPUSievePrimes            88629
  GPUSieveProcessSize       24Ki bits
  GPUSieveSize              96Mi bits
  FlushInterval             8
  WorkFile                  worktodo.txt
  ResultsFile               results.txt
  Checkpoints               enabled
  CheckpointDelay           300s
  Stages                    enabled
  StopAfterFactor           class
  PrintMode                 compact
  V5UserID                  none
  ComputerID                none
  TimeStampInResults        no
  VectorSize                2
  GPUType                   AUTO
  SmallExp                  no
  UseBinfile                mfakto_Kernels.elf
Compiletime options
Select device - Get device info - Loading binary kernel file mfakto_Kernels.elf
Compiling kernels.

OpenCL device info
  name                      Capeverde (Advanced Micro Devices, Inc.)
  device (driver) version   OpenCL 1.2 AMD-APP (1526.3) (1526.3 (VM))
  maximum threads per block 256
  maximum threads per grid  16777216
  number of multiprocessors 10 (640 compute elements)
  clock rate                1020MHz

Automatic parameters
  threads per grid          256
  optimizing kernels for    GCN

Started a simple selftest ...
######### testcase 1/17 (M60003157[58-59]) #########
######### testcase 2/17 (M60008017[58-59]) #########
######### testcase 3/17 (M60009827[58-59]) #########
######### testcase 4/17 (M50863909[69-70]) #########
######### testcase 5/17 (M51375383[69-70]) #########
######### testcase 6/17 (M51406301[70-71]) #########
######### testcase 7/17 (M47644171[70-71]) #########
######### testcase 8/17 (M51038681[71-72]) #########
######### testcase 9/17 (M49717271[71-72]) #########
######### testcase 10/17 (M50752613[72-73]) #########
######### testcase 11/17 (M50908933[72-73]) #########
######### testcase 12/17 (M53076719[73-74]) #########
######### testcase 13/17 (M53123843[74-75]) #########
######### testcase 14/17 (M60009109[34-35]) #########
######### testcase 15/17 (M60002273[34-35]) #########
######### testcase 16/17 (M60004333[63-64]) #########
######### testcase 17/17 (M3321928703[90-91]) #########
Selftest statistics                                    
  number of tests           108
  successful tests          108

selftest PASSED!

got assignment: exp=10515217 bit_min=64 bit_max=65 (0.36 GHz-days)
Starting trial factoring M10515217 from 2^64 to 2^65 (0.36GHz-days)
Using GPU kernel "cl_barrett15_69_gs_2"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
May 30 20:24 |    0   0.1% |  0.160   2m33s |    199.87    88629    0.00%
May 30 20:24 |    8   0.2% |  0.170   2m43s |    188.12    88629    0.00%
May 30 20:24 |   12   0.3% |  0.170   2m43s |    188.12    88629    0.00%
May 30 20:24 |   15   0.4% |  0.170   2m43s |    188.12    88629    0.00%
May 30 20:24 |   20   0.5% |  0.170   2m42s |    188.12    88629    0.00%

M10515217 has a factor: 30604828569994966241

found 1 factor for M10515217 from 2^64 to 2^65 (partially tested) [mfakto 0.14-MGW cl_barrett15_69_gs_2]
tf(): total time spent:  0.840s (36548.21 GHz-days / day)

got assignment: exp=10720453 bit_min=65 bit_max=66 (0.70 GHz-days)
Starting trial factoring M10720453 from 2^65 to 2^66 (0.70GHz-days)
Using GPU kernel "cl_barrett15_69_gs_2"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
May 30 20:24 |    0   0.1% |  0.340   5m26s |    184.51    88629    0.00%
May 30 20:24 |    3   0.2% |  0.330   5m16s |    190.11    88629    0.00%
May 30 20:24 |    8   0.3% |  0.340   5m25s |    184.51    88629    0.00%

M10720453 has a factor: 46314683458850504129

found 1 factor for M10720453 from 2^65 to 2^66 (partially tested) [mfakto 0.14-MGW cl_barrett15_69_gs_2]
tf(): total time spent:  1.010s (59629.21 GHz-days / day)

got assignment: exp=10030583 bit_min=66 bit_max=67 (1.49 GHz-days)
Starting trial factoring M10030583 from 2^66 to 2^67 (1.49GHz-days)
Using GPU kernel "cl_barrett15_69_gs_2"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
May 30 20:24 |    0   0.1% |  0.710  11m21s |    188.87    88629    0.00%
May 30 20:24 |   12   0.2% |  0.710  11m20s |    188.87    88629    0.00%
May 30 20:24 |   13   0.3% |  0.710  11m19s |    188.87    88629    0.00%
May 30 20:24 |   16   0.4% |  0.710  11m19s |    188.87    88629    0.00%
May 30 20:24 |   21   0.5% |  0.710  11m18s |    188.87    88629    0.00%
May 30 20:24 |   25   0.6% |  0.700  11m08s |    191.57    88629    0.00%
May 30 20:24 |   28   0.7% |  0.720  11m26s |    186.25    88629    0.00%
May 30 20:24 |   33   0.8% |  0.710  11m16s |    188.87    88629    0.00%
May 30 20:24 |   37   0.9% |  0.700  11m06s |    191.57    88629    0.00%

M10030583 has a factor: 77685810496363030583

found 1 factor for M10030583 from 2^66 to 2^67 (partially tested) [mfakto 0.14-MGW cl_barrett15_69_gs_2]
tf(): total time spent:  6.390s (20146.36 GHz-days / day)

got assignment: exp=10715053 bit_min=67 bit_max=68 (2.79 GHz-days)
Starting trial factoring M10715053 from 2^67 to 2^68 (2.79GHz-days)
Using GPU kernel "cl_barrett15_69_gs_2"

Found a valid checkpoint file.
  last finished class was: 15

Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
May 30 20:24 |   20   0.6% |  1.330  21m09s |    188.77    88629    0.00%
May 30 20:24 |   23   0.7% |  1.330  21m07s |    188.77    88629    0.00%
May 30 20:24 |   27   0.8% |  1.330  21m06s |    188.77    88629    0.00%
May 30 20:24 |   36   0.9% |  1.330  21m05s |    188.77    88629    0.00%

LaurV · 2014-05-31, 03:45

The things are not related. I mean what you (kracker) say, and what I say. (I am still at the coffee

)

The benchmark is good as it is.

But I just complained (read: nitpicking) about the fact that class-0 has no reason to be excluded. There are plenty of factors in class 0 too:

Code:

gp> d=4620; p=7; while(1, p=nextprime(p+1); forstep(k=d,10^10,4620,q=2*k*p+1; if([0,1,0,0,0,0,0,1][q%8+1],if(Mod(2,q)^p==1, print(p","q),
break))); printf("...%d...%c",p,13))
11621,107378041
294403,2720283721
1797589,16609722361
1995317,18436729081
2233547,20637974281
2639761,24391391641
3361639,31061544361
3391831,31340518441
3395123,31370936521
4022371,37166708041
4785769,44220505561
4797073,44324954521
4902077,45295191481
5114959,47262221161
5735981,53000464441
5817389,53752674361
5863271,54176624041
6082619,56203399561
7651261,70697651641
7907231,73062814441
8181751,75599379241
...8708663...
  ***   at top-level: d=4620;p=7;while(1,p=nextprime(p+1);f
  ***                                  ^--------------------
  ***   user interrupt after 45,481 ms.

In the code above, "d" is the class and you can supply any number between 1 and 4619. Of course, for class-0, you have to supply 4620, otherwise all factors are 2*0*p+1=1, and you may get some error - maybe that was James' issue when he queried the DB ?

edit: Or does mfaktX test the class 0 at the end, same as in my example?? (I don't remember that, but I think is tested first).

NickOfTime · 2014-05-31, 04:30

hmm, It might be easier to just parse the results.txt and just have people supply the benchmark.txt for the startup variables/card specs :-)

hmm, mfacto showing 87792 GHz-days / day :)

Code:

got assignment: exp=10515217 bit_min=64 bit_max=65 (0.36 GHz-days)
Starting trial factoring M10515217 from 2^64 to 2^65 (0.36GHz-days)
Using GPU kernel "cl_barrett15_69_gs_2"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
May 30 22:58 |    0   0.1% |  0.124   1m59s |    257.90    82485    0.00%
May 30 22:58 |    8   0.2% |  0.120   1m55s |    266.50    82485    0.00%
May 30 22:58 |   12   0.3% |  0.118   1m53s |    271.01    82485    0.00%
May 30 22:58 |   15   0.4% |  0.118   1m53s |    271.01    82485    0.00%
May 30 22:58 |   20   0.5% |  0.117   1m52s |    273.33    82485    0.00%
M10515217 has a factor: 30604828569994966241
found 1 factor for M10515217 from 2^64 to 2^65 (partially tested) [mfakto 0.14-Win cl_barrett15_69_gs_2]
tf(): total time spent:  0.689s (44558.05 GHz-days / day)

got assignment: exp=10720453 bit_min=65 bit_max=66 (0.70 GHz-days)
Starting trial factoring M10720453 from 2^65 to 2^66 (0.70GHz-days)
Using GPU kernel "cl_barrett15_69_gs_2"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
May 30 22:58 |    0   0.1% |  0.227   3m38s |    276.37    82485    0.00%
May 30 22:58 |    3   0.2% |  0.222   3m33s |    282.59    82485    0.00%
May 30 22:58 |    8   0.3% |  0.223   3m33s |    281.32    82485    0.00%
M10720453 has a factor: 46314683458850504129
found 1 factor for M10720453 from 2^65 to 2^66 (partially tested) [mfakto 0.14-Win cl_barrett15_69_gs_2]
tf(): total time spent:  0.686s (87792.28 GHz-days / day)

Code:

[Fri May 30 22:58:51 2014]
M10515217 has a factor: 30604828569994966241 [TF:64:65*:mfakto 0.14-Win cl_barrett15_69_gs_2]
found 1 factor for M10515217 from 2^64 to 2^65 (partially tested) [mfakto 0.14-Win cl_barrett15_69_gs_2]
M10720453 has a factor: 46314683458850504129 [TF:65:66*:mfakto 0.14-Win cl_barrett15_69_gs_2]
found 1 factor for M10720453 from 2^65 to 2^66 (partially tested) [mfakto 0.14-Win cl_barrett15_69_gs_2]
M10030583 has a factor: 77685810496363030583 [TF:66:67*:mfakto 0.14-Win cl_barrett15_69_gs_2]
found 1 factor for M10030583 from 2^66 to 2^67 (partially tested) [mfakto 0.14-Win cl_barrett15_69_gs_2]
[Fri May 30 22:59:31 2014]

hmm, I guess the timestamps are not after every result in results.txt

axn · 2014-05-31, 06:58

Quote:

Originally Posted by chalsall

And for any who can bring additional fire-power to bear, it would be welcome, needed and appreciated.

We _really_ need that front page pimpage on GIMPS. George?

2014-05-31, 03:45	#2968
LaurV Romulan Interpreter Jun 2011 Thailand 3·3,221 Posts	The things are not related. I mean what you (kracker) say, and what I say. (I am still at the coffee ) The benchmark is good as it is. But I just complained (read: nitpicking) about the fact that class-0 has no reason to be excluded. There are plenty of factors in class 0 too: Code: gp> d=4620; p=7; while(1, p=nextprime(p+1); forstep(k=d,10^10,4620,q=2kp+1; if([0,1,0,0,0,0,0,1][q%8+1],if(Mod(2,q)^p==1, print(p","q), break))); printf("...%d...%c",p,13)) 11621,107378041 294403,2720283721 1797589,16609722361 1995317,18436729081 2233547,20637974281 2639761,24391391641 3361639,31061544361 3391831,31340518441 3395123,31370936521 4022371,37166708041 4785769,44220505561 4797073,44324954521 4902077,45295191481 5114959,47262221161 5735981,53000464441 5817389,53752674361 5863271,54176624041 6082619,56203399561 7651261,70697651641 7907231,73062814441 8181751,75599379241 ...8708663... * at top-level: d=4620;p=7;while(1,p=nextprime(p+1);f * ^-------------------- *** user interrupt after 45,481 ms. In the code above, "d" is the class and you can supply any number between 1 and 4619. Of course, for class-0, you have to supply 4620, otherwise all factors are 20p+1=1, and you may get some error - maybe that was James' issue when he queried the DB ? edit: Or does mfaktX test the class 0 at the end, same as in my example?? (I don't remember that, but I think is tested first). Last fiddled with by LaurV on 2014-05-31 at 03:51

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Status	Primeinator	Operation Billion Digits	5	2011-12-06 02:35
62 bit status	1997rj7	Lone Mersenne Hunters	27	2008-09-29 13:52
OBD Status	Uncwilly	Operation Billion Digits	22	2005-10-25 14:05
1-2M LLR status	paulunderwood	3*2^n-1 Search	2	2005-03-13 17:03
Status of 26.0M - 26.5M	1997rj7	Lone Mersenne Hunters	25	2004-06-18 16:46

2014-05-30, 23:25	#2962
TheMawn May 2013 East. Always East. 6BF₁₆ Posts	You have dozens of lines like Factor=10409843,79,80 That benchmark will take weeks to complete...

2014-05-31, 03:17	#2966
LaurV Romulan Interpreter Jun 2011 Thailand 3·3,221 Posts	Pure stupid question, I just woke up and didn't have my coffee yet, but I don't get it why the class-0 has to be excluded (?!) Forgive my brain fart if any.