Linear algebra reservations, progress and results

[RichD]

I'll take 8447_61_minus1 next when sieving completes.

[YuL]

GW_5_349 is factored.

Wed Apr 22 06:45:31 2015  p81 factor: 237562004704624448311927204157228513514925488183767855024614900967198016835145381
Wed Apr 22 06:45:31 2015  p158 factor: 43659826426781109186821952208471499900308562367193670976127593637886811117325706924723430319107505928901799076884346915471971428254916449628410689467775783303

LA took ~86.5h on Dual Xeon E5-2620, td=116. Log here or as attached file.

[YuL]

I can deal with GW_6_317 next.

[debrouxl]

8111_61_minus1 ends in two large factors:

p116 = 60435115091268432419921810433567521709399973026268300309736782518186865465572378911253558495130382497401902094470611
p119 = 57973334096353723857681299801353322247365987257208968796859922108859487196394462843274636706184959884252826374430352251

~9.7M matrix, 91h elapsed time on Xeon E3-1240 V2 @ 3.6 GHz, the computer was far from being otherwise idle.

[fivemack]

Originally Posted by YuL

GW_5_349 is factored.

Wed Apr 22 06:45:31 2015  p81 factor: 237562004704624448311927204157228513514925488183767855024614900967198016835145381
Wed Apr 22 06:45:31 2015  p158 factor: 43659826426781109186821952208471499900308562367193670976127593637886811117325706924723430319107505928901799076884346915471971428254916449628410689467775783303

LA took ~86.5h on Dual Xeon E5-2620, td=116. Log here or as attached file.

I am curious that you're running eight threads per MPU job on a machine with six cores per processor; is -t 8 running more quickly than -t 6?

[YuL]

Originally Posted by fivemack

I am curious that you're running eight threads per MPU job on a machine with six cores per processor; is -t 8 running more quickly than -t 6?

I don't really know, -t 6 may well be faster. What is clear is that two -t 8 processes
is way faster than one -t 8 (can't give you the numbers, I don't have access to
the logs I'm not at home). I'll try to (partially) rerun LA with -t 6 to see how to goes.

[fivemack]

Given the shape of the memory system on a dual-Xeon, I would certainly expect two jobs to be quicker than one, provided some combination of the OS and your command-line arguments have arranged the memory for each job to be allocated on its socket rather than across QPI.

I would expect to be able to run taskset -c 0-5 numactl -l msieve -nc2 -t 6 and taskset -c 6-11 numactl -l msieve -nc2 -t 6 on two separate jobs without either being significantly slowed down by the other; don't know whether that is more efficient in the longer term. Will have to wait until the 48-core-Opteron dies before I can justify getting my own dual-Xeon to experiment with :)

[fivemack]

Fri Apr 24 05:29:32 2015  prp105 factor: 955692549334371243078921268353865183554241021336069816183031517214259811008037954393555319325774360841289
Fri Apr 24 05:29:32 2015  prp131 factor: 18832132445422467167516526556671851593866412901247899936420027484723903995356687614204911038001552861944332446424568114763187513817

247 hours for 15.0M density-100 matrix on i7/2600 -t3

Taking GW_8_271 (13.9M density-110 matrix; ETA 5 May)

[VictordeHolland]

F1929

Fri Apr 24 13:47:21 2015  prp82 factor: 2205948162915930188728253758406230768089063035987047653868064633322223630864596833
Fri Apr 24 13:47:21 2015  prp100 factor: 4669428221120836052207901829071262580709954000445036139624291939756122352739143892182753698617571373

255,158,127 (raw) relations
188,124,378 unique relations
Target_density=125
matrix is 16,110,488 x 16,110,713 (7497.8 MB)
209 hours 4 cores 3770k

[henryzz]

Originally Posted by xilman

15.1M matrix with td=110 (120 didn't build). ETA 320 hours on an AMD Phenom(tm) II X6 1090T using 6 threads and 6.9G of 8G available RAM.

It only just fits and I'm going to have to purchase a bigger system for running jobs any bigger than this.

I would suggest either waiting for skylake or getting Haswell-E. If you go for DDR4 now then you will have much more expansion room in the future. This is especially true as it seems you hold onto pcs for a while. I suffered by buying a DDR2 pc just after DDR3 came out. Upgrading was a pain. If I had 8-16GB now my pc would run much faster. I am stuck with 4GB now though.

[xilman]

Originally Posted by henryzz

I would suggest either waiting for skylake or getting Haswell-E. If you go for DDR4 now then you will have much more expansion room in the future. This is especially true as it seems you hold onto pcs for a while. I suffered by buying a DDR2 pc just after DDR3 came out. Upgrading was a pain. If I had 8-16GB now my pc would run much faster. I am stuck with 4GB now though.

Thanks. My general approiach is to buy new PCs every 5 years or so and run them into the ground. I tend to go for something somewhat behind the bleeding edge, but not by much, to achieve a sensible price/performance ratio. Second-hand PCs are an entirely different matter; there price/performance is much more critical.

Power consumption is an issue, but not a big one, as PCs are effective fan-heaters.