Small(ish) matrices on fast computers

fivemack · 2011-03-14, 11:13

I ran an easy (C131) problem on the new dual-Xeon-5660 machine at work, with -t 3 through -t 12:

Hyperthreading enabled, though there are 12 'real' threads and I would hope I could trust Linux (Ubuntu 10.04) to use real threads before doubling up.

Code:

Number of threads : reported BLanczosTime in seconds
3  3070
4  2898
5  2346
6  2392
7  2212
8  2358
9  2324
10 2330
11 2270
12 2436

Batalov · 2011-03-14, 23:43

The Xeon X56x0's (I've had a brief access to one or two of these, both dual and a single X5680) throw a handful of sand into msieve's inner workings by self-reporting their cache sizes strangely - which in turn affects the chosen block size for LA. I've seen that when they ran with block size manually set to 32768 (and compiled without -DLARGEBLOCK), they seemed to scale better.

2011-03-14, 11:13	#1
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2×7×461 Posts	Small(ish) matrices on fast computers I ran an easy (C131) problem on the new dual-Xeon-5660 machine at work, with -t 3 through -t 12: Hyperthreading enabled, though there are 12 'real' threads and I would hope I could trust Linux (Ubuntu 10.04) to use real threads before doubling up. Code: Number of threads : reported BLanczosTime in seconds 3 3070 4 2898 5 2346 6 2392 7 2212 8 2358 9 2324 10 2330 11 2270 12 2436 Attached Thumbnails Last fiddled with by fivemack on 2011-03-14 at 23:49

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Do normal adults give themselves an allowance? (...to fast or not to fast - there is no question!)	jasong	jasong	35	2016-12-11 00:57
Plugging Matrices into Functions	Joshua2	Homework Help	50	2009-11-17 02:42
Count of Matrices	davar55	Puzzles	2	2007-06-13 20:46
matrices question	fuzzy	Miscellaneous Math	1	2005-03-19 11:12
Fast calculations modulo small mersenne primes like M61	Dresdenboy	Programming	10	2004-02-29 17:27

2011-03-14, 23:43	#2
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 19·23² Posts	The Xeon X56x0's (I've had a brief access to one or two of these, both dual and a single X5680) throw a handful of sand into msieve's inner workings by self-reporting their cache sizes strangely - which in turn affects the chosen block size for LA. I've seen that when they ran with block size manually set to 32768 (and compiled without -DLARGEBLOCK), they seemed to scale better.