New .dat time?

benjackson · 2007-10-31, 05:23

Looks like SoB knocked another one of their primes off:

http://www.free-dc.org/forum/showthread.php?t=12670

Citrix · 2007-10-31, 07:40

Nice news!

(I think it was PSP's turn to find a prime. I guess we get an extra turn)

thommy · 2007-10-31, 08:36

Is there a way sr2sieve can deleta a k sequence like sr5sieve does?

hhh · 2007-10-31, 08:45

Patience... please.

The last thing we want are human caopy and paste errors in change of a 3% speed increase for 2 days. The new dat will be out as soon as Lars has enough time to do it, and do it correctly. Until then, we should stick to the old one.

H.

ltd · 2007-10-31, 13:40

New combined dat file is online at the usual place.

http://www.psp-project.de/test/sievecomb.zip

monst · 2007-10-31, 14:26

I'm not sure if this matters or not. I just noticed that the first line has "18" on it. I thought it was supposed to be the number of k's in the file, which is 17.

6 SoB + 14 PSP - 3 duplicates = 17

ltd · 2007-10-31, 15:11

Thanks for finding this one.

If you use srsieve it is no problem but when using jjsieve or proth_sieve you will get an error.

I have corrected the file.

Brucifer · 2007-10-31, 17:07

heh... :)

The speed difference for me was like adding 3.5 more cores to the sieving effort! Very nice!

geoff · 2007-11-02, 00:14

The speedup from using the new combined SoB.dat with sr2sieve should be approx sqrt(18/17) or about 2.9% if the removed sequence was typical.

But the new .dat also results in the maximum hashtable density dropping below a threshold level on machines with 16Kb or 64Kb L1 caches, and so on those machines there could be a non-proportional change in speed (up or down) due to a smaller hashtable being used.

If anyone finds that the speed dropped with the new .dat, or that it didn't improve as much as expected, they could try overriding the hashtable size. Run with the -v switch to see what size hashtable was selected, then override it with the -H switch. E.g. -H32 will force a 32Kb hashtable to be used. Try half or double the default size. (Maximum is 64Kb).

Dead J. Dona · 2008-07-24, 11:37

I'm sorry, how much k will be in a current dat file.

In a welcomу thread says that we have only 14 k left.

Joe O · 2008-07-24, 11:47

Quote:

Originally Posted by Dead J. Dona

I'm sorry, how much k will be in a current dat file.

In a welcomу thread says that we have only 14 k left.

There are 16 k in the current dat file.

2007-10-31, 05:23	#1
benjackson Aug 2007 Princeton, NJ 10₁₀ Posts	New .dat time? (Answer: currently not) Looks like SoB knocked another one of their primes off: http://www.free-dc.org/forum/showthread.php?t=12670 Last fiddled with by hhh on 2008-07-25 at 18:33

2007-11-02, 00:14	#9
geoff Mar 2003 New Zealand 13·89 Posts	The speedup from using the new combined SoB.dat with sr2sieve should be approx sqrt(18/17) or about 2.9% if the removed sequence was typical. But the new .dat also results in the maximum hashtable density dropping below a threshold level on machines with 16Kb or 64Kb L1 caches, and so on those machines there could be a non-proportional change in speed (up or down) due to a smaller hashtable being used. If anyone finds that the speed dropped with the new .dat, or that it didn't improve as much as expected, they could try overriding the hashtable size. Run with the -v switch to see what size hashtable was selected, then override it with the -H switch. E.g. -H32 will force a 32Kb hashtable to be used. Try half or double the default size. (Maximum is 64Kb). Last fiddled with by geoff on 2007-11-02 at 00:16

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Time to End	davar55	Lounge	4	2013-02-23 02:40
P95 est time to go seems off	bcp19	Software	1	2012-08-03 22:39
How much time	Unregistered	Information & Answers	4	2008-12-20 21:00
Time	Xyzzy	Science & Technology	26	2008-01-19 03:28
P3 TF time	PrimeCruncher	Software	30	2003-12-21 05:26

2007-10-31, 07:40	#2
Citrix Jun 2003 2·7·113 Posts	Nice news! (I think it was PSP's turn to find a prime. I guess we get an extra turn)

2007-10-31, 08:36	#3
thommy Dec 2006 3×11 Posts	Is there a way sr2sieve can deleta a k sequence like sr5sieve does?

2007-10-31, 08:45	#4
hhh Jun 2005 373 Posts	Patience... please. The last thing we want are human caopy and paste errors in change of a 3% speed increase for 2 days. The new dat will be out as soon as Lars has enough time to do it, and do it correctly. Until then, we should stick to the old one. H.

2007-10-31, 13:40	#5
ltd Apr 2003 2²·193 Posts	New combined dat file is online at the usual place. http://www.psp-project.de/test/sievecomb.zip

2007-10-31, 14:26	#6
monst Mar 2007 179 Posts	I'm not sure if this matters or not. I just noticed that the first line has "18" on it. I thought it was supposed to be the number of k's in the file, which is 17. 6 SoB + 14 PSP - 3 duplicates = 17

2007-10-31, 15:11	#7
ltd Apr 2003 2²·193 Posts	Thanks for finding this one. If you use srsieve it is no problem but when using jjsieve or proth_sieve you will get an error. I have corrected the file.

2007-10-31, 17:07	#8
Brucifer Dec 2005 313 Posts	heh... :) The speed difference for me was like adding 3.5 more cores to the sieving effort! Very nice!

2008-07-24, 11:37	#10
Dead J. Dona Feb 2007 Kiev, Ukraine 3×7 Posts	I'm sorry, how much k will be in a current dat file. In a welcomу thread says that we have only 14 k left.