help designing prime95 28.10 feature - Page 4

vsuite · 2016-05-11, 04:02

As a separate feature, how about an estimate of GHz-Days/Day each 10000 iterations or whatever?

CuriousKit · 2016-05-12, 01:28

I don't know how practical it is, but it might be best to do some kind of FFT test run during start-up, before anything else starts, so it is interrupted only minimally. You know those Windows applications that run in text mode before the log-on screen even appears, for example... such as CHKDSK during work on the system partition. Not sure how straightforward that is though, and it may require administrator rights.

I wonder... would it be practical at all to run an LL test for 2^1,257,787 - 1 for each FFT algorithm available, or something similarly small but which has a known result and won't take forever (or too short a time that it's not an accurate representation of a typical task)? It will also spot bugs with an implementation or CPU at the same time (not that there should be any at this stage).

P.S. 28.9 seems to be faster for me generally, and seems to use fewer CPU cycles as well (something to do with hyperthreading, I think).

vsuite · 2016-05-12, 07:02

The rationale is

ms/iteration is calculated for each 10000. However it is only easily converted to GHz-days/day by the program which will know the estimated GHz-days for that exponent (range).

ATH · 2016-05-12, 11:59

Quote:

Originally Posted by vsuite

The rationale is

ms/iteration is calculated for each 10000. However it is only easily converted to GHz-days/day by the program which will know the estimated GHz-days for that exponent (range).

ms/iteration is calculated as often as it is set in the options.

TObject · 2016-05-12, 17:02

Quote:

Originally Posted by CuriousKit

I don't know how practical it is, but it might be best to do some kind of FFT test run during start-up, before anything else starts, so it is interrupted only minimally. You know those Windows applications that run in text mode before the log-on screen even appears, for example... such as CHKDSK during work on the system partition. Not sure how straightforward that is though, and it may require administrator rights.

Doable and already done. This is what the Windows 10/8/7/Vista/XP/2008 64-bit Service and Windows 10/8/7/Vista/XP/2008/2003/2000/NT 32-bit Service versions of Prime95 are for. They run regardless of whether a user is logged in or not.

May not be practical to run tests at the boot time though, as a lot goes on during computer start up.

vsuite · 2016-05-13, 06:37

I didn't want to mention the software to avoid seeming like comparing them.

Mfaktc for example shows the GHz-days/day rate every screen write, which is every customizable number of 'iterations' or whatever.

Could Prime95 calculate the at the moment Ghz-days/day rate every screen write as well?
The current GHz-days/day can be calculated from something like this [I hope the equation is correct].
. (Total GHz-days for exponent / total iterations for exponent / (60 x 60 x 24) [constant]) x # of iterations performed / time taken (in seconds)

It won't increase the speed of the LL test. It won't encourage more people to join GIMPS. Yet some people might appreciate the addition of that feature.

2016-05-12, 07:02	#36
vsuite Jan 2010 2·3·19 Posts	The rationale is ms/iteration is calculated for each 10000. However it is only easily converted to GHz-days/day by the program which will know the estimated GHz-days for that exponent (range). Last fiddled with by vsuite on 2016-05-12 at 07:03

2016-05-13, 06:37	#39
vsuite Jan 2010 72₁₆ Posts	I didn't want to mention the software to avoid seeming like comparing them. Mfaktc for example shows the GHz-days/day rate every screen write, which is every customizable number of 'iterations' or whatever. Could Prime95 calculate the at the moment Ghz-days/day rate every screen write as well? The current GHz-days/day can be calculated from something like this [I hope the equation is correct]. . (Total GHz-days for exponent / total iterations for exponent / (60 x 60 x 24) [constant]) x # of iterations performed / time taken (in seconds) It won't increase the speed of the LL test. It won't encourage more people to join GIMPS. Yet some people might appreciate the addition of that feature. Last fiddled with by vsuite on 2016-05-13 at 06:52

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Prime95 - stop all workers on error [feature request]	kql	Software	1	2020-12-31 15:15
New Feature!	Xyzzy	Lounge	0	2017-01-07 22:52
Feature request: Prime95 priority higher than 10	JuanTutors	Software	19	2006-10-29 04:09
Prime95 Version 24.13 "Feature"	RMAC9.5	Software	2	2006-03-24 21:12
Designing a home system for CNT.	xilman	Hardware	6	2004-10-21 19:41

2016-05-11, 04:02	#34
vsuite Jan 2010 114₁₀ Posts	As a separate feature, how about an estimate of GHz-Days/Day each 10000 iterations or whatever?

2016-05-12, 01:28	#35
CuriousKit "J. Gareth Moreton" Feb 2015 Nomadic 2·3²·5 Posts	I don't know how practical it is, but it might be best to do some kind of FFT test run during start-up, before anything else starts, so it is interrupted only minimally. You know those Windows applications that run in text mode before the log-on screen even appears, for example... such as CHKDSK during work on the system partition. Not sure how straightforward that is though, and it may require administrator rights. I wonder... would it be practical at all to run an LL test for 2^1,257,787 - 1 for each FFT algorithm available, or something similarly small but which has a known result and won't take forever (or too short a time that it's not an accurate representation of a typical task)? It will also spot bugs with an implementation or CPU at the same time (not that there should be any at this stage). P.S. 28.9 seems to be faster for me generally, and seems to use fewer CPU cycles as well (something to do with hyperthreading, I think).