| Author |
Message |
|---|
John Honorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
                 
|
|
FYI...the PPS (Sieve) task has doubled from 1G to 2G's. Credit has been adjusted accordingly.
____________
|
|
|
|
|
|
PPS CUDA Sieving:
As expected runtime increased 2 x
Not expected cpu-time increased 2.4 x
Not an issue with so low cpu usage, but perhaps you can explain this.
____________
 |
|
|
 Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 0
                           
|
PPS CUDA Sieving:
As expected runtime increased 2 x
Not expected cpu-time increased 2.4 x
Not an issue with so low cpu usage, but perhaps you can explain this.
On my system, I'm not seeing that -- both elapsed and CPU times increased by just under 2.0. I'm definitely not seeing a disproportionate increase in CPU times.
Might be time to start comparing hardware:
Core2Quad @2.4GHz and GTX 280.
____________
My lucky number is 75898524288+1 |
|
|
VatoVolunteer tester
Send message
Joined: 2 Feb 08
Posts: 785
ID: 18447
Credit: 263,436,450
RAC: 0
              
|
|
has deadline dropped too? combined with the doubling of WU size, that's really crappy for CPU folks with part-time machines.
____________
 |
|
|
|
|
On my system, I'm not seeing that -- both elapsed and CPU times increased by just under 2.0. I'm definitely not seeing a disproportionate increase in CPU times.
Might be time to start comparing hardware:
Core2Quad @2.4GHz and GTX 280.
http://www.primegrid.com/show_host_detail.php?hostid=145391
The cpu should be less than 2 x, because relatively a lot of cpu is used by starting and stopping the process and that is done only once now ;) |
|
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
                
|
http://www.primegrid.com/show_host_detail.php?hostid=145391
The cpu should be less than 2 x, because relatively a lot of cpu is used by starting and stopping the process and that is done only once now ;)
http://www.primegrid.com/results.php?hostid=115189
@CP
Your gpu seems to be a little bit faster with win64...
[edit]
link to host
____________
Best wishes. Knowledge is power. by jjwhalen
 |
|
|
|
|
|
There are several kinds of a GT240 :)
Mine is GT240(GT215) 512MB GDDR5 40nm technology.
So it's hard to compare.
Even when the clockrates are different:
Core: 652 Memory:1700 Shader:1588
____________
|
|
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
|
|
GT240(GT215) with 1GB GDDR3 and all clockrates are lower.
Core: 550 Mem:850 Shader:1340
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
has deadline dropped too? combined with the doubling of WU size, that's really crappy for CPU folks with part-time machines.
Yes, at p=1300T, deadline dropped to 2 days. This is only temporary. We are approaching p=1405T and will stop there.
We will be switching from ppsieve to tpsieve and moving to the combined Proth and Riesel effort of 2M < n < 3M. It is currently in PST being prepared. In order to switch to tpsieve, all the tasks based on ppsieve must be completed and validated. Dropping the deadline to 2 days will help facilitate the "clean up."
We expect to be at p=1405T by the end of the week with the "clean up" complete sometime next week. Sorry for the inconvenience...hopefully it will be short lived. :)
____________
|
|
|
|
|
|
What happens if some Fermi GPU / OpenCL crunchers forget to update their app_info.xml files to use the tpsieve app? Will the validator check for the presence of factors of k*2^n-1 numbers? It would be not nice to end up with hundreds of (small) ranges that were not sieved on the Riesel side.
____________
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
What happens if some Fermi GPU / OpenCL crunchers forget to update their app_info.xml files to use the tpsieve app? Will the validator check for the presence of factors of k*2^n-1 numbers? It would be not nice to end up with hundreds of (small) ranges that were not sieved on the Riesel side.
A solution is in place. :) Thanks for bringing the issue up.
____________
|
|
|
VatoVolunteer tester
Send message
Joined: 2 Feb 08
Posts: 785
ID: 18447
Credit: 263,436,450
RAC: 0
|
Yes, at p=1300T, deadline dropped to 2 days. This is only temporary.
That makes sense - no need for me to be grouchy! :-) Thanks John.
____________
|
|
|
|
|
We will be switching from ppsieve to tpsieve and moving to the combined Proth and Riesel effort of 2M < n < 3M.
This sounds like that there is no need for two subprojects.
What does this actually mean in terms of the 'life expectancy' of Proth and Riesel sieve subprojects?
Is there still going be time to hit a few milestones?
|
|
|
Ken_g6Volunteer developer
Send message
Joined: 4 Jul 06
Posts: 915
ID: 3110
Credit: 183,164,814
RAC: 0
          
|
|
If you're worried about "TRP Sieve", don't be. That's a completely separate project sieving a few K's deeply.
PPS Sieve will sieve numbers of the Proth and Riesel formats, in a different range of K's (a smaller range but a larger set), for the PPS LLR project and the No Prime Left Behind project on Mersenneforum. From John's plan, it looks like PPS Sieve will continue for a good long while too.
____________
|
|
|
|
|
|
Question about sieving in general.
In order to sieve, either p remains constant or the PPS remains constant. It's clear multiple PPSes can be sieved out in one sieve workunit (~200, last time I participated), so I would guess there is one or more p per sieve workunit. The question, then, is, what range of n does the sieve filter?
Or, y'know, if I was wrong well before the last sentence, feel free to backtrack. :) |
|
|
|
|
Question about sieving in general.
In order to sieve, either p remains constant or the PPS remains constant. It's clear multiple PPSes can be sieved out in one sieve workunit (~200, last time I participated), so I would guess there is one or more p per sieve workunit. The question, then, is, what range of n does the sieve filter?
Or, y'know, if I was wrong well before the last sentence, feel free to backtrack. :)
Times have changed a little bit. We are doing a tiny bit of more work these days :)
Neither p nor k is constant. With BOINC we sieve p ranges of 2G(=1 WU, the length was just increased from 1G) of ks from 5 to 9999 and N goes up to 2M. PrimeGrid's PSA participants currently work on n=2M-3M. For example: I've just uploaded k 5-9999, n 2-3M, p 115T-116T over at PST. Only 288,537 factors for this 1T range :)
By the way: Over at PST - Project Staging's sieving department :) - we already sieve simultaneously the +1 and -1 forms (Proth and Riesel) and BOINC is about to switch these days (see the posts above).
---
1M = 10^6, 1G = 10^9, 1T=10^12
____________
|
|
|
