New Application Coming Soon
log in

Advanced search

Message boards : News : New Application Coming Soon

1 · 2 · Next
Author Message
Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 11192 - Posted: 28 Jan 2011, 22:40:26 UTC

In a few days, there will be a second app option named "mini_collatz" which uses the same apps but checks a much smaller range of numbers per workunit. Those with fast GPUs should unselect mini_collatz in their project preferences as they will likely run out of work each day if running the shorter WUs.
Those choosing to disable the longer Collatz WUs should make sure to check the "If no work for selected applications is available, accept work from other applications?" option so they will continue to get until prior to the new app being enabled.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 11193 - Posted: 28 Jan 2011, 23:03:13 UTC

The daily high steps page won't work correctly until it is upgraded to match the server changes made today which are needed to support splitting the app into large and small versions. I hope to get that working again this weekend.

Profile Crunch3r
Volunteer moderator
Project developer
Project tester
Avatar
Send message
Joined: 30 Jun 09
Posts: 219
Credit: 7,515,663,568
RAC: 8,898
Message 11195 - Posted: 28 Jan 2011, 23:50:08 UTC - in response to Message 11192.

In a few days, there will be a second app option named "mini_collatz" which uses the same apps but checks a much smaller range of numbers per workunit. Those with fast GPUs should unselect mini_collatz in their project preferences as they will likely run out of work each day if running the shorter WUs.
Those choosing to disable the longer Collatz WUs should make sure to check the "If no work for selected applications is available, accept work from other applications?" option so they will continue to get until prior to the new app being enabled.


So a new plan class ... no "automatic" detection of gpu features ?

____________

Team BOINC United.Join Science that matters.

Profile SLAYER OF DEATH
Avatar
Send message
Joined: 12 Dec 10
Posts: 134
Credit: 17,993,517
RAC: 0
Message 11197 - Posted: 29 Jan 2011, 7:41:23 UTC
Last modified: 29 Jan 2011, 7:42:30 UTC

Hello all...it seams that the mini_c, in my case, are going to take 2 hours on an hd 4670 and gts 250 versus the regular WU taking just as long, fixing to get some hd6870 if not the hd69xx. Not commplaining, well as long as it is still the same credit..lol. On a side note what differentiats being a wing man versus doing the original WU? Any differance anyway?
And thx Slicker for your time and work.
____________

Proud Member of Team Carl Sagan.

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3246
Credit: 1,702,269,192
RAC: 4,749,050
Message 11199 - Posted: 29 Jan 2011, 13:00:33 UTC - in response to Message 11197.

Hello all...it seams that the mini_c, in my case, are going to take 2 hours on an hd 4670 and gts 250 versus the regular WU taking just as long, fixing to get some hd6870 if not the hd69xx. Not commplaining, well as long as it is still the same credit..lol. On a side note what differentiats being a wing man versus doing the original WU? Any differance anyway?
And thx Slicker for your time and work.


I got a 6850 and it is not nearly as fast as its cost would indicate it should be!! I have a 5770, 4 actually, and they are getting twice the credits!

Profile SLAYER OF DEATH
Avatar
Send message
Joined: 12 Dec 10
Posts: 134
Credit: 17,993,517
RAC: 0
Message 11202 - Posted: 29 Jan 2011, 15:14:00 UTC - in response to Message 11199.

Huu..I thought that more gpu cores wud crunch faster, thats the only reasoning for the 68xx. My initial wanton was the 5770,5780? cuz of the more core count. Do you run your 68xx on other projects or just Collatz? At some time I wud like to get back into MY@H needing a DP gfx, also DP cards really do not work to good on SP projects from my understanding, I cud be wrong. And I am itchy to build an Intel i7 950 rig with at least 2 pci_e's, x64 bit of course.
____________

Proud Member of Team Carl Sagan.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 11204 - Posted: 29 Jan 2011, 15:41:25 UTC - in response to Message 11195.

So a new plan class ... no "automatic" detection of gpu features ?


The ATI cards might be able to be done by using the id, but keeping up with them either requires fixing the BOINC code so that the GPU info is data driven in a table verses hard-coded as it is today, or doing a build every month.

cc > 1.2 might work, but aren't all the newer nVidia GPUs cc 1.3+ even the low end ones?

Profile [AF>EDLS] Polynesia
Avatar
Send message
Joined: 2 Sep 09
Posts: 13
Credit: 13,018,423
RAC: 0
Message 11205 - Posted: 29 Jan 2011, 16:07:46 UTC

I did not understand the difference in duration between these two applications ....!

2h? So the credit should be a little more than 4 times ... say 4.5 at least to reward those who calculate the longer units ....
____________

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 11206 - Posted: 29 Jan 2011, 17:04:07 UTC - in response to Message 11202.

Huu..I thought that more gpu cores wud crunch faster, thats the only reasoning for the 68xx. My initial wanton was the 5770,5780? cuz of the more core count. Do you run your 68xx on other projects or just Collatz? At some time I wud like to get back into MY@H needing a DP gfx, also DP cards really do not work to good on SP projects from my understanding, I cud be wrong. And I am itchy to build an Intel i7 950 rig with at least 2 pci_e's, x64 bit of course.


68xx cards will also run on DNETC. They will not run on MW.

More cores = faster, but only if memory is the same speed and the cores run at the same speed. The die size can also affect the speed. 100 cores @ 100 Mhz w/ 50nm transistors will be slower than 100 cores @ 100 Mhz w/ 40nm transistors. Collatz uses texture memory a lot, so increased memory can have more of an affect here than on other GPU projects, but only if memory is the limiting factor on the GPU.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 11208 - Posted: 29 Jan 2011, 17:45:58 UTC - in response to Message 11205.

I did not understand the difference in duration between these two applications ....!

2h? So the credit should be a little more than 4 times ... say 4.5 at least to reward those who calculate the longer units ....


The work generator controls the size of the WU. There is a separate work generator for each app. I had to change both so that they now use a stored proc to lock the record, increment the next starting number, and then return since using a single file would no longer work.

The validators know from the WU size what the max credits should be should tow boxes happen to return garbage that matches which exceeds the expected total steps but isn't high enough to trigger an overflow error.

The assimilators are just duplicates of each other, although they were changed to be now use the database to store results rather than a per-day result file since I couldn't have both writing to the same file at the same time.

I might change the validator to grant a little more credit to the large ones to encourage people to run those instead of the mini ones (either that or reduce the credit a little for the smaller ones). I'm sure people would complain that they are being penalized for not having newer faster equipment that can run the longer ones. But you can't please all of the people all of the time...

The other idea I've had is to send out the mini ones to start and then, if the box is completing a WU in under an hour, switching the host to only use the large ones. It would have to take into account averages for CPUs vs GPUs in order to account for those with boxes crunching both CPU and GPU Wus where the box has a fast GPU and a slow CPU. For example, Kevint used to have a bunch of machines where he bought the cheapest MB/CPU her could find just so he could put a GPU inside.

Profile sir sant
Send message
Joined: 17 Nov 10
Posts: 40
Credit: 2,733,063,385
RAC: 6,190,076
Message 11210 - Posted: 29 Jan 2011, 20:01:55 UTC - in response to Message 11202.

Huu..I thought that more gpu cores wud crunch faster, thats the only reasoning for the 68xx. My initial wanton was the 5770,5780? cuz of the more core count. Do you run your 68xx on other projects or just Collatz? At some time I wud like to get back into MY@H needing a DP gfx, also DP cards really do not work to good on SP projects from my understanding, I cud be wrong. And I am itchy to build an Intel i7 950 rig with at least 2 pci_e's, x64 bit of course.

I have an 6870 cf and running it 24\7 gives about 330000 credit in 24 hours. Would recommend for you to go get some old 6950 2gbt ones and unlock them to 6970-s, a pair on unlocked and clocked 6950-s would rock imo.

Profile Beyond
Avatar
Send message
Joined: 30 Jul 09
Posts: 213
Credit: 939,653,252
RAC: 2,671
Message 11220 - Posted: 30 Jan 2011, 14:19:11 UTC - in response to Message 11208.

I might change the validator to grant a little more credit to the large ones to encourage people to run those instead of the mini ones

One vote for this strategy.

The other idea I've had is to send out the mini ones to start and then, if the box is completing a WU in under an hour, switching the host to only use the large ones.

One vote against this method. Please let us choose the WU size according to our needs for the particular host.

Thanks for initiating this dual WU size strategy. It sure shows a concern for the needs of the people running your project. You're the best.

Profile sir sant
Send message
Joined: 17 Nov 10
Posts: 40
Credit: 2,733,063,385
RAC: 6,190,076
Message 11224 - Posted: 30 Jan 2011, 17:05:10 UTC - in response to Message 11208.

I did not understand the difference in duration between these two applications ....!

2h? So the credit should be a little more than 4 times ... say 4.5 at least to reward those who calculate the longer units ....


I might change the validator to grant a little more credit to the large ones to encourage people to run those instead of the mini ones (either that or reduce the credit a little for the smaller ones). I'm sure people would complain that they are being penalized for not having newer faster equipment that can run the longer ones. But you can't please all of the people all of the time...


Yes, more credit please, switching to long wu-s now. Please dont reduce credit for short wu-s, that will discriminate people, and they will be less inclined to run the project with weaker gpu-s. Or keep the credit system the same, doesn't bother me, as long as there will not be credit penaltys for running short wu-s (which i wont be running anyway).

Clod Patry
Send message
Joined: 4 Jan 10
Posts: 4
Credit: 55,980,673
RAC: 0
Message 11225 - Posted: 30 Jan 2011, 17:50:26 UTC - in response to Message 11208.
Last modified: 30 Jan 2011, 17:51:35 UTC


I might change the validator to grant a little more credit to the large ones to encourage people to run those instead of the mini ones (either that or reduce the credit a little for the smaller ones). I'm sure people would complain that they are being penalized for not having newer faster equipment that can run the longer ones. But you can't please all of the people all of the time...


+1 for me.

I've a 5970 dedicated to BOINC, that could be a good reason to swich that machine from DNETC to collatz.

When running 24/7, at the end, the result between project should be similar for the same hardware.

terencewee*
Send message
Joined: 20 Dec 10
Posts: 12
Credit: 39,717,409
RAC: 0
Message 11226 - Posted: 30 Jan 2011, 21:04:09 UTC - in response to Message 11208.

I might change the validator to grant a little more credit to the large ones to encourage people to run those instead of the mini ones (either that or reduce the credit a little for the smaller ones).


Yes, please. DNetC ratio vs. time = ~2.61, while Collatz is ~1.95
I prefer Collatz due to GPU being cooler by 3 Celsius. :)

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3246
Credit: 1,702,269,192
RAC: 4,749,050
Message 11229 - Posted: 31 Jan 2011, 13:23:33 UTC - in response to Message 11208.

I might change the validator to grant a little more credit to the large ones to encourage people to run those instead of the mini ones (either that or reduce the credit a little for the smaller ones). I'm sure people would complain that they are being penalized for not having newer faster equipment that can run the longer ones. But you can't please all of the people all of the time...

The other idea I've had is to send out the mini ones to start and then, if the box is completing a WU in under an hour, switching the host to only use the large ones. It would have to take into account averages for CPUs vs GPUs in order to account for those with boxes crunching both CPU and GPU Wus where the box has a fast GPU and a slow CPU. For example, Kevint used to have a bunch of machines where he bought the cheapest MB/CPU her could find just so he could put a GPU inside.


You could also make it so the smaller ones are restricted in the total number of units any one pc can get per day, so if someone were to chose the smaller units they would only get say 10 units per 24 hours where as if they chose the bigger units they could get upto 100 units per 24 hours, assuming they could od that many. The numbers I just used are made up of course but if you restricted the total number of small units allowed per day to something a single core, or real slow dual core, pc could do, then everyone else would switch to the longer, bigger units.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 11235 - Posted: 31 Jan 2011, 19:57:23 UTC - in response to Message 11226.

I might change the validator to grant a little more credit to the large ones to encourage people to run those instead of the mini ones (either that or reduce the credit a little for the smaller ones).


Yes, please. DNetC ratio vs. time = ~2.61, while Collatz is ~1.95
I prefer Collatz due to GPU being cooler by 3 Celsius. :)


Collatz credit is based on a 100 credit per day machine earning 100 credits per day with the stock app running 32-bit Windows. DNETC credit is based upon.... well, they just picked a number bigger than Collatz but less than MW (see http://www.dnetc.net/forum_thread.php?id=25&nowrap=true#108). If I raise Collatz credit randomly, they will just do the same as their credit system is based upon being higher than Collatz.

Profile Beyond
Avatar
Send message
Joined: 30 Jul 09
Posts: 213
Credit: 939,653,252
RAC: 2,671
Message 11236 - Posted: 31 Jan 2011, 21:49:39 UTC - in response to Message 11235.

DNETC credit is based upon.... well, they just picked a number bigger than Collatz but less than MW (see http://www.dnetc.net/forum_thread.php?id=25&nowrap=true#108). If I raise Collatz credit randomly, they will just do the same as their credit system is based upon being higher than Collatz.

That's what they claimed but in reality the credits DNETC gives out are way higher than MW at least for HD 58xx cards (and DNETC isn't DP). For their new project (AndrOINC) it looks like they just picked a random number between DNETC and infinity ;)

Gary Craig
Send message
Joined: 30 Jan 10
Posts: 6
Credit: 3,626,547
RAC: 0
Message 11245 - Posted: 1 Feb 2011, 10:57:44 UTC

Well, I'll probably keep running the larger units even though they take 13+ hours on my Mac (which really isn't that old, maybe 16 months or so). I guess I don't care if I get X credits in N hours or X/d credits in N/d hours.

My other box could chew up the larger units really fast (GTX 570) but it is running Ubuntu 10.10 and I didn't see an obvious way to get Collatz to run on that via the normal BOINC mechanisms. I think I saw a post somewhere (can't find it now) about a manual install which I would be willing to do.

Sad to read in prior posts that the credit system is so chaotic from one project to the next. I read on wikipedia that the cobblestone measure was set up relative to a couple well-known benchmarks; would have thought that would have encouraged some consistency, but I guess not. Sorry, wandering off-topic into a quagmire, I'm sure :-)

--Gary

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3246
Credit: 1,702,269,192
RAC: 4,749,050
Message 11246 - Posted: 1 Feb 2011, 12:12:14 UTC - in response to Message 11245.

Well, I'll probably keep running the larger units even though they take 13+ hours on my Mac (which really isn't that old, maybe 16 months or so). I guess I don't care if I get X credits in N hours or X/d credits in N/d hours.

My other box could chew up the larger units really fast (GTX 570) but it is running Ubuntu 10.10 and I didn't see an obvious way to get Collatz to run on that via the normal BOINC mechanisms. I think I saw a post somewhere (can't find it now) about a manual install which I would be willing to do.

Sad to read in prior posts that the credit system is so chaotic from one project to the next. I read on wikipedia that the cobblestone measure was set up relative to a couple well-known benchmarks; would have thought that would have encouraged some consistency, but I guess not. Sorry, wandering off-topic into a quagmire, I'm sure :-)

--Gary


There are ways to make the gpu work just fine under Linux but you are correct it must be done manually, it is not automatic like in Windows. Here is something that might help you "There isn't a Stock Linux Cuda app that automatically gets dished out by the project,
you have to download and install the Linux Cuda Power app yourself".
Does that makes sense to you? I am not a Linux guy so am not able to provide more help but I do know that libraries, user permissions and not having the 32 bit files even on a 64 bit system are a common reason for asking for help.

As for credits or cobblestones that is a subject that has been a problem since day one.

1 · 2 · Next
Post to thread

Message boards : News : New Application Coming Soon


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.