| View previous topic :: View next topic |
| Author |
Message |
StarGal
Knight
Joined: 14 Jun 2012
Posts: 37
Location: Lunar System
|
 Posted: Sat Jun 16, 2012 10:26 am Post subject: MilkyWay@home |
 |
|
Any feedback welcome.
 |
|
| Back to top |
|
 |
StarGal
Knight
Joined: 14 Jun 2012
Posts: 37
Location: Lunar System
|
| Posted: Sat Jun 16, 2012 11:25 am Post subject: |
|
|
Maybe I need to be specific?
How does it run?
What does it do? Aside from its goal. |
|
| Back to top |
|
|
Concrete-mixing Moose
Prince
Joined: 30 Apr 2012
Posts: 567
Location: The Joyce Grenfell Home for the Distressed
|
| Posted: Sat Jun 16, 2012 12:44 pm Post subject: |
|
|
Never run it myself, but if you're interested in different distributed computing projects have a look here http://www.distributedcomputing.info/projects.html
There's a lot to choose from and note that not all use Boinc.
_________________
Save my home - click every day on the picture!
 |
|
| Back to top |
|
|
StarGal
Knight
Joined: 14 Jun 2012
Posts: 37
Location: Lunar System
|
| Posted: Sat Jun 16, 2012 2:56 pm Post subject: |
|
|
Great site! Thanks.
Still would like to know how MW@H uses our computers. |
|
| Back to top |
|
|
dow
Knight
Joined: 10 Jan 2011
Posts: 56
Location: Under the stairs
|
| Posted: Sat Jun 16, 2012 3:22 pm Post subject: |
|
|
| Lots of really hard sums, and utilising both the cpu (slow, and will use muliple cores, if available, or nvidia / double precision only amd /ati gpu). |
|
| Back to top |
|
|
StarGal
Knight
Joined: 14 Jun 2012
Posts: 37
Location: Lunar System
|
| Posted: Sat Jun 16, 2012 3:26 pm Post subject: |
|
|
OK
in other words
analyzing data like S@H?  |
|
| Back to top |
|
|
PhastPhred
Prince
Joined: 22 Mar 2006
Posts: 6017
Location: Northwest AR (USA)
|
| Posted: Sat Jun 16, 2012 8:50 pm Post subject: |
|
|
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=317#3521
| Quote: | The paper describes in rigorous detail (and in mathematical detail) the method and models in which we do all of our calculations: a very brief summary is that we use a very small volume to study spatial distributions of stars. This allows us to model the stream debris as a cylinder within some background distribution which are described by a number of parameters. We then use a technique called maximum likelihood to "search" the likelihood "space" and find the values of the parameters that have the highest likelihood. These correspond to the "correct" values for those parameters.
Using this method we were able to reproduce the parameters for a simulated data set, and closely reproduce earlier results on the stripe 82 data that we used. However, we were able to greatly reduce the errors regarding these values and determine new quantities as well. All total we were able to present values for the position, orientation, and size (spatial and number) of the stream and the values that best define the background distribution.
|
You can view/download the Cornell University paper here...
http://arxiv.org/abs/0805.2121
Not exactly on-point, but close, I am sure, albeit a bit dated...
_________________
 |
|
| Back to top |
|
|
Killerrabbit
Major Oblivion
Joined: 23 May 2002
Posts: 4656
Location: in a rabbit hole near you!!
|
| Posted: Sat Jun 16, 2012 11:48 pm Post subject: |
|
|
It is not about how to eat Milkyway candy bats which I thought at first.
Ni
_________________
 |
|
| Back to top |
|
|
Pooh Bear 27
Prince
Joined: 28 Jan 2005
Posts: 1358
Location: Fond du Lac, WI
|
| Posted: Wed Jun 27, 2012 1:28 pm Post subject: |
|
|
OK, got my new shurbber online. NVidia GT 640 card is running MW but so far has done a couple units and is inconclusive on them. I am on driver 301.42.
Any suggestions? I put that it can use a full CPU with the GPU. Any other settings I should check?
[edit]
Well, first one that got a wingperson validated. So I know I can product good units. Maybe they are just double checking me until my card is proven? Even though the quorum is 1 on them all. I'll continue on for a bit.
[edit 2]
It does look like they do adaptive replication. New hosts have to be validated against a few times before they are deemed clean. Just had my first one validate with no wingperson needed. |
|
| Back to top |
|
|
LanDroid
Prince
Joined: 11 Jun 2002
Posts: 4476
Location: Cincinnati, OH U.S.
|
| Posted: Wed Jun 27, 2012 8:39 pm Post subject: |
|
|
| Quote: | | It does look like they do adaptive replication. New hosts have to be validated against a few times before they are deemed clean. Just had my first one validate with no wingperson needed. |
This leads me to an off-topic beef with Seti and some other projects that seem to double, triple, or quad-validate ALL work units. Is this still going on? I've always thought the "adaptive validation" described by Pooh Bear 27 would be MUCH more efficient, allowing those projects to chainsaw through huge amounts of data in much less time, uncovering the Wisdom in the raw data that will launch humanity Beyond InfiNIty!
   |
|
| Back to top |
|
|
stooper101
Prince
Joined: 16 May 2009
Posts: 1248
Location: Cheese Country
|
| Posted: Thu Jun 28, 2012 12:25 am Post subject: |
|
|
| LanDroid wrote: | | Quote: | | It does look like they do adaptive replication. New hosts have to be validated against a few times before they are deemed clean. Just had my first one validate with no wingperson needed. |
This leads me to an off-topic beef with Seti and some other projects that seem to double, triple, or quad-validate ALL work units. Is this still going on? I've always thought the "adaptive validation" described by Pooh Bear 27 would be MUCH more efficient, allowing those projects to chainsaw through huge amounts of data in much less time, uncovering the Wisdom in the raw data that will launch humanity Beyond InfiNIty!
|
_________________
>stats< |
|
| Back to top |
|
|
Killerrabbit
Major Oblivion
Joined: 23 May 2002
Posts: 4656
Location: in a rabbit hole near you!!
|
| Posted: Thu Jun 28, 2012 12:35 am Post subject: |
|
|
Free thinkers a re dangerous and I thought they stopped doing that.
Ni
_________________
|
|
| Back to top |
|
|
Pooh Bear 27
Prince
Joined: 28 Jan 2005
Posts: 1358
Location: Fond du Lac, WI
|
| Posted: Thu Jun 28, 2012 4:08 am Post subject: |
|
|
Adaptive replication can be useful in some instances but in others double checking can also be useful. Some sciences and math projects are very sensitive in the answers, so double checking is good.
I have no issues with Primegrid, Einstein and several others double checking the work. |
|
| Back to top |
|
|
Nuadormrac
Prince
Joined: 13 Sep 2009
Posts: 506
|
| Posted: Sat Jul 21, 2012 1:37 pm Post subject: |
|
|
As to any question, I've downloaded the lattest beta drivers from nVidia and haven't had an issue. On the Kepler cores, the retail drivers seem to be for Fermi, so they weren't exactly designed around the new product. It'll take a lil time before the drivers for nVidia's recent part gets out of beta....
One thing on credits I have noticed, Milkyway completes in about 35 minutes, and GPUgrid was in 9 hours. Milkyway gives 176 credits or so per WU, and GPUgrid was over 16,000... Now do the math, I'll do it quick and dirty which actually favors Milkyway's return (aka count it as 1/2 hour, not 35 mins, and
18 half hours * 176 credit per task == 3, 168 credits, which is way less then 16,000. At least on the Kepler cores, GPU grid is paying WAY BETTER. There's no comparison.
But and the flip side to this, I think GPUgrid is using more of the GPU, because the air comming out of the fan on this laptop felt warmer to the touch, and that area by the palm rest (which is where the HDD also resides) felt a touch warmer. So GPUgrid does run things a bit hotter, hence is probably using a lot more GPU, but returns way better credits.
Personally, would still like to hit a mil in MW at some point, but and at the same time can't totally ignore that sort of a credit difference
_________________
. |
|
| Back to top |
|
|
|
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
