Page 1 of 1
#1 MilkyWay@Home Heads Up
Posted: Sun Feb 22, 2009 11:18 am
by jockmacmad2
As I'm sure some of you are aware MilkyWay@Home has an unoffical GPU engine for ATI. MilkyWay also had a daily limit of 1000 WU per day per CPU.
I see from their forums 2 things are changing.
1. The workload limit is being raised to something like 10,000 (currently upped to 5k) per GPU and I say GPU as they are talking abount not raising the CPU as an I7 cannot manage the 1000 WU already. So there will be 2 max values a CPU one and a GPU one.
2. A CUDA version of the app is on the way.
Afaik, there is already a student starting to work on a CUDA app. As this is easier to work with, I guess we could see some results soon
But don't expect times much below 25s per WU for nvidias GTX line. And older ones won't work at all (lack of double precision units).
Yeah hopefully within the next week or two we'll have an alpha CUDA application for you guys to crash
So 25 seconds per WU on CUDA vs. I believe 6s on an ATI HD4870. Also only works on nVidia GTX-2xx due to the requirement of Double Precision results.
The full link is at
GPU app teaser
Enjoy
#2
Posted: Sat Mar 14, 2009 7:27 pm
by jockmacmad2
ATI 4850 running Version 19d of the Milkyway mod. Nice points.
Wish I could get these damn nVidia working as well as a single ATI. At this rate they can go in the bin to be replaced by ATI. 4 GPU on the nVidia but only 1 working returning 1/4 the points of the ATI. So thats £1000 quid of nVidia been beaten by £150 of ATI.
Seems the nVidia boat is being spoilt for the happeth of tar not spent on creating proper drivers. 2 x GTX-295 != 4 x GTX-260 but = 1 x GTX-260.
Why can nVidia never sort their 4%#? out with consumer drivers. their workstation ones are awesome but consumer ones are coded by the cast of the muppet show
#3
Posted: Sat Mar 14, 2009 9:22 pm
by Megacruncher
Woohoo! another Klondyker joining the Milkyway ATI bonanza!
In fairness to Nvidia at least with Cuda you can carry on crunching with your CPU cores without crashing everything.
Also I've got to say that the credits being issued by MW are on the decidely iffy side of generous. I simply don't believe that my £135 HD4850 is really doing more work than my roomful of 260GTXs and Q6600 which together must have cost me more thousands than I can comfortably calculate. They have a history of excessive credit giving which TSBT have made the most of while it lasted and until Boinc told them to get back in line.
It won't last, so don't chuck out all your Nvidea cards, I'm sure better drivers will be along soon, and the Boinc police will no doubt be feeling the Milkyway collar very soon.
#4
Posted: Sun Mar 15, 2009 1:25 am
by jockmacmad2
Tempting fate but using the 19d Milkyway Mod of the ATI I have not had a single issue in 24 hours.
#5
Posted: Sun Mar 15, 2009 7:33 am
by steve
jockmacmad2 wrote:Tempting fate but using the 19d Milkyway Mod of the ATI I have not had a single issue in 24 hours.
The server is overloaded .... The only answer is to keep banging the update button . Project admin is away with the flu so it is Monday before you see any improvement......
#6
Posted: Mon Mar 16, 2009 8:37 am
by Megacruncher
I've had no problems getting work for Milkyway and since I got my wired broadband back up I've been managing a fairly steady and totally outrageous 3,000 credits per hour. Of course Steve's demands are rather heavier!
#7
Posted: Mon Mar 16, 2009 9:07 am
by jockmacmad2
Same here went to bed at 2:30. All stopped at 3am.
Allegidly there is a script on the forums that automates the update button pushing.
#8
Posted: Mon Mar 16, 2009 2:46 pm
by jockmacmad2
Okay so I have a weasel like plan.
I have downloaded Z-Cron for Windows and set a job up to run every 5 minutes to call:-
to force an update of the project.
#9
Posted: Mon Mar 16, 2009 8:57 pm
by Megacruncher
My luck seems to have held up - no middle of the nigh interuptions for me.
#10
Posted: Mon Mar 16, 2009 11:37 pm
by steve
Its all these dam credit whores grabbing all the work units that has caused this problem....
errr ummmm
( Guilty )
http://boinc.berkeley.edu/chart_list.php
#11
Posted: Mon Mar 16, 2009 11:57 pm
by Megacruncher
Steve's link is to the Boinc Top 100 Multi-Project Volunteers. You will see he is number 1
. I come in at 77th & John McLean is 96th.
If you look carefully you will see that if you are running GPUGrid you get two equal sized slices of pie for this, one labled GPUGrid and one marked PS3Grid!
Hopefully by the time they spot & sort out this bit of double counting my MilkyWay RAC will be high enough to keep me in this elite listing.
#12
Posted: Tue Mar 17, 2009 10:02 pm
by jockmacmad2
Klondyker indeed.
Weeeeeeeeeeeeeeeeeeeeeeeee. If only they could supply enough WU's.
Cards are idle for large amounts of the day
All I need is 24 WU's per 30 seconds shame they cannot keep up lol.
Development of secret weapon for updating under way and gives me chance to learn C# along the way lol.
#13
Posted: Tue Mar 17, 2009 11:57 pm
by jockmacmad2
I see we passed BOINCStats on Milkyway@Home today to go 9th.
Who is betting they try to rally?
Team Norway should be overhauled tomorrow to get 8th.
In the UK we are 32 days away from becomming No. 1 passing UK BOINC Team
[waits for said team to post comment :)]
#14
Posted: Wed Mar 18, 2009 8:29 pm
by Ben
jockmacmad2 wrote:I see we passed BOINCStats on Milkyway@Home today to go 9th.
Who is betting they try to rally?
Team Norway should be overhauled tomorrow to get 8th.
In the UK we are 32 days away from becomming No. 1 passing UK BOINC Team
[waits for said team to post comment :)]
In the words of Homer Simpson... D'oh!
#15
Posted: Wed Mar 18, 2009 11:41 pm
by Megacruncher
There's no doubt about it we are kicking Milkyway ass!
We are 4th by RAC & the RAC still reflects "just" Steve's awesome performance and has yet to incorporate fully the additional awesomeness of the recent contributions of JockMacMad & meself.
If anybody else wants to escalate their Boinc RAC by 70K per day they could not do better than add an HD4850 to one of their PCs.
As I've previously said the ludicrous returns being offered by MW for ATI card users can't last but while it does - grab it!
#16
Posted: Thu Mar 19, 2009 6:27 am
by jockmacmad2
Mega you ruinning a single HD4850?
I am struggling with getting WU's at the moment big time. I have a 4850 and 2 x 4870 but am not getting the full performance due to WU lean patches. I just checked my 4850 machine and it did no work between 1am and 6am
#17
Posted: Thu Mar 19, 2009 8:53 am
by Megacruncher
Single card and not too many outages (although it's not been quite so full on in the last 48 hours & I checked just now and the tank was empty).
I've got my network usage preferences set to connect to the network about every 5 days and additional work buffer of 1 days which should keep Boinc Manager anxiously seeking work-units.
#18
Posted: Thu Mar 19, 2009 7:28 pm
by jockmacmad2
I wish I had built my rack servers rather than buying them
As I bought Dell they are all 1U and 2U with small power supplies. The one I did buid myself has a nice 650W OCZ and the 4850 in it.
I currently have 5 servers powered down as the 8k ish they provide is not worth the electricity bill seeing the output the ATI's are managing. Yes yes I know I'm whimping out.
I have 3 nVidia (2 x GTX-295 and the GTX-260) on the shelf as they will not fit nor be powered in the Dell racks.
#19
Posted: Sun Mar 29, 2009 12:41 am
by Megacruncher
Uh-Oh!
Milkyway has gone tits up!
No-one has had credit since 6pm yesterday evening.
Has the bubble burst or will the project be up and straining or ATI cards by morning. Let us look forward to some straining. :)
#20
Posted: Sun Mar 29, 2009 9:57 am
by jockmacmad2
Alas nada here this morning.
#21
Posted: Mon Mar 30, 2009 12:28 am
by Megacruncher
The MW website is back up again & I see that peeps are getting credit again. I'd plugged an old 9600 into my test rig so I'll let it finish it's GPUGrid test WU before I put the 4850 back in again.
It's quite impressive that we managed over 658K yesterday in a day when we got next to nothing from our best paying project and only had 23hrs to crunch in!
#22
Posted: Mon Mar 30, 2009 2:39 am
by jockmacmad2
I see the nVidia version is being worked on and there will be no GPU drop in credits when they have finished reworking GPU vs. CPU WU's.
And max. allowable WU's for GPU's will be increased. Whoo Hoo.
Nice.
#23
Posted: Mon Mar 30, 2009 12:33 pm
by steve
SERVER LOAD!
#24
Posted: Mon Mar 30, 2009 12:57 pm
by jockmacmad2
Well I imagine it is a side effect of:-
Second, we're going to be doing a few server-side changes to how things are run, which should significantly help with workunit reporting and work availability. We'll be splitting MilkyWay@Home into two separate applications/backends, Milkyway regular and Milkyway GPU. This will allow us to keep current workunit sizes for non-GPU applications, while farming a lot more work out to the GPUs (so they need to contact the server less frequently).
We'll be doing different optimization methods on the regular Milkway and the GPU Milkyway so they'll both be doing equally valid research. In the next day or two I'll be posting the code that the application will be using for the GPU version. After testing to make sure that it's working correctly we'll swap to only awarding credit to the GPU applications for GPU milkyway (that way our servers wont be as bogged down, which will mean better work availability).
Also, before a credit firestorm erupts again, we have no plans to award less credit/work to the GPU applications, so you don't need to worry about that.
#25
Posted: Mon Mar 30, 2009 11:42 pm
by Megacruncher
Hopefully it'll be worth waiting for. In the meanwhile I can get the square root of SFA from the project. If I could get the whole farm running it and CUDA is as generous as the ATI port then we will indeed have lift off.
For now we don't.
#26
Posted: Tue Apr 28, 2009 11:00 am
by jockmacmad2
nVidia will not quite be as generous as ATI simply due to the number of double-precision FPU's between the 2 cards.
Now there is some FUD around how many of these Milkyway will actually need, but it will be anywhere from the same time to 5x longer depending on how many it does use. Seems ATI didn't skimp on the FPU's.
Like others I just wish I could keep the cards I have busy
