MilkyWay@home

[slavko.sk]

Prave som zistil, ze uz vznikol konecne projekt pre GPU, zatial nema ziadne jednotky. Tak som sa taktiez zaregistroval, snad niekedy kupim graficku kartu, ktora tento projekt bude podoporovat. Nas tym je uz zalozeny founderem je Hefto99
http://milkyway.cs.rpi.edu/milkyway_gpu/

No maju v plane aj CUDA takze ja to nahodim tiez.

[mbs]

Nemam teraz cas sa v tom rypat, tak sa strucne spytam:
ATI Radeon HD 2600 Pro v iMac-u bude pouzitelne? o grafickych kartach neviem nic...

[slavko.sk]

Nemam teraz cas sa v tom rypat, tak sa strucne spytam:
ATI Radeon HD 2600 Pro v iMac-u bude pouzitelne? o grafickych kartach neviem nic...

Musim ta sklamat:
1. Funguje to len pod Windows
2. ATI Radeon graphic card (HD3800 series or HD4850/HD4870, Firestream 9200 series)
(HD4890 has not yet been confirmed as working for these optimized apps)
Catalyst 8.12, 9.1 or for 9.2 and above

[mbs]

Tak dik, tak nic...

[slavko.sk]

Pomaly ale isto sa pripravuje CUDA GPU verzia priojektu, fuixuje sa klient, momentalne je verzia 0.04, dalsie kroky v najblizsich dnoch:
- dorobit check pointy
- nahodit forum a celu stranku
- prve testovacie WU
osobne mam masinu s CUDA pripojenu na projekt a cakam na prvu WU, snad to odlahci klasicky projekt a bude dostatok WU pre vsetkych.

[Kiwi]

No ja uvazujem o mesiac, dva, kupu nejakej grafiky, hlavne pre Milky,
a kedze maju ATI double precision, tak asi budu vyrazne rychlejsie, ako
porovnatelne karty od nVidie.
Skoda ze zatial sa ATI nedaju pouzit v inych projektoch.
A tiez som zacal testovat Windows 7, lebo na Linuxe sa to zatial s ATI
pocitat neda.
Viete o tom niekto nieco viac, nejake porovnania vykonu ?

[Gissmo]

inak tento projekt vam stahuje nejake jednotky ?

[gabberattack]

inak tento projekt vam stahuje nejake jednotky ?

Stahoval mi tak 5-6 na den, tak som sa odpojil.

[Kiwi]

Tam na fore sa hadaju, ako rozlisit, ci nejake ulohy taha CPU alebo GPU,
lebo vykon napr. ATI 3850 a Quadov po OC, je porovnatalny, takze treba
nejako zabezpecit, aby "ostali" aj nejake WU pre CPU, lebo vacsinou to vychytaju
ti so silnymi GPU, napr. 2 x 4870 a pod., a server je ich poziadavkami tak pretazeny,
ze pre CPU cruncherov ostava malo WU.

[gabberattack]

Tam na fore sa hadaju...

To bol dalsi dovod, preco som sa odpojil.

[slavko.sk]

Je to konkretne a presne tak, nejde o hadanie, je to realita:
Na tento projekt su take optimalizovane aplikacie ze na Q9300 sa da urobit denne 600-700WU a na masinach s grafikami aj 10 krat tolko (Q9300 mi robi WU za 9 minut, nejak ATI tu ustu urobi za 9 sekund). Preto nie su WU - server ich nestiha generovat.
Momentalne sa dokoncieva MilkyWayGPU ktory bude poskytovat LEN WU pre GPU a budu zrejme dlhsie ako normalne - to odlahci povodny projekt ktory potom bude CPU only a mal by stihat zasobovat WU.

[Kiwi]

Mozno to nejako poriesia a bude dostupna aj takato GPU,
http://eshop.x-computers.sk/zbozi/ati-s ... p=z:149505
alebo tato:
http://www.alza.sk/asus-eah4770-htdi-512md5-d120051.htm
http://benchmarkreviews.com/index.php?o ... mitstart=2
http://benchmarkreviews.com/index.php?o ... itstart=10
tak potom by sa dalo dost spocitat s malou spotrebou.

[gabberattack]

Je to konkretne a presne tak, nejde o hadanie, je to realita:
Na tento projekt su take optimalizovane aplikacie ze na Q9300 sa da urobit denne 600-700WU a na masinach s grafikami aj 10 krat tolko (Q9300 mi robi WU za 9 minut, nejak ATI tu ustu urobi za 9 sekund). Preto nie su WU - server ich nestiha generovat.
Momentalne sa dokoncieva MilkyWayGPU ktory bude poskytovat LEN WU pre GPU a budu zrejme dlhsie ako normalne - to odlahci povodny projekt ktory potom bude CPU only a mal by stihat zasobovat WU.

Ale prd robia s GPU verziou - stranka je online uz asi 2 mesiace a nespravili doteraz nic. Som tam zo zaciatku chodil kazdy den, no teraz som na Milkty uplne zanevrel.

[slavko.sk]

no teraz som na Milkty uplne zanevrel.

A to je chyba lebo by si sa docital toto:
CUDA Update -- Working Compile Script
May 22, 2009
I've updated v0.04 to the code release directory. This should work correctly with the compile script. The next steps on our to-do list are getting checkpointing working correctly for it (i'm pretty sure its not working), and compile time specified single or double floating point precision. In the next couple days I should get the milkyway_gpu forums up and running and some test work units available.
CUDA Update
May 22, 2009
I updated another version of the GPU code. It should have the evaluation directory now.
CUDA Release v0.02
May 21, 2009
I missed a couple files in the previous release, so theres a v0.02 in the code release directory now.
CUDA Application Alpha Release
May 21, 2009
I've released a preliminary version of the CUDA application in the code release directory. For more detail, there's a forum post here. This should run on the same test samples, however I'm having some issues getting a working makefile/build script using both BOINC and the CUDA libraries, so if anyone has some knowledge about how to do this it would be really appreciated.
CUDA Application Update
May 20, 2009
Looks like I've finally gotten the CUDA application working. I'm in the process of cleaning up the code, and it should be available for you to play around with tonight. It's my first crack at a CUDA application, so it's a bit rough, but currently it's crunching the current stripe 82 WUs in around a minute using a GeForce 9600M GT. I'm sure there's quite a bit more optimization we can to do get it running even faster, as this is the first version and not optimized at all. Part of the reason this has taken so long was that I was doing quite a bit of testing comparing double and single precision math on the GPU. I've come up with some schemes that will allow us to use older graphics cards which don't support double precision math (like the GeForce I've been using); which should let a lot more of our users be able to use the GPU application. More news later when I post the code.
--Travis

[gabberattack]

no teraz som na Milkty uplne zanevrel.

A to je chyba lebo by si sa docital toto:
CUDA Update -- Working Compile Script
May 22, 2009
I've updated v0.04 to the code release directory. This should work correctly with the compile script. The next steps on our to-do list are getting checkpointing working correctly for it (i'm pretty sure its not working), and compile time specified single or double floating point precision. In the next couple days I should get the milkyway_gpu forums up and running and some test work units available.
CUDA Update
May 22, 2009
I updated another version of the GPU code. It should have the evaluation directory now.
CUDA Release v0.02
May 21, 2009
I missed a couple files in the previous release, so theres a v0.02 in the code release directory now.
CUDA Application Alpha Release
May 21, 2009
I've released a preliminary version of the CUDA application in the code release directory. For more detail, there's a forum post here. This should run on the same test samples, however I'm having some issues getting a working makefile/build script using both BOINC and the CUDA libraries, so if anyone has some knowledge about how to do this it would be really appreciated.
CUDA Application Update
May 20, 2009
Looks like I've finally gotten the CUDA application working. I'm in the process of cleaning up the code, and it should be available for you to play around with tonight. It's my first crack at a CUDA application, so it's a bit rough, but currently it's crunching the current stripe 82 WUs in around a minute using a GeForce 9600M GT. I'm sure there's quite a bit more optimization we can to do get it running even faster, as this is the first version and not optimized at all. Part of the reason this has taken so long was that I was doing quite a bit of testing comparing double and single precision math on the GPU. I've come up with some schemes that will allow us to use older graphics cards which don't support double precision math (like the GeForce I've been using); which should let a lot more of our users be able to use the GPU application. More news later when I post the code.
--Travis

Ja mam ATI grafiku, takze CUDA mi je uplne naprd.