Voting for ATI Radeon

|MatMan|
|MatMan|
Joined: 22 Jan 05
Posts: 24
Credit: 249005261
RAC: 0

RE: The ATI GPU clients

Message 94736 in response to message 94735

Quote:
The ATI GPU clients running at Milkyway and Collatz were coded and compiled by Gipsel, and work very well.


Indeed they do. But I'm pretty sure (just my guess!) that the Einstein client code is way more complex and (thus) harder to parallelize than the rather straight-forward (mathematical) problems at Milkyway and Collatz...

Reuben Gathright
Reuben Gathright
Joined: 15 Feb 06
Posts: 23
Credit: 6851215
RAC: 0
Paul D. Buck
Paul D. Buck
Joined: 17 Jan 05
Posts: 754
Credit: 5385205
RAC: 0

RE: RE: The ATI GPU

Message 94738 in response to message 94736

Quote:
Quote:
The ATI GPU clients running at Milkyway and Collatz were coded and compiled by Gipsel, and work very well.

Indeed they do. But I'm pretty sure (just my guess!) that the Einstein client code is way more complex and (thus) harder to parallelize than the rather straight-forward (mathematical) problems at Milkyway and Collatz...


Yes, well, if I followed Gispel / CP's chastisement of me for my assumption that that was true he uses the same code structures for both versions at MW ... or to put it a different way, done well there is little difference between a program written to take advantage of CUDA as Stream.

That does not make it easy ... or simple ... and I wonder if he would have time to do project number three ... :)

refla
refla
Joined: 21 Oct 06
Posts: 5
Credit: 11024073
RAC: 0

RE: I made it up to 1

Message 94739 in response to message 94727

Quote:


I made it up to 1 million cobblestones on a single Radeon HD4850
in less than 2 months over at Milky Way.

Best Regards,

Bill

Hi, Bill592

nice to see you here again! :)

we meet at milky way before. http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=1212

refla
refla
Joined: 21 Oct 06
Posts: 5
Credit: 11024073
RAC: 0

RE: it takes a lot of time

Message 94740 in response to message 94731

Quote:
it takes a lot of time and effort to build programs..

OpenCL was established about 1 year ago.

I think scientific computation should use open standard, not a private APIs....

refla
refla
Joined: 21 Oct 06
Posts: 5
Credit: 11024073
RAC: 0

RE: The ATI GPU clients

Message 94741 in response to message 94735

Quote:
The ATI GPU clients running at Milkyway and Collatz were coded and compiled by Gipsel, and work very well.

A single knight cannot defeat a army!

I don't know why does ATi/AMD not pay attention to BOINC. You know the first GPU app, Folding@Home, is based on ATi X1000 series.

I hope ATi/AMD don't treat Gipsel as Don Quixote...... :-(

refla
refla
Joined: 21 Oct 06
Posts: 5
Credit: 11024073
RAC: 0

RE: Yes, well, if I

Message 94742 in response to message 94738

Quote:

Yes, well, if I followed Gispel / CP's chastisement of me for my assumption that that was true he uses the same code structures for both versions at MW ... or to put it a different way, done well there is little difference between a program written to take advantage of CUDA as Stream.

That does not make it easy ... or simple ... and I wonder if he would have time to do project number three ... :)

I deem Einstein@Home staff need PRECISION not speed.....

refla
refla
Joined: 21 Oct 06
Posts: 5
Credit: 11024073
RAC: 0

RE: I deem Einstein@Home

Message 94743 in response to message 94742

Quote:

I deem Einstein@Home staff need PRECISION not speed.....

BTW, S5R5 is still in test stage. That means the code we are using is
likely not the final version!

Bill592
Bill592
Joined: 25 Feb 05
Posts: 786
Credit: 70825065
RAC: 0

RE: I deem Einstein@Home

Message 94744 in response to message 94742

Quote:

I deem Einstein@Home staff need PRECISION not speed.....

Howdy Refla,

I think you are right about that since the Einstein Cuda

app is slower than a snail crawling through molasses .

Glad to see you are also crunching here on Einstein !

Best Regards,

Bill

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 727950019
RAC: 1225710

RE: BTW, S5R5 is still in

Message 94745 in response to message 94743

Quote:

BTW, S5R5 is still in test stage. That means the code we are using is
likely not the final version!

Not exactly, S5R5 is OVER :-), we are now beginning S5R6 (see the science section for more infos...well, it's the same app as S5R5 only the data is extended to include higher frequencies). Anyway, the S5R5/6 app is only available for CPU, and is so mature that it has remained stable for quite some time now. The CUDA beta apps are for the Arecibo Binary Pulsar Search in radio data (ABP1). There will definitely be major speed improvements in the future for this.

CU
Bikeman

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.