DSPs and where the work stood.

Message boards : Number crunching : DSPs and where the work stood.
Message board moderation

To post messages, you must log in.

AuthorMessage
NWLB.net

Send message
Joined: 3 Apr 99
Posts: 7
Credit: 0
RAC: 0
United States
Message 61782 - Posted: 7 Jan 2005, 19:54:10 UTC

I've tried searching under DSP and GPUs, but I was not able to find any definitive word on if there were any efforts underway to find a means to using graphics cards to crunch WUs. I've been trying to find sites I could track that topic on, without much success.

Also, one thread indicated somebody had tried but the results were not great. Was that the end of the effort, or did they just think it needed more work?

Either way, with the powerful new cards coming out in the last year or so, there has to be a way. Has anybody contacted somebody at ATI or other card makers?
ID: 61782 · Report as offensive
Profile AthlonRob
Volunteer developer
Avatar

Send message
Joined: 18 May 99
Posts: 378
Credit: 7,041
RAC: 0
United States
Message 61794 - Posted: 7 Jan 2005, 20:14:24 UTC - in response to Message 61782.  

> I've tried searching under DSP and GPUs, but I was not able to find any
> definitive word on if there were any efforts underway to find a means to using
> graphics cards to crunch WUs. I've been trying to find sites I could track
> that topic on, without much success.

There is work being done on this, although I don't know what kind of success they've had: http://setiboinc.sourceforge.net/

Rob
ID: 61794 · Report as offensive
Profile Doris and Jens
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 21 Nov 99
Posts: 362
Credit: 3,539,386
RAC: 13
Germany
Message 61815 - Posted: 7 Jan 2005, 20:41:34 UTC - in response to Message 61782.  

There was a project together with Nvidia last year. But I believe it was not really successfuly. They tried to do the FFT computation on the GPU, what seems to work, but mostly the data transfer between GPU and main memory was to slow for good results.

One way for future work is, to do all parts of the computation on the GPU, for reducing the data transfer. But this needs more high level language support, the Brook (Stanford University) GPU programming language for example. Folding@home is working too on this.

A other interesting point at the moment is the rapid hardware development of new bus systems for the graphics adapters. This may help in the future to split the computation of WUs in a CPU and a GPU part.



Greetings from Bremen/Germany
Jens Seidler (TheBigJens)

ID: 61815 · Report as offensive
Profile ghstwolf
Volunteer tester
Avatar

Send message
Joined: 14 Oct 04
Posts: 322
Credit: 55,806
RAC: 0
United States
Message 61819 - Posted: 7 Jan 2005, 20:44:05 UTC - in response to Message 61782.  

> Also, one thread indicated somebody had tried but the results were not great.
> Was that the end of the effort, or did they just think it needed more work?
>
> Either way, with the powerful new cards coming out in the last year or so,
> there has to be a way. Has anybody contacted somebody at ATI or other card
> makers?
>

I seem to remember the issue had to do with the AGP bus (asymetrical bandwidth???). Of course PCIexpress changes the game, and the huge potential of such a beast will keep people working on it.

Any successful attempt will likely require vendor support (ATI/NVIDIA/SIS and so on) to work well. The pitch would be the 1st video card to accelerate your system. Bringing Adobe, Autodesk, and Oracle on board (start planting a bug in their ear now) would definately add some marketability to the venture. Kind of a step back to the days of seperate math coprocessors but, when it works, it works.


Still looking for something profound or inspirational to place here.
ID: 61819 · Report as offensive
NWLB.net

Send message
Joined: 3 Apr 99
Posts: 7
Credit: 0
RAC: 0
United States
Message 61846 - Posted: 7 Jan 2005, 21:21:43 UTC - in response to Message 61819.  

Ahhhh, Coprocessors. I'd have liked having a MCP for my old 386sx 25.

I wonder how SLI and ATI counterpart to it might effect such an approach. On the other hand, the concept creates dreams of systems with dual quad core CPUs, two or four GPU's, and crushing Units like tanks in a Godzilla movie.

I have to think at some point they will look at making the processing power of GPUs open to the OS and hardware. Those things are getting way to much power to go to waste anymore. Not everybody will need the card for Doom 3 and Halo all of the time.

Now if they had a way for X-Box and PS3s to crunch units...... Aw heck, I have a history degree, not like I can do anything about it.
ID: 61846 · Report as offensive
Janus
Volunteer developer

Send message
Joined: 4 Dec 01
Posts: 376
Credit: 967,976
RAC: 0
Denmark
Message 61857 - Posted: 7 Jan 2005, 21:49:40 UTC - in response to Message 61846.  

> Now if they had a way for X-Box and PS3s to crunch units...... Aw heck, I
> have a history degree, not like I can do anything about it.

Potentially there is nothing wrong in compiling BOINC and seti@home on X-box and PS2. It should be perfectly possible to do so (even without much trouble). The only thing is that noone has ever done it yet (that I know off).

You would have to have one of those free linux distros for X-box/PS2 installed of course, be one of those lucky ones with a harddrive(/usb flashdrive?) and then you simply build and run the linux client and required libs.
You would also need to have an internet connection for the console. Which requires an adaptor in the PS2 case.

Theoretically there's a chance that it could even run in a ramdrive (which I would recommend if using the flashdrive approach. but dunno specs for X-box and ps2 except that their CPU power is not that impressive...)

I'm still looking forward to seeing the first screenshots of CLI seti running on PS2 - I'm so sad that I don't own one myself =)

__
Aw well, sorry for hijacking the thread...
ID: 61857 · Report as offensive
7822531

Send message
Joined: 3 Apr 99
Posts: 820
Credit: 692
RAC: 0
Message 61868 - Posted: 7 Jan 2005, 22:13:45 UTC - in response to Message 61857.  

Aw well, sorry for hijacking the thread...
Not at all! From what I have heard of the PlayStation 3, it will have eight PowerPC 970fx processors. SETI, Einstein, Pirates, Climate Prediction, et cetera, are all repetitive tasks which could take advantage of the Single-Instruction Multiple-Data capabilities of the processor. I would love to see that kind of power harnessed for a BOINC project (When I'm not playing games, of course!)

Returning back to topic, is the GPU burdening the project's graphics or is it being handled by the CPU? Under Mac OS 9, the CPU had to create SETI's graphics, not the GPU. Will there be a shift towards applying Quartz Extreme in the Mac version?
ID: 61868 · Report as offensive
Profile Paul D. Buck
Volunteer tester

Send message
Joined: 19 Jul 00
Posts: 3898
Credit: 1,158,042
RAC: 0
United States
Message 61882 - Posted: 7 Jan 2005, 22:44:21 UTC - in response to Message 61857.  


Re: x-box, I have a note on how to install Linux on an x-box ... so you can take it from there ... it should be in the BOINC FAQ under other platforms.
ID: 61882 · Report as offensive
John McLeod VII
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 15 Jul 99
Posts: 24806
Credit: 790,712
RAC: 0
United States
Message 62078 - Posted: 8 Jan 2005, 5:11:51 UTC - in response to Message 61882.  

>
> Re: x-box, I have a note on how to install Linux on an x-box ... so you can
> take it from there ... it should be in the BOINC FAQ under other platforms.
>
If I recall correctly, there was a thread that the XBox used a unix variation, and was not that hard to hack. However, it would probably void any warranty that you had left.


BOINC WIKI
ID: 62078 · Report as offensive
Hans Dorn
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 3 Apr 99
Posts: 2262
Credit: 26,448,570
RAC: 0
Germany
Message 62160 - Posted: 8 Jan 2005, 9:40:29 UTC - in response to Message 62078.  

> >
> > Re: x-box, I have a note on how to install Linux on an x-box ... so you
> can
> > take it from there ... it should be in the BOINC FAQ under other
> platforms.
> >
> If I recall correctly, there was a thread that the XBox used a unix variation,
> and was not that hard to hack. However, it would probably void any warranty
> that you had left.
>

The xbox only has 64MB of RAM and will hit the new memory size limit.
There's some tweaking required to overcome this ;-)

A 733 Celeron isn't exactly the best seti CPU either.

Regards Hans

P.S:
This doesn't mean I won't add a xbox to my crunching zoo soon, just for the heck of it :o)
ID: 62160 · Report as offensive

Message boards : Number crunching : DSPs and where the work stood.


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.