Bruce, a question about An Optimized Application

Zilli Samuel
Zilli Samuel
Joined: 2 Mar 06
Posts: 6
Credit: 21,583
RAC: 0

RE: Read below, in the

Message 23571 in response to message 23570

Quote:

Read below, in the same thread we are now, a post from Bernd Machenschalk. He has tested it, to find that SSE2 does not give a significant improvement over SSE in Linux compiling with gcc.

Yes, but if they've to decide which set of instruction to use for improvements, I think SSE2 is the best one, because every fast CPU have it.

Quote:

And for Windows, he says there is no significant improvement when using SSE or SSE2 over the default optimizations from MSVC compiler.

Ok, but it's only automatic recompiling. I think that if the "core code" is optimized by hand to support SSE or SSE2, it should be possible to get boost in performance.
Surely, it's not possible to handcode a lot of different versions, but if they chose one (such as SSE2 in example) I think it's quite easy to maintain that optimized version up-to-date.

Anyway... it's only a suggestion! ;-)

Fletch
Fletch
Joined: 2 Mar 06
Posts: 2
Credit: 619,442
RAC: 0

While we are at it how about

While we are at it how about a client that can tap the power of my GPU? I saw an article some time in the past that with PCI-e video cards that it should be possible to use the CPU power of the GPU's.

Or is this too far fetched. I dont understand that much about coding just know many of us have a very powerfull videocard or 2 that is not being used when this client is running.

Add it to the wish list...

Troy

Wurgl (speak^Wcrunching for Special: Off-Topic)
Wurgl (speak^Wc...
Joined: 11 Feb 05
Posts: 321
Credit: 140,550,008
RAC: 0

RE: While we are at it how

Message 23573 in response to message 23572

Quote:
While we are at it how about a client that can tap the power of my GPU? I saw an article some time in the past that with PCI-e video cards that it should be possible to use the CPU power of the GPU's.

How many different GPUs are in the wild? Who will test the application?

I think this is the biggest problem.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 560
Credit: 4,514,103
RAC: 218

RE: Yes, but if they've to

Message 23574 in response to message 23571

Quote:

Yes, but if they've to decide which set of instruction to use for improvements, I think SSE2 is the best one, because every fast CPU have it.

Ok, but it's only automatic recompiling. I think that if the "core code" is optimized by hand to support SSE or SSE2, it should be possible to get boost in performance. Surely, it's not possible to handcode a lot of different versions, but if they chose one (such as SSE2 in example) I think it's quite easy to maintain that optimized version up-to-date.

I tried out a SSE2 versions on my Pentium-M. It was slower.

AMD-USR_JL
AMD-USR_JL
Joined: 24 Jan 06
Posts: 13
Credit: 27,886
RAC: 0

RE: RE: While we are at

Message 23575 in response to message 23573

Quote:
Quote:
While we are at it how about a client that can tap the power of my GPU? I saw an article some time in the past that with PCI-e video cards that it should be possible to use the CPU power of the GPU's.

How many different GPUs are in the wild? Who will test the application?

I think this is the biggest problem.


I heard from someone over at SIMAP that there are two major flavors of GPUs. ATI and Nvidia. There are also some integrated intel gpus.

Apparently they have already been using gpus in Folding@home. One guy has a link to a gpu compiler, looks like it will definetly work for newer gpus of both flavors, and it might work for older ones.

At the bottom of his post he said

Quote:

"Above readings may enable u to mount u 2 teraflops desktop at low cost
that is: about $20K

ps: 1 teraflop == 1000 gigaflops
Happy crunching :D"


I am a little skeptical of 2 TFLOPS, but if he is right then gpus could help a lot.
Here is the link to the SIMAP page about it. The gpu posts are at the bottom of the page.
EDIT: Sorry, I forgot to give you the link to download it incase you are interested.

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157,718
RAC: 0

RE: How many different GPUs

Message 23576 in response to message 23573

Quote:

How many different GPUs are in the wild? Who will test the application?

I think this is the biggest problem.

There is currently only one line of GPUs (ATI) and only one family within that line (the 1600, 1800, 1900, etc) that is designed from scratch to be used for General Computing purposes, and AFAIK ATI has not yet released the specs for the General Computing API. According to the articles that I read at the time, their schedule called for releasing those API specs in late-1st qtr - early-2nd qtr of 2006, so any time soon...

This family of GPUs covers a broad cost range, from moderate to very expensive, and has now been out for about 4 months, so there are probably several already installed in some of our crunching rigs. I'm quite sure that those of you who have them would be happy to test.

Michael

microcraft
"The arc of history is long, but it bends toward justice" - MLK

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 760
Credit: 182,671,861
RAC: 7,313

I agree: ATIs X1000 series is

I agree: ATIs X1000 series is currently the most suitable for GPGPU (general purpose GPU). I know ATI wants to push this, however, I don't know how this will look specifically. It would be great if they'd provide highly optimized math libraries.
I think the biggest problem with GPGPU today is the 32bit precision (fp).

MrS

Scanning for our furry friends since Jan 2002

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157,718
RAC: 0

RE: I agree: ATIs X1000

Message 23578 in response to message 23577

Quote:

I agree: ATIs X1000 series is currently the most suitable for GPGPU (general purpose GPU). I know ATI wants to push this, however, I don't know how this will look specifically. It would be great if they'd provide highly optimized math libraries.
I think the biggest problem with GPGPU today is the 32bit precision (fp).

MrS

I think I'll try to contact them (ATI) to see how they're coming along toward releasing developers' specs. I'm not a dev by any stretch of the imagination, but if we could get something more solid to work with ...

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 560
Credit: 4,514,103
RAC: 218

A36

A36

UBT - Halifax--lad
UBT - Halifax--lad
Joined: 10 Nov 04
Posts: 20
Credit: 3,162
RAC: 0

RE: A36 whats the link

Message 23580 in response to message 23579

Quote:
A36

whats the link for??

Join us in Chat (see the forum) Click the Sig

Join UBT

Comment viewing options

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