PDA

View Full Version : My video card has a processor, right?


Krusty
10-08-02, 02:18 AM
Random thought. Video cards have processors on them. Video cards are programmable.

Think it's possible to program a video card processor to run seti or folding?

It would be rather interesting to have the first contact with alien life forms found by a geforce3 gpu.

none1
10-08-02, 02:24 AM
Umm. ya huh? yep don't know.

Good idea, yes GPU are processors ...... thinking ...... probably not. The seti client source code is NOT available to the public. Given seti is nearing its end, my guess is it would be hard to get the seti folks to spend time writing / compiling the client to work on new hardware.

Good idea though, that would be cool.

JigPu
10-08-02, 09:04 AM
Perhaps somebody could write a program to launch an executable on the GPU. All it would have to do is make the card run it. Kinda like (you said they're programmable...) programming the card to run a program that you've loaded into VRAM. Would a modified Video BIOS do the trick?

Insane ideas..... With insane effort IMO to make it work (though would be awesome if it could work!!)

BTW, it won't be a GF3 GPU, it'll be a certian member's 8500 GPU ^_^
JigPu

Mictlan
10-08-02, 09:34 AM
Well, let me drop a random though here.

Yes, the GPUs are programable but there are programable within certain bounds. They are constructed with circuits designed to "crunch" triangles and pixeles, not numbers. They do have several units into mathematical processing, but not for any kind of mathematical operation like a CPU. This way they can be better than the CPU at lower speed.....just like any kind of IC (Integrated Controller) is easier to use than a general chip (like the CPU) for certain tasks (modem and sound for example)

ozzlo
10-08-02, 02:39 PM
we might be able to get our graphics cards to crunch optical seti:D

jstutman
10-08-02, 04:34 PM
so if it did work it would be like a fast ass hell pentium 100mhz :)

none1
10-08-02, 11:05 PM
Ya but..... the GPU is programable. HMMMMM....... BUT, it is not i386 compatible. So programs written for intel CPU will not run natively on a GPU.

THEORETICALLY ...... maybe you could get a compiler for the GPU. THEN, you could compile a LINUX KERNAL that runs on the GPU. NOW, we could NOT run the setiathome linux client, cause it needs intel CPU. BUT, we could then compile a windows emulator under linux (e.g. WINE). NOW, we could run the WINDOWS SETIATHOME client, under wine, with wine EMULATING an intel CPU but running on the GPU. The GPU of today probably does have enough horsepower, and certainly enoug RAM ........

Beats me how to get a compiler for a graphics card, or how to interface to it to run linux (e.g. need keyboard support, somehow, through the graphic bus)

SO, it could probably be done, but would need a hardware / linux expert with more mojo than me. :)

Digital Pimp
10-08-02, 11:30 PM
hmm... you can still get pci vid cards right.... some mb's have 6 slots......and an agp....... imagine crunching 8 wu's at a time (9 for dually) :eek:

Digital

TC
10-08-02, 11:44 PM
I hate to rain on the party, but a gpu is an asic. It would be nearly impossible to make it run an OS and process something like seti. Modern gpu's are quite good at what they do, but that's more because they're designed to do one thing very well. It's not comparable to an x86 processor.

papounet
10-09-02, 03:18 AM
I concurr with TC.

Modern GPU aren't as programmable as the PR department of their company are making them appear.

As the heart, they are managing bitplanes, and they have a BITBlitter to perform all logical combination on various bitplanes sources. As the pointers to bitplanes can be redirected around, you can do fancy double/tripple buffering tricks.

A long time ago, GPU originated as sprite engines for game machine , i.e. Hardware sprite which could be display as overlay on regular bitplanes, and moved around freely.

First programming possible was thus to be able to setup sprites, move them around and setup interrupt in case of collision (with other sprite or bitplanes).

They were already called programmable because the CPU could load sevreal instructions in the GPU (or point its vector to a serie of instructions) and have the GPU executes that set of instruction without the direct control of the CPU

Now with HW 3D acceleration, the same thing happens, complex vertex and rendering instructions are setup in the GPU memory, the GPU is set to render them, and meanwhile the CPU can go and do something else.

A good starting point to learn more would be to look into DirectX documentation.

But don't hold your breath, GPU can't help SETI .

On the other hand, computers with DSP could probably do it easily.

Demont
10-10-02, 11:15 PM
so why doesnt someone just make us a cpu that can only crunch SETI, then we could have seti cards with "SPU's" on them. 4-6 500 mhz SPU cards filling your pci bus. that would be the best. :D

Codeman05
10-10-02, 11:30 PM
Originally posted by Demont
so why doesnt someone just make us a cpu that can only crunch SETI, then we could have seti cards with "SPU's" on them. 4-6 500 mhz SPU cards filling your pci bus. that would be the best. :D

lol that would be pretty sweet

Greg M
10-10-02, 11:35 PM
I would rather see them leverage 3D Now! and SSE in their code.

Codeman05
10-10-02, 11:39 PM
I think he meant integrating seti into a cpu on a pci card...not into the actual "amd" cpu

none1
10-11-02, 12:38 AM
Originally posted by Codeman05
I think he meant integrating seti into a cpu on a pci card...not into the actual "amd" cpu

But if seti were to take advantage of the extra AMD features, then every current AMD seti machine would become much faster. Greg's right, an big improvement times the current number of amd users would make much more progress than making it easier to add more processors.

Wondering if the DSP idea would work ....... there already are DSP cards, we could fill up all our PCI slots with DSPs ..... wonder if we can get linux running on a card inside a system ........

Key is, seti won't be writing any new seti clients (for current seti) .... so we need to have a way to run one of the current clients ...... sort of like in my earlier note ... hmmmmm

papounet
10-11-02, 10:08 AM
DSP would definitely work but who's got DSP ???

The Killer idea of seti@home was:
better to have 10's of 1000 of general purpose CPU computing than 100' of DSP in a few boxes (which would have to be paid for).
On top of that, instead of getting 10's of 1000, Seti@home got more than 4.000.000 users * average CPU per users.

In other words, the brute force approach (many GPU) has won over the expert approach (few DSP) because of scalability.

jokrswild
10-11-02, 11:57 AM
you guys sound like you want to create the set accelerator all over again.. Too bad it was a hoax.. hehehe.. Do any of you guys remember that?

JigPu
10-11-02, 12:36 PM
Yeah, that was awesome. Sure wish that there was some way to actualy make one though. Would be great to have a SETI cruncher on a single PCI card! :D

JigPu

DeltaSierra
10-11-02, 12:46 PM
Originally posted by jokrswild
you guys sound like you want to create the set accelerator all over again.. Too bad it was a hoax.. hehehe.. Do any of you guys remember that?

No. I didn't hear about this one. Got more info? Sounds like a good story!