DC on graphics cards more possible now?

[Swanny]

Hello all,

All new DX9 compatible video cards will need to be able to do floating point calculations. I know DC on graphics cards has been discussed here before, one of the problems being the lack of floating point capabilities. Now that barrier is gone. Video cards also get more programable with every product cycle. Is is possible that we will be seeing some kind of DC client (SETI for me) that will be able to take advantage of the power of ATi R300 or nVidia NV30 cards? Or is there still some obsticle in the way....




Thanks for your thoughts on the subject!

Swan

 

[Confused]

Hey it would be cool to get a DC program running, like Seti. Sheesh with the RAM speed and GPU speed that should really fly, as well as having a (hopefully) great FPU unit!

Hrm, running Seti on the 4GHz CPU and the 1GHz GeForce 6

 

[EagleKeeper]

Originally posted by: Confused
Hey it would be cool to get a DC program running, like Seti. Sheesh with the RAM speed and GPU speed that should really fly, as well as having a (hopefully) great FPU unit!

Hrm, running Seti on the 4GHz CPU and the 1GHz GeForce 6

fire up a dually with mulitple vid cards. Watch those work units move on out!!!

 

[m2kewl]

lol @ eaglekeeper

 

[Swanny]

Yeah, Confused has a good point with the fast RAM speeds. Since SETI really loves RAM, I think it would really run fast! Does anyone here know how graphics cards are programed? How hard would it be to adapt a SETI client to run on it?

 

[Robor]

With the prices of modern badarse video cards these days it seems like you'd be better off buying some spare MB/CPU/RAM combos to make a crack rack.

< == Just spent $200 on a GF4 TI4200 (Gainward Golden Sample)
< == Spent $300 on a GF3 (Gainward Golden Sample) last year.
< == Will never learn.

 

[Sukhoi]

It looks like the hardware can now do it. The problem will be getting the people that write clients to recompile them for the different video card. It could be a significant amount of work depending on how similar the code for the video card compilers is to the code the clients are already in.

 

[SpaceWalker]

I don't know about SETI (not doing) but RC5 ( am doing) is interger based calculations not floating point... so a video card client is no good...

 

[Baldy18]

Thats a good question. We could really turn up the production easily as I'm sure that there will be many in this forum that will upgrade to a DX9 video card whether you can run DC on it or not. Maybe we should post this in the highly technical forum and see what the brainiacs think?

 

[Swanny]

Originally posted by: SpaceWalker
I don't know about SETI (not doing) but RC5 ( am doing) is interger based calculations not floating point... so a video card client is no good...

SETI is floating point dependant

 

[Wellcky]

<== Just spent $139 on a GF4 TI4200 (Gainward GeForce4 PowerPack Ultra/650)
<== Already spent $300 on ATI 8500 64mb a year ago! :| I'll never learn either..

 

[Confused]

Originally posted by: Baldy18
Maybe we should post this in the highly technical forum and see what the brainiacs think?

I think that will be a good idea! And we'd have a reason to knock on the door now!

 

[Assimilator1]

Originally posted by: Baldy18
Thats a good question. We could really turn up the production easily as I'm sure that there will be many in this forum that will upgrade to a DX9 video card whether you can run DC on it or not. Maybe we should post this in the highly technical forum and see what the brainiacs think?

Good idea

Btw are GPU coolers built to take constant 100% loading? (me imagines fried ATI 9700!:Q )

 

[Swanny]

Ok, I'll post a thread in HT for us

 

[Swanny]

Ok, here's the thread in HT.

 

[Corsairpro]

Originally posted by: Swanny
Ok, here's the thread in HT.

Here's one to keep an eye on!

 

[ViRGE]

Originally posted by: Assimilator1
Btw are GPU coolers built to take constant 100% loading? (me imagines fried ATI 9700!:Q )

In all honesty, I'm willing to believe not. With the way video cards seem to be designed, my gut feeling is that the entire card is never under full load, only a certain unit.

 

[Swanny]

technician posted a response in my thread in HT. Here's what he had to say:

The Radeon 9700 could have that capability. If you read the 9700 preview (here on Anandtech I believe), it's mentioned that applications such as 3D Studio MAX have the potential of being run completely off the GPU.


Quote

--------------------------------------------------------------------------------
For now, where this comes in handy the most is in high-end 3D rendering applications. Due to the extremely programmable nature of the R300 (and other forthcoming DX9 GPUs), it will be possible to actually compile and run code (for example 3D Studio Max or RenderMan code) on the GPU itself. Currently 3D rendering is done entirely by your host CPU, but with proper software support, you will be able to render 3D scenes much quicker on these powerful GPUs.
--------------------------------------------------------------------------------

Link to article


Quote

--------------------------------------------------------------------------------
ATI's RenderMonkey Application will take code from any high level language and compile it to R300 assembly (e.g. RenderMan > R300 asm)
--------------------------------------------------------------------------------



RenderMonkey screenshot.

Hope that explains your question. I think RenderMonkey would be especially of interest to you...

Looks like pretty good news!

 

[Assimilator1]

Sounds promising

 

[Swanny]

Here's a bump because there's more discussion going on in HT.

 

[Sukhoi]

Thanks for the bump as I had forgotten about the HT thread.

 

[vss1980]

As far as I understand it, the main problem for this theory is that there is a set level of steps in any possible shader program that can be performed by the GPU in any particular shader program. The more complicated the program/calculations that would need to be done, the more steps needed, hence the more unlikely it would be to be done. Its not a question of having enough RAM to store these instructions, its more a case of what DX9 spec calls for. In one way, the 'nfinite FX' engine of nvidia is not really infinite enough for some. Although I suppose it may be possible to make the GPU complete a second, third, etc, set of instuctions on a particular memory area if needed.

I refer you to this page of Anand's ATI 9700 review which shows the limitations of the shader (a possible way of programming a GPU).
Floating Point precision will help with most DC projects, although RC5 is an example of where that wouldn't be needed.

The main thing most have missed though is the fact that there are limited instructions that you can perform. In mosts cases, the instruction set of a CPU is general enough to do most jobs (thats the idea of a general purpose CPU like we have - good at everything but not particularly special at any one thing at the same time). With a GPU, it will be more tailored to doing graphical orientated things, although I expect there will be even the most simple instructions in some ways in these API's and these are mostly maths based calculations performed in 3D it may not be a problem.

It may be possible, but consider the validation necessary by SETI or D.net for these programs on GPU's every time a new driver set comes out. The power is there, but will it be practical to try and tap into it in this case??

 

[Swanny]

Originally posted by: vss1980
As far as I understand it, the main problem for this theory is that there is a set level of steps in any possible shader program that can be performed by the GPU in any particular shader program. The more complicated the program/calculations that would need to be done, the more steps needed, hence the more unlikely it would be to be done. Its not a question of having enough RAM to store these instructions, its more a case of what DX9 spec calls for. In one way, the 'nfinite FX' engine of nvidia is not really infinite enough for some. Although I suppose it may be possible to make the GPU complete a second, third, etc, set of instuctions on a particular memory area if needed.

I refer you to this page of Anand's ATI 9700 review which shows the limitations of the shader (a possible way of programming a GPU).
Floating Point precision will help with most DC projects, although RC5 is an example of where that wouldn't be needed.

The main thing most have missed though is the fact that there are limited instructions that you can perform. In mosts cases, the instruction set of a CPU is general enough to do most jobs (thats the idea of a general purpose CPU like we have - good at everything but not particularly special at any one thing at the same time). With a GPU, it will be more tailored to doing graphical orientated things, although I expect there will be even the most simple instructions in some ways in these API's and these are mostly maths based calculations performed in 3D it may not be a problem.

It may be possible, but consider the validation necessary by SETI or D.net for these programs on GPU's every time a new driver set comes out. The power is there, but will it be practical to try and tap into it in this case??

So if we can't use the GPU for a whole WU, could we use it for just parts? Maybe have it does some of the FFTs like tenchim suggested? Or will it not be able to do it at all?

 

[vss1980]

As I said, it depends on the instruction set DX9/OpenGL2 use for the programmable vertex/pixel shaders. I dont know if FFT's will appear in 3D calculations that often at all. We mainly use them for fast frequency analysis and I dont know if that applies to 3D at all. However, as I said earlier, maybe the same effect can be achieved using some of the other instructions available.

I think it will be more a case of the GPU only being able to process certain types of DC contests. I think SETI may not be too complex as such, but may make use of instructions not included.