Everything is in the title, what is the state of progress of this optimized application?

Good Question, I have 10 200 Series GTX's ready to go when the Application is ready to go ... :)

As I understand, even if a CUDA application were to be released, it wouldn't provide nearly the same crunching power as the current ATI cards. I've read statements by both Cluster Physik and the admin of GPUGRID saying that. Essentially, as I understand it, ATI cards have many more units that can handle double-precision calculations than the comparable GT200 based cards. As for the actual status of the application... Travis?

---

As I understand, even if a CUDA application were to be released, it wouldn't provide nearly the same crunching power as the current ATI cards. I've read statements by both Cluster Physik and the admin of GPUGRID saying that. Essentially, as I understand it, ATI cards have many more units that can handle double-precision calculations than the comparable GT200 based cards. As for the actual status of the application... Travis?

My estimate would be that a GTX280 runs at about 40% the speed of a HD4870. To get above 50% one would have to do some magic ;)
But maybe nvidia handles the exp and sqrt functions exceptionally well (or the implementation on ATI is suboptimal), we will see.

Well, lets's get an app that will run on both and have a go. My money is still on Nvdia at the finish line.

---

Then why do the folding@home nVidia apps absolutely crush the ATI apps? My 4850 could barely chug out 1000 ppd there, while my 9800GTX+ does around 5000 ppd. Also, if the ATI cards are so much more powerful for this, why does GPUGRID not have something like what you guys have for ATI cards reather than just crunching on the PS3(nVidia) and nVidia GPUs? I'm not trying to criticize or question your knowlegde, I genuinely want to understand. :)

I'm hoping for an nVidia stock App.Would it be a bit faster than stock cpu App?
Seti-beta's gpu is aprox. 4 x faster than stock cpu app.

---

Then why do the folding@home nVidia apps absolutely crush the ATI apps? My 4850 could barely chug out 1000 ppd there, while my 9800GTX+ does around 5000 ppd. Also, if the ATI cards are so much more powerful for this, why does GPUGRID not have something like what you guys have for ATI cards reather than just crunching on the PS3(nVidia) and nVidia GPUs? I'm not trying to criticize or question your knowlegde, I genuinely want to understand. :)

Because both Folding@home and GPUGRID probably use single-precision calculations. While Nvidia packed their latest GT200 GPUs and even their last-gen G92 GPUs with shaders, most are capable of only single-precision calculations. So while they can probably do more single-precision calculations, ATI cards rock out over double-precision, which is what's used at MW.

Note that all the info above is just gathered by my reading stuff. Anyone else feel free to correct me.

---

Then why do the folding@home nVidia apps absolutely crush the ATI apps? My 4850 could barely chug out 1000 ppd there, while my 9800GTX+ does around 5000 ppd. Also, if the ATI cards are so much more powerful for this, why does GPUGRID not have something like what you guys have for ATI cards reather than just crunching on the PS3(nVidia) and nVidia GPUs? I'm not trying to criticize or question your knowlegde, I genuinely want to understand. :)

AFAIK the ATIs over at folding are held back by software issues. The resources of the GPU are not used that much (same as with SETIs CUDA app now, GPUGrid reaches about 3 times the flops on the same card :o).

The general problem of ATI cards for GPGPU computing is not the shader power, they have plenty of it (for single precision roughly the same or even slightly more than nvidia). The problem is that it's not easy to program for ATI with the Stream SDK and its documentation.

It's sometimes quite hard to implement an efficient solution. Some hardware features of the ATI GPUs can only be utilized if one resorts to a low level programming approach as they are not accesible through the high level Brook. For a somewhat larger project this might be too much effort and the sometimes patchy documentation doesn't help either. One has to admit that nvidia has done a good job with CUDA.

Hopefully ATI will do better with newer incarnations of the Stream SDK and the support of OpenCL arriving in 3 months or so.

Ahhh, that makes sense. Thank you very much for expalining things to me. :)

Come on my GTX285 wait for the optimized application ^^