Well while I enjoy crunching for MW I can never ever stay with a constant feed of work. I mean I 'might' get an 30 minutes time out of an entire day.
This is just becoming unbearable so I have to ask...
are there any other projects that run on the ATI client besides MW and Folding? I absolutely refuse to run folding.
What has prompted this is I have gone from 40-60K days to not even 20K. While the extra boost is fine, it always seems to hit when I am actually wanting to do something on the computer. If it were steady it would be something entirely different but it is far from steady.

---

Not yet. MW will soon deploy a GPU side, with longer WUs. So when you get some, they'll last a few hours each hopefully.

---

As far as I know, there are no other projects that run on the ATI.

The only other project that has announced that they will specifically target the ATI cards is AI but they are not there yet.

The Lattice Project has indicated that they may target the ATI card in the future, but they just rolled out a CUDA application and I suspect that they will be focused on that for a little time.

Einstein has indicated that it may have an OpenCL version in the works but that the CUDA version will likely see the sun first and no idea when that will happen...

Sorry we do not have happy news.

But, this too shall pass. MW did a neat thing with allowing the source to be open. That led to the ATI application being made (thanks CP and all others that made that possible) with the unintended consequence that we killed the project with speed as we ramped up the amps and the servers just could not keep up.

Yes I don't get the work I want and I have fallen from 60K per day to whatever... when Travis gets this worked out he gets it worked out and we move on ... patience is a vulture and I wish I had more of it ...

patience is a vulture and I wish I had more of it ...

I just do not understand why all these projects hoped on the CUDA but not ATI... they are both just as capable as the other.
Well I guess I can save up and buy a tesla C1060 if I am gonna go CUDA...

---

I just do not understand why all these projects hoped on the CUDA but not ATI... they are both just as capable as the other.
Well I guess I can save up and buy a tesla C1060 if I am gonna go CUDA...

CUDA is slightly easier to program, or at least that is the rumor. Secondly Nvidia supplied some resources to get applications migrated ... right now Dr. Anderson is waiting for ATI to send him a card so he will add ATI to BOINC ... no idea if that has happened yet.

But, the point is, you want the advertisement and product placement you have to pony up ... Nvidia did, ATI did not ...

But your larger point is correct, the cards are pretty much equally capable with some differences.

As to the best card you might consider the GTX 295 instead, cheaper and you can run two tasks at the same time. If I read the tables right it is actually more powerful ... of course, they are harder to find ...

[quote]

As to the best card you might consider the GTX 295 instead, cheaper and you can run two tasks at the same time. If I read the tables right it is actually more powerful ... of course, they are harder to find ...

There is absolutely no comparing the GTX 295 to a tesla C1060. Absolutely no comparing. Yes twin 295 would certainly be cheaper, but that would not be the purpose.
The Tesla is capable of over a terra gigaflops per sec. It is a GPU base, but absolutely not a video card it is a parallel CPU.
I have a very nice GPU card for video.
But with that Tesla I get far more than just a hybrid GPU. I get a computer that can do anything I want it to and damn near nothing would slow it down.

---

As to the best card you might consider the GTX 295 instead, cheaper and you can run two tasks at the same time. If I read the tables right it is actually more powerful ... of course, they are harder to find ...

There is absolutely no comparing the GTX 295 to a tesla C1060. Absolutely no comparing. Yes twin 295 would certainly be cheaper, but that would not be the purpose.
The Tesla is capable of over a terra gigaflops per sec. It is a GPU base, but absolutely not a video card it is a parallel CPU.
I have a very nice GPU card for video.
But with that Tesla I get far more than just a hybrid GPU. I get a computer that can do anything I want it to and damn near nothing would slow it down.

Uhmm, April fools day is over, isn't?

A Tesla C1060 is actually just a GTX280 with more RAM and missing video connectors. It is a complete rip off from the price performance ratio.
It is the same with AMD firestream cards, the 9170 is a HD3870, the 9250 is virtually identical to the HD4850 and the Firestream 9270 is a HD4870. But at least AMD didn't remove the video conectors, so you can use them as a normal graphics card, too ;)

But with that Tesla I get far more than just a hybrid GPU. I get a computer that can do anything I want it to and damn near nothing would slow it down.

Um, I was looking at the price / performance ratio and for the minor increase the Tesla is clearly not worth it on two fronts.

First, using Wikipedia as my source for lack of a better one with the data, the GTX 295 is rated at 1788.48 GFLOPS while the Tesla C1060 is rated at 936 GFLOPS.

So, you lose one slot and one GPU "core" meaning the C1060 can only run one task while the 295 can run two ... though each 295 "core" runs at only 894 GFLOPS...

So, if it were 25% faster I might agree that it would be better, but, its cost is far too high for to little performance and the fact that you only get one "core".

I had considered getting one to see how well it performed, but, at this moment, looking at this data ... well, I think I would rather get my third GTX295 (or its successor) instead. Of course, one problem is that rumors abound about availability of these cards and I am only hoping that I can find some when I want them ...

It's also a case of what you want you're work to do.

(It seems I keep banging on about precision, but bear with me)

The nVidia cards excel at single precision, but ATI cards storm double precision. Depending on the calculations involved, you need one or the other. Most GFLOP numbers you see floating around are for single precision, and don't tell the whole picture when computing for science. For gaming, single precision is sufficient in most cases.

---

It's also a case of what you want you're work to do.

(It seems I keep banging on about precision, but bear with me)

The nVidia cards excel at single precision, but ATI cards storm double precision. Depending on the calculations involved, you need one or the other. Most GFLOP numbers you see floating around are for single precision, and don't tell the whole picture when computing for science. For gaming, single precision is sufficient in most cases.

You are absolutely correct.

WIth one caveat ... more projects are supporting CUDA and development thereof than the AIT cards ... FOR THE MOMENT ... and if and when that changes there may be a re-evaluation of what is the best strategy ...

For example, there is a dry-spell going on here on almost all systems I run. I don't complain because these things happen. WOuld I like the 60K per day I got at one time? Sure, who would not?

But, while waiting for MW to get back on track, well, I am still GPU Grid-ing quite happily.

But, we still have to wait for BOINC to start to recognize ATI cards to get the most out of them ... and I am not at all convinced that the changes that they have been making to BOINC are going to address several fundamental issues that the developers have been steadfastly ignoring...

Also the Telsa C1060's are rated for over a terraflop per sec. I believe it is the same as the HD4850 at 1.2.
http://www.youtube.com/watch?v=UlZ_IoY4XTg

Wikipedia is not the greatest source for high specifics. I will trust a computer testing company.

Also back to semi-topic MW is the only worthwhile project running on ATI right now. Where it seems that there are many projects that are going CUDA. I say keep MW ATI. Cuda has enough projects (or will have) int he near future.

---

I would have the support for ATI if not because of the absence of a GPU client on Mac OS. Looking at the prospect of the new Mac Pro supporting the ATI 4870 against the ever-outdated 8800GT on the nVidia side, this is something I'd wish to participate but the lack of a Mac client means I'll most probably have to make do with an nVidia card, 9800GTX+ at best and the upcoming CUDA support as promised by the MW developers.

I would have the support for ATI if not because of the absence of a GPU client on Mac OS. Looking at the prospect of the new Mac Pro supporting the ATI 4870 against the ever-outdated 8800GT on the nVidia side, this is something I'd wish to participate but the lack of a Mac client means I'll most probably have to make do with an nVidia card, 9800GTX+ at best and the upcoming CUDA support as promised by the MW developers.

Don't give up the ship yet.

We want the developers to move at the speed of light ... but, it does not happen that way. When was the first CUDA application out? September last year? August? Good heavens it is less than a year ... And yes I want to run the GPU in my Mac Pro or get a better card for it too ...

But we need to take a breath here ...

I mean come on folks ... the current flapdoodle about not having work is a teapot in a tempest ... much ado about doing ... if there is work we will do it ... I suspect that the silence and the lack of work is cause Travis is doing what he suggested he might be doing ... working madly on getting us a new GPU application that will relieve our issues and get us back on track with lots and lots of work ...

Don't give up the ship yet.

But we need to take a breath here ...

I mean come on folks ... the current flapdoodle about not having work is a teapot in a tempest ... much ado about doing ... if there is work we will do it ... I suspect that the silence and the lack of work is cause Travis is doing what he suggested he might be doing ... working madly on getting us a new GPU application that will relieve our issues and get us back on track with lots and lots of work ...

I do remember doing Nanohive and the work was usually in a short burst and then a wait. And from what I've read other current projects are like that. There is still a large amount of work being offered, just with the addition of Gpu's it can't feed them too.

What about lowering the limit of wu's a gpu can do in a day? Bring it down a half? Wouldn't that help to spread some of the work around and back off the stress of the constant requests?

---

Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.

I tend to think that has already happened.