I was wondering if there are any plans to support CAL/Brook+ based-cards (ie ATI cards)? Maybe further down the track support for OpenCL?

I expect the developers have their hands full just getting the GTX260's to work with the 190.xx drivers, but thought i'd ask anyway.

---

BOINC blog

i was wondering this too, but haven't brought the issue forward; however, my gtx 260 will be taken off in the next month to build attempt an app for dd@h with ati. i'll be starting from scratch.

i was wondering this too, but haven't brought the issue forward; however, my gtx 260 will be taken off in the next month to build attempt an app for dd@h with ati. i'll be starting from scratch.

Well I figured that most of the projects that have a cuda app will want to support ATI.

As for writing one from scratch that would be a steep learning curve. Do you know cuda? Have you had a look at the Seti cuda app, maybe you could start there.

---

BOINC blog

i was wondering this too, but haven't brought the issue forward; however, my gtx 260 will be taken off in the next month to build attempt an app for dd@h with ati. i'll be starting from scratch.

Well I figured that most of the projects that have a cuda app will want to support ATI.

As for writing one from scratch that would be a steep learning curve. Do you know cuda? Have you had a look at the Seti cuda app, maybe you could start there.

dd@H is using gromacs/openmm, 2 different systems that no other projects have even touched if my mind serves me.

jack shultz is working on the cuda app for dd@H, i'm going to attempt an ati build. jack is sortof stuck with cuda. we figure that if we can get both going, well be able to get the project out of alpha quicker. btw, my mind hasn't even learned cuda, so i may as well not learn it unless i plan to build an app for it.

What's dd@home?

MrS

---

Scanning for our furry friends since Jan 2002

The ideal would actually be to first, try to get developers like Nick A. and Crunch3r to implement OpenCL detection to the BOINC base code and then to try to get project to migrate to OpenCL so that they will only have to write the application once to get it to run on both targets (though if I understand it correctly there may still be two deliverables) ...

This is especially timely in that Snow Tiger the latest version of OS-X is out (and installed on my Power Mac) and it supports OpenCL so we now can get the Macs into the mix ...

I will note though that the impact may be less than I would hope in that many of the standard Apple installed cards are not that "hot" meaning that the output may not as much as we would like. For example my card is a 8800GT in the Mac... which is slower than the slowest PC GPU I have ...

Of course, I might be tempted to get a faster GPU if I can put it to use ...

The ideal would actually be to first, try to get developers like Nick A. and Crunch3r to implement OpenCL detection to the BOINC base code and then to try to get project to migrate to OpenCL so that they will only have to write the application once to get it to run on both targets (though if I understand it correctly there may still be two deliverables) ...

From what I gather OpenCL will be the longer term goal, but would require seperate binaries for each vendor's brand of card. It isn't quite as developed as CUDA from what I gather, but probably will catch up in time. In the mean time we have CUDA fairly well developed and CAL/Brook+ not so advanced.

And then there is the issue of apps. Einstein have started an OpenCL one, but they also have a CUDA one for 1st beta (which by the way works on the GTX260 with the 190.xx drivers). As far as I know none of the other projects have a OpenCL app in the works. Maybe GPUgrid will follow the same path?

---

BOINC blog

The ideal would actually be to first, try to get developers like Nick A. and Crunch3r to implement OpenCL detection to the BOINC base code and then to try to get project to migrate to OpenCL so that they will only have to write the application once to get it to run on both targets (though if I understand it correctly there may still be two deliverables) ...

From what I gather OpenCL will be the longer term goal, but would require seperate binaries for each vendor's brand of card. It isn't quite as developed as CUDA from what I gather, but probably will catch up in time. In the mean time we have CUDA fairly well developed and CAL/Brook+ not so advanced.

And then there is the issue of apps. Einstein have started an OpenCL one, but they also have a CUDA one for 1st beta (which by the way works on the GTX260 with the 190.xx drivers). As far as I know none of the other projects have a OpenCL app in the works. Maybe GPUgrid will follow the same path?

I have not yet been brave enough to venture into EaH's CUDA beta yet. And My last recollection of a post by Brend was that the OpenCL version was lagging. But, my only point was it would be nice if BOINC anticipated a bit the need rather than this hodge-podge we have with the ATI cards.

I mean, 6.10.1 at long last sorta works... but Cunch3r noted on the MW forums that he has seen a number of bugs and missing code. Not the least of which is the ATI tasks don't seem to update debts correctly, why am I not surprised...

Sadly for GPU Grid will be the day when I am more confident of other projects CUDA offerings and I start to turn more attention to those projects. This will be relieved a little bit by up grades to my suite of cards and as I upgrade MB by the addition of more GPUs...

Anyway, the other point is that Snow Leopard has OpenCL included (according to the glossy advertising) and it would be nice to start to pull the macs out there into the mix...

I mean, 6.10.1 at long last sorta works... but Cunch3r noted on the MW forums that he has seen a number of bugs and missing code. Not the least of which is the ATI tasks don't seem to update debts correctly, why am I not surprised...

v6.10.3 is now out with some fixes to that code you mention. See seperate message thread here for whats in it.

---

BOINC blog