just saw this on the milkyway site....

April 8, 2009
There might be some downtime this afternoon as we should be meeting with labstaff to get the MilkywayGPU@Home project up and running. I'll also be posting CPU code for the GPU application this afternoon so the people who have made the GPU application for the ATI cards can get their hands on it and modify it so it works with ATI cards. We should have the CUDA code out sometime this week.

looks interesting... especially as it's the first project that's aiming for both Nvidia and ATi GPU clients....

But it will stay the only one because

the people who have made the GPU application for the ATI cards

were neither MV nor ATI people.

It will be *interesting* in the way that the ATIs are perfect for the job, whereas nVidia is much slower.. if the CUDA app will still use double precision, as the current client does. ATI is about a factor of 3 faster here (assuming both are programmed equalliy well).

MrS

---

Scanning for our furry friends since Jan 2002

It will be interesting to say the least. If they stick to their modus operandi the source for the applications will be published and if history is a guide the community at large might be able to suggest and squeeze some additional performance out of the applications.

What is not at all clear is if the published applications will be CUDA only, ATI only, OpenCL Only or some mix of these options.

As I understood Travis wrote the CUDA version and will publish this code and leave the ATI CAL implementation for Gipsel.

MrS

---

Scanning for our furry friends since Jan 2002

Indeed Travis wrote MW makes the cuda application, Gipsel and friends hopefully make the Ati one.
And yes if we are right on the performance of the cards the Ati will be better on MW and nvidia gonna rock here on Gpugrid.
I will have to choose which project will be run on my little nvidia :D, at least ill test the MW@home on my machine if i can.
If the MW application is gonna run on a 9600 GT we have to wait till it gets released and see.

... and nvidia gonna rock here on Gpugrid.

Well, it's easy to rock when ATI can't run this project ;)
(or did you mean NV rock here compared to seti?)

MrS

---

Scanning for our furry friends since Jan 2002

Yes Seti gives 1000 points after running for 220.000 seconds of work.
And is flooded with fake results, i have run these AP units with stock application and see that many of my results where reported as invalid.
For that reason i deleted seti as a project on my machines as long as users like anonymous and administrator are in the user lists by seti.
Everybody with common sense knows that if a T2200 with user anonymous or a P4-2.3 Ghz outperforms a AMD 9850 BE 2.5 Ghz something fishy is going on
Hell even with a optimized application from KWSN i could not get it to finish at their reported finishing time.

On the other hand MW had ruled the boinc community with Gipsels ATI GPU client.
Clients with a 4850 made at least 30000 points a day
But it seems AMD/ATI does not have a open/free developers standard or something, untill now i have not heard the reason.
Anyway why MW choose to go with CUDA was not mentioned.

If Gipsel will or will not make another GPU client for ATI cards we have to see in the future. I would rather have seen that MW would have choosen to go with the ATI standard since there are many ATI cards sitting useless out there :)
If an ATI app will be made we can see who is the fastest on double precision calculations ;)

Anyway untill now on MW not one unit has been send to the clients yet xD.

Anyway why MW choose to go with CUDA was not mentioned.

Because it's much easier. CUDA is basically C, whereas ATI CAL is more like assembler.

MrS

---

Scanning for our furry friends since Jan 2002