Suspect lots of us will be joining you Fred .... an AMD app at GPUGRID long been a hope in many minds - and as long as AMD come up with the goods on 79XX compared to NVidia, lots will.

If they dont, suspect many will change to NVidia in a heartbeat - I certainly will. 28nm is a defining moment, its a huge step change from 40nm, and if NVidia screw this up aka 3.XX and 4.XX .... I suspect they will be in trouble. I hope not, high prices on cards in the last two years is because basicly AMD had it all their own way for years, and that didnt change until 560/570/580 came out, they were a last min saviour for NVidia.

We need good offerings from NVidia in the markeplace ....

Regards
Zy

Its 7.0.3 and above. I am running 7.0.8 and its fine. I've seen 7970s at POEM running 7.0.3, so as long as 7.X.X is loaded - should be fine

Regards
Zy

Yes, I'm able to run both POEM & Milkyway with the 7.0.3 Client ...

---

STE\/E

Any updates? :)

Ohlalalaaaa :) - gpugrid on AMD gpu is just my DREAM..

In that moment, when you`ll release a stable gpugrid app for 7950 ill buy one!

Good luck guys!

p.

Any progress on the AMD side of things?

Any progress on the AMD side of things?

AMD HD7970 announced

Any progress on the AMD side of things?

Up.

I have tested donate@home but i like run my ati card here. Any news from ati app?

Hello dear all,

just want to post a link to the top hosts list @ einstein:
http://einstein.phys.uwm.edu/top_hosts.php

AMD is leading there and at positions 3,4 and 6.

And here? Still talking about poor performance of AMD openCL?

Reality has overtaken you, developers!

Alexander

Lol

Top supporters != most efficient

I've run Einstein, and I can tell you that their CUDA app is more efficient.

Edit: also, Jeroen is #2 because he is running at pcie 2.0, since you cannot run 3.0 with Kepler on SB-E on Linux. The reg hack windows users can use obviously does not work for linux.

Lol

Top supporters != most efficient

I've run Einstein, and I can tell you that their CUDA app is more efficient.

Edit: also, Jeroen is #2 because he is running at pcie 2.0, since you cannot run 3.0 with Kepler on SB-E on Linux. The reg hack windows users can use obviously does not work for linux.

anyway, even if inefficient, many users have requested an amd OpenCL app. Maybe, as in einstein, even if inefficiently, that app could take the top spot (and in the meanwhile you can always improve your code...), thanks to the extreme performances of GCN computationally wise (much more than Kepler, btw, if you're honest - just GK110 can be comparable but it does not exist in any consumer card for the moment).
But it seems that here the devs are absolutely not interested. And this is a pity.

But that app has the top spot because the 1 other person there, who would blow past the other guy, is on Linux.

Im not saying that ATI GPUs are not decent, it's just that OpenCl isn't.

EDIT: And the difference between PCIe 2.0 and 3.0 on Einstein app is something like 20% better runtimes when running multiple tasks. And computational performance does not always equate to better performance as can be seen at Einstein.

But that app has the top spot because the 1 other person there, who would blow past the other guy, is on Linux.

Im not saying that ATI GPUs are not strong, it's just that OpenCl isn't.

EDIT: And the difference between PCIe 2.0 and 3.0 on Einstein app is something like 20% better runtimes when running multiple tasks.

OpenCL is a language. It cannot be strong or weak.
The problem, maybe, is that so many very well developed libraries exist only for cuda. And many devs have an expertise only in that language.
And also the tools were much more developed for cuda, from the compiler to the debugger. Things have improved dramatically, but here it seems that they're never interested. Every gen that comes out they say: this is the perfect one, we will look into it. They also said, once, "we have an app ready, please enable beta". But nothing ever happened then. No real exchange of true informations. I'm sure that in this community there're also some good programmers. Just release the code opensource and someone can also help you.

OpenCL is a language. It cannot be strong or weak.
The problem, maybe, is that so many very well developed libraries exist only for cuda. And many devs have an expertise only in that language.
And also the tools were much more developed for cuda, from the compiler to the debugger. Things have improved dramatically, but here it seems that they're never interested. Every gen that comes out they say: this is the perfect one, we will look into it. They also said, once, "we have an app ready, please enable beta". But nothing ever happened then. No real exchange of true informations. I'm sure that in this community there're also some good programmers. Just release the code opensource and someone can also help you.

You got the point, cenit.
First time I asked for an app for ATI-cards the HD5xxx series was state of the art.
When the HD6xxx series was up to date, they posted, their openCL app is working, but performance is poor. But still no beta apps.
Now we have the HD7xxx in the market, next rework (of the GCN) is announced, openCL has proofed to be fast and reliable, and the reaction here: not existent.
It was the time when I installed my first HD5xxx card when they advised us to enable beta apps.

I gave up the hope for OpenCL-app since the start of donate@home. IMHO the question is: Is it wise to release an APP for AMD, when you can earn money from the performance of AMD at donate? That will canabalize donate, which is only a support project to fund GPUGrid.

This question can only be answered by the devs.

OpenCL is a language. It cannot be strong or weak.
The problem, maybe, is that so many very well developed libraries exist only for cuda. And many devs have an expertise only in that language.
And also the tools were much more developed for cuda, from the compiler to the debugger. Things have improved dramatically, but here it seems that they're never interested.

This is not true. Even your next statement disproves it.

Every gen that comes out they say: this is the perfect one, we will look into it. They also said, once, "we have an app ready, please enable beta". But nothing ever happened then.

While I can recall the developers said that they have an app ready, they said the same time it's not reliable and/or very slow (compared to the CUDA app), therefore it's not ready to release even as beta.

No real exchange of true informations. I'm sure that in this community there're also some good programmers. Just release the code opensource and someone can also help you.

According to this post, the code of the ACEMD client won't be open source.

I gave up the hope for OpenCL-app since the start of donate@home. IMHO the question is: Is it wise to release an APP for AMD, when you can earn money from the performance of AMD at donate? That will canabalize donate, which is only a support project to fund GPUGrid.

This question can only be answered by the devs.

Yes, that is exactly what I have been thinking since Donate was rumored and since Kepler was glorified; now they found an excuse for all AMD owners, as poor as that is, that they gave the volunteers an AMD application to support this project, well, indirectly by running none science related bitcoining at yet another project spoiling the volunteers with credits instead of doing valuable science. The developers are simply too inexperienced with AMD OpenCL, so they blame it on everything else, that is the only thing I can conclude from that.
I am not an OpenCL developer, but then again this is not my task, as I am donating my hardware, power and time to a project refusing to adapt to another programming platform than CUDA...
This is all I have to say about that.
Now the CUDA fanboys are free to call me names, insult me and bash me again, as that is what happens every time AMD owners complain rightfully... :))

---

I personally would switch my 7950 instantly after the end of WCG HCC GPU to GPUGrid IF they would run OpenCL. But not to donate (only tested it there in the downtime of WCG last week). It would bring me very much Points (620k per day) but im not a credit hunter, so i want to run extra buyed dedicated BOINC Hardware to science.

Its a good way to fund the project by volunteers who has problems to donate to boinc projects when they have to much hardware lieing around ;) I tried one time to donate money over my creditcard to, i dont know was it seti or something?, and mastercard banned my creditcard (not reversible!) because they think it was a unwished transaction. i dont want to do that anymore again -_-

Pllzz make an OpenCL Support, even when it is not the best performanced one. Im struggling a little bit what i can make with the 7950 after HCC. At the time i only can use it for...MW then..I dont have the hardware to support a 7950 on POEM :/ *sign*

---

DSKAG Austria Research Team: http://www.research.dskag.at

I would suggest Einstein.

I'm currently in a bind of wishing to buy a 7950 for HCC, but at the same time, I feel I would have no place to take said card after HCC GPU app ends or during its "dry spells"

Guys.. let's stay cool, take a step back and recapitulate what we're dealing with here. GPU-Grid

- has limited ressources and needs to decide how to make the best use of them
- requires manpower on the science side of things: analyse the results, formulate new problems to solve, write publications, supervise thesis'.. all the usual stuff
- requires manpower to improve the CUDA code
- requires manpower to maintain the CUDA code (work around issues popping up with new driver releases and support of many combinations of cards, driver and OS)

They built an AMD app, but it performed poorly due to hardware and software issues. The hardware side may be solved by GCN, but the software wasn't some time ago. The software certainly has improved, but I don't know if the encountered stability issues are solved yet. And someone made something work in OpenCL - fine, but they're using different code, function calls & libraries. This is nice to know, but doesn't guarantee the GPU-Grid code will be fine.

In any way, developing and supporting an AMD code path using other development tools and a different language will require additional manpower. It even introduces much more complex cross-platform testing.

Surely this can be done - given enough manpower. But GDF has said in the past that they'd rather properly support nVidia cards than nVidia and AMD in a mediocre way.

Of course one would prefer perfect support & optimization for both.. but this is not a realistic option, as it seems. You're of course welcome to report on OpenCL progress and ask about the current state. But don't get mad at the developers if they can't realize everything you want.

MrS

---

Scanning for our furry friends since Jan 2002

Guys.. let's stay cool, take a step back and recapitulate what we're dealing with here. GPU-Grid

- has limited ressources and needs to decide how to make the best use of them
- requires manpower on the science side of things: analyse the results, formulate new problems to solve, write publications, supervise thesis'.. all the usual stuff
- requires manpower to improve the CUDA code
- requires manpower to maintain the CUDA code (work around issues popping up with new driver releases and support of many combinations of cards, driver and OS)

GDF has said in the past that they'd rather properly support nVidia cards than nVidia and AMD in a mediocre way.

Of course one would prefer perfect support & optimization for both.. but this is not a realistic option, as it seems. You're of course welcome to report on OpenCL progress and ask about the current state. But don't get mad at the developers if they can't realize everything you want.

MrS

Well said MrS.