I especially like the new feature "asynchronous dispatch". It lets different apps be executed on the GPU at the same time. Fermi only allows different kernels from the same app. This could help with GPU utilisation, if the problem is using more Multiprocessors, or if the GPU is not used all the time. And it could make crunching GPUs more responsive, which is always welcome. Might even help with the CPU support / synchronization issues we're seeing at GPU-Grid now. By that I mean the CPU is not acting fast enough to keep the GPU busy - which would be much easier if the GPU was working on 2 WUs concurrently and with half a step phase difference. That way there'd be a rather large buffer or time window until the next CPU support would be required.

The feature needs OpenCL, though.

MrS

---

Scanning for our furry friends since Jan 2002

I especially like the new feature "asynchronous dispatch". It lets different apps be executed on the GPU at the same time. Fermi only allows different kernels from the same app. This could help with GPU utilisation, if the problem is using more Multiprocessors, or if the GPU is not used all the time. And it could make crunching GPUs more responsive, which is always welcome. Might even help with the CPU support / synchronization issues we're seeing at GPU-Grid now. By that I mean the CPU is not acting fast enough to keep the GPU busy - which would be much easier if the GPU was working on 2 WUs concurrently and with half a step phase difference. That way there'd be a rather large buffer or time window until the next CPU support would be required.

WOW, that feature alone is peaking my interest in GPU coding. I imagine this a result of AMD's Fusion direction.

More reading to do...

M.

From Extra Terrestrial Apes:

The feature needs OpenCL, though
---

Looks like time to ask the BOINC to provide one of two possible new features, whichever one they'd prefer:

1. Add an OpenCL interface to BOINC.

2. Allow workunits to tell BOINC to reserve a GPU for them, then turn loose of that GPU in all other ways until the workunit finishes. The application program would then have to contain all the code needed for interfacing to that GPU, but at least that would be better than not being able to use it at all in the desired fashion.

Another idea to consider: Provide workunits with two independent parts taking about as much time each. On Fermi processors, use the new feature to run one a while, then while the other one is running, do the CPU interfacing for the first one. On older processors, run one after the other instead.

Boinc Projects are going to want to facilitate Open-CL:

The forthcoming Sandy Bridge contains 6 or 12 integrated Execution Units (an on-die GPU). These 32nm processors will replace the mainstream LGA 1155 i7’s, i5’s and i3’s. While they are unlikely to be useful to GPUGrid for GPU tasks, other projects such as Einstein might be able to utilize them to good effect, and I expect some big projects will use Open-CL in order to employ these CPU’s more fully. Intel have doubled the SSE data path and introduced Advanced Vector eXtensions (AVX), an advanced SSE.

Open-CL is knocking at the door.

So the beta for ATi cards is expected some time within January?
I'm really looking forward to it.

Anny news for beta for ATi cards ?

excuse me if this has already been answered. Or if I should know better. But, i need a bit of help here.

Just installed a Radeon ATI HD 5570 GPU on my husband's computer (Intel Pentium dual cpu E2160). Increased the Windows experience rating from 3.8 to 5.9 immediately. Boinc says for ATI 5500 series GPU's you need to have driver V8.12. Windows vista went on the web and found ATI HD 5500 driver V8.791. So, I installed it. Next, when I installed the driver that came with the graphic card, it went to V8.77. Whatever :-| When *that* updated itself, it went to driver V8.791, then again to V8.8 when I installed what they call, "Catalyst Control".

HOWEVER - - When starting up Boinc V6.10.58, it gives a message "No usable GPU's found".

I like the 5570. It was cheap ($69.99), fits our Dell 530s SLIM profile case, however occupying the only single PCI X16 slot on the MB (the other three are PCI1, PCI2, and PCIX). It is passive (no fan - no noise)with massive heat fins on both sides of the card (and over the top - still fitting the SLIM profile case), and has a nifty utility for panning across multiple displays http://ati.amd.com

The stock Dell 530s has only a 250 watt PSU, but I have been assured (on another forum) that it will work OK. So far so good!

Two HIS Radeon ATI HD 5570's can be installed on a MB with two PCI X16 slots if you have them; but for that a 400 Watt PSU is recommended by ATI. Then, it lets you run, what they call, "Crossfire Mode" which, I suppose really zooms it up.

Did I say, - - When starting up Boinc V6.10.58, it gives a message "No usable GPU's found".

Am I barking up the wrong tree?

then again to V8.8 when I installed what they call, "Catalyst Control".

HOWEVER - - When starting up Boinc V6.10.58, it gives a message "No usable GPU's found".

Is that the only graphic card in your PC or is there also an on-board card? AMD (unlike nVidia) needs a monitor plugged in (or a dummy-plug) to be detected by BOINC.

Boudicca, ATI cards do not presently work on this project (GPUGrid), hence the message, "No usable GPU's found".

Try Folding@home if you want to run a molecular folding project; they also do useful science.

Boudicca,

You may want to determine which model of HD 5570 card that is, find it on this list and click on that model name , then click on Details on the resulting page to get a page including more accurate information about what rating of power supply is needed. For example, if you scroll down in the details for the first model on the list, you'll see that it needs at least a 400 watt power supply (possibly less if the rest of the computer is on a separate power supply).

http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=HD+5570&x=11&y=25

Boudicca,

you must not install BOINC in "protected mode" in order to use GPUs. It's an option during installation which is active by default. Reinstall 6.10.58 (to the same paths, so you can keep all your configuration files and WUs), look for this option and deactivate it.
Regarding the driver: just go to amd.com and get the latest from there.

MrS

---

Scanning for our furry friends since Jan 2002

SDK 2.3 is out, some news about AMD app?

i would really love to use my 5870 + 6970 for bio crunching ...
these mathematical and astro projects are boring -.-

and its okay if the app is 2-3 times slower, i would do it for science, not for credits and i think many AMD cruncher think the same way...


greetings

There has been continuous speculation as to when ATI cards would be facilitated by this project. While another SDK has been released and the programmers might be working on this or at least planning to, it would be wise to accept that so far there have been lots of disappointments and false dawns. I don’t want to get anyone’s hope up on the ATI front because of this, but I also understand some of the hardware requirements might not be in place to support such an addition to the project. Also note that if ATI cards are supported at some stage, the initial phase of their introduction would be bumpy with lots of testing, problems and periods of inactivity. So don’t think someone can just flick a switch and suddenly GPUGrid will support ATI cards.

Over the last year and a half I replaced my ATI cards with NVidia cards, to support this project. While I can’t see many people leaving the mature NVidia apps for infant ATI apps, I agree there is a lack of Biological and Medical GPU projects for ATI crunchers and respect that many would be happy to crunch here even at 30 to 50% the performance of an NVidia card. However, the problem is not just development; it is implementation, maintenance and support and these things require funding, equipment, space, time and staff.

So while I look forward to having at least one ATI card in the future, it's likely to be at least a year before I fork out on one, and I do have a GTX260 and several GT240's that could be replaced should something come along.

SDK2.3 is still unstable unfortunately. I am getting a bit tired of this level of software support from ATI.

gdf

SDK2.3 is still unstable unfortunately. I am getting a bit tired of this level of software support from ATI.

gdf

could you please share the problems you found with the community? I think that many of us would like to read something about the problems you are finding and how you're tackling them.
May I also say that maybe some problems rely on your side? I'm reading many positive comments about SDK 2.2 and 2.3... Surely a lot of work is still required for optimization, but an app should be working by now!

I’m not in the least surprised to hear that ATI support is insufficient, and I seriously doubt the merits of perusing an ATI application suitable to the complexities of this projects research. I suggest you concentrate on CUDA for the foreseeable future; I know of several aspects to the present CUDA apps that could do with your attention (I spoke too soon, being addressed here). If an ATI app is not going to happen, put it to rest and let it be. Some projects have spent over a year trying to test and develop ATI apps only to find it is impossible/worthless.

to let the AMD-app die, is not a good choice!
the developers said, an AMD-app is possible, only the background driver is instable
so, if the app will work in future, you can (at least) double the amount of results ... there are so much people which have ATI/AMD cards only to make millions of credits per day and what is their result? they prove if a mathematical theory is right or wrong ...

imagine the millions of Watts (Energy) which are useless wasted, this energy could be used to heal people, to take away the pain of so much people

i agree with cenit, please tell us the problems, maybe someone has a solution ... a little patch can help millions of people...

I’m not in the least surprised to hear that ATI support is insufficient, and I seriously doubt the merits of perusing an ATI application suitable to the complexities of this projects research. [..] Some projects have spent over a year trying to test and develop ATI apps only to find it is impossible/worthless.

Looks like a real fan-boy speaking...

Is ATI only a mathematical GPU???

By the way: Why not using external support? there are excellent examples in the history of grid-computing:

Akatosh: Einstein@Home
ClusterPhysik(Gipsel): Milkyway/Collatz

Looks like a real fan-boy speaking...

I expect you came to that conclusion by reading one post.

Is ATI only a mathematical GPU???

You answered yourself,

Akatosh: Einstein@Home
ClusterPhysik(Gipsel): Milkyway/Collatz

By the way: Why not using external support? there are excellent examples in the history of grid-computing

Funds and expert ATI OpenCL programmers familiar with molecular dynamics don’t grow on trees.
Might have been an excellent example if it came with a link, but I doubt it; MW, Collatz and Einstein dont run molecular dynamic simulations, their apps are more simplistic and very different.

We have done our bit, it's just them that have to produce a working SDK. We have filed so many bug reports in the process.
We will keep the possibility to run ATI,a lot of effort has been put it to get there with large structural changes.
The last attempt we have planned to do is to see if Windows is more stable than Linux. All our tests were on Linux so far.
ATI gpus are quite good, especially the latest generation. So hopefully we should be able to run on them.

gdf