That points to this thread which contains this message. "I hope there will be one later" isn't much of a timeline IMHO. I would call that "some talk".
Yes, which indicates that it's in the testing phase. If anyone has accurate predictions of the results of that testing - prior to the tests being done that is, so that we can then perhaps speed up development even more by obviating their need - then the developers would love to hear from you. Don't be shy, step forward ! :-)
Cheers, Mike.
I have tried to speed up development in this thread. However, it seems it has fallen on deaf ears as I'm the only one posting in it as of late. I have searched the forums over a few times. If there's a better thread for the OpenCL GPU app development, I'd like to know.
That points to this thread which contains this message. "I hope there will be one later" isn't much of a timeline IMHO. I would call that "some talk".
Yes, which indicates that it's in the testing phase. If anyone has accurate predictions of the results of that testing - prior to the tests being done that is, so that we can then perhaps speed up development even more by obviating their need - then the developers would love to hear from you. Don't be shy, step forward ! :-)
Cheers, Mike.
I have tried to speed up development in this thread. However, it seems it has fallen on deaf ears as I'm the only one posting in it as of late. I have searched the forums over a few times. If there's a better thread for the OpenCL GPU app development, I'd like to know.
So you're personally volunteering to do the development and testing work, right ?? ..... :-) :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
So you're personally volunteering to do the development and testing work, right ?? ..... :-) :-)
Cheers, Mike.
Considering Aqua@home posted their OpenCL code for download, I figured that was a huge leap forward for "how-to's" that the Einstein project people could handle.
To directly answer your question, if they would like to pay me to quit my job and program, sure.
So you're personally volunteering to do the development and testing work, right ?? ..... :-) :-)
Cheers, Mike.
Considering Aqua@home posted their OpenCL code for download, I figured that was a huge leap forward for "how-to's" that the Einstein project people could handle.
To directly answer your question, if they would like to pay me to quit my job and program, sure.
Ah. So that's a no then. 'Volunteer' means no pay you see ... :-)
We only have two developers and they're pretty busy. It's not the case that OpenCL, or any other development, is being held up for lack of sufficient advocacy or the presence of deaf ears. Thanks for your thoughts in any case. :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
We only have two developers and they're pretty busy. It's not the case that OpenCL, or any other development, is being held up for lack of sufficient advocacy or the presence of deaf ears.
* Of the two developers, one is on his well-deserved and long-planned vacation now, and the other is currently sick. They both had a pretty high workload before with two new searches (BRP4 and FGRP1) to issue.
* We're still sorting out issues with OpenCL support in BOINC. In fact, we are actually funding one BOINC developer to get this done. I don't know what exactly Aqua@home does, it couldn't be more than a workaround until there is official OpenCL support in BOINC.
* OpenCL is meant to be an open, generic, vendor-independent standard, meaning that you can write code for e.g. NVidia and ATI devices using OpenCL. However an OpenCL program that runs efficiently on one device may not run as efficient (or, if using certain extensions, run at all) on a different architecture. ATI cards have fewer but 'wider' cores than NVidia, you have to use ATI-specific code (extensions I think) to harvest the theoretically larger computing power of ATI cards. As the first version of OpenCL apps will probably be a plain port of the CUDA app and won't run as good on ATI as on NVidia, I suspect that for quite some time the CUDA app will continue to be the most efficient app on E@H. It's certainly not worth to buy an ATI card in the near future to crunch for E@H.
AQUA started with CUDA code quite early, but found that for the particular science they require, multi-threaded CPU applications were more appropriate and more efficient. The fact that AQUA still appears on BOINC's list of GPU projects is merely a vestige of those early experiments - they haven't generated any CUDA workunits for at least 18 months.
They did try re-building their GPU app as OpenCL, and it got as far as off-line testing a year ago (May 2010), but it turned out to be even less efficient than CUDA for their purposes, so they didn't persue it.
I appreciate the responses. I didn't want to be a message board gorilla, but some times you have to kick the hornet's nest in order to evoke responses. Also, having previously worked in academia myself, I am painfully aware of resource/people/funding issues and sincerely meant no disrespect.
Thanks for the updates.
Edited to Add:
GPU computing is on the horizon as AMD is gearing up to fuse (Llano APU), then integrate, GPU cores into their CPUs. Soon GPU computing won't be done on just graphics cards, but the APU also.
I don't know what exactly Aqua@home does, it couldn't be more than a workaround until there is official OpenCL support in BOINC.
AQUA has decided to stop development of GPU apps because what they want to run is not suitable for high-parallel apps currently.
Quote:
OpenCL is meant to be an open, generic, vendor-independent standard, meaning that you can write code for e.g. NVidia and ATI devices using OpenCL. However an OpenCL program that runs efficiently on one device may not run as efficient (or, if using certain extensions, run at all) on a different architecture. ATI cards have fewer but 'wider' cores than NVidia, you have to use ATI-specific code (extensions I think) to harvest the theoretically larger computing power of ATI cards.
OpenCL is currently not much more than a plan. AMD is not putting in enough effort, NVIDIA is not even half-harted because CUDA is working quite well and INTEL - oh well. even if they would - it will need a long time to develope compilers which get good performance out of all those different architectures.
currently i think CUDA is the one and only way to go..
OpenCL is meant to be an open, generic, vendor-independent standard, meaning that you can write code for e.g. NVidia and ATI devices using OpenCL. However an OpenCL program that runs efficiently on one device may not run as efficient (or, if using certain extensions, run at all) on a different architecture. ATI cards have fewer but 'wider' cores than NVidia, you have to use ATI-specific code (extensions I think) to harvest the theoretically larger computing power of ATI cards.
ATI's started talking about their new GPU core; and it appears that they're moving towards a fermiesque architecture. Hopefully that will result in easier convergence apps in the future.
RE: RE: That points to
)
I have tried to speed up development in this thread. However, it seems it has fallen on deaf ears as I'm the only one posting in it as of late. I have searched the forums over a few times. If there's a better thread for the OpenCL GPU app development, I'd like to know.
RE: RE: RE: That points
)
So you're personally volunteering to do the development and testing work, right ?? ..... :-) :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: So you're personally
)
Considering Aqua@home posted their OpenCL code for download, I figured that was a huge leap forward for "how-to's" that the Einstein project people could handle.
To directly answer your question, if they would like to pay me to quit my job and program, sure.
RE: RE: So you're
)
Ah. So that's a no then. 'Volunteer' means no pay you see ... :-)
We only have two developers and they're pretty busy. It's not the case that OpenCL, or any other development, is being held up for lack of sufficient advocacy or the presence of deaf ears. Thanks for your thoughts in any case. :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
On OpenCL: RE: We
)
On OpenCL:
* Of the two developers, one is on his well-deserved and long-planned vacation now, and the other is currently sick. They both had a pretty high workload before with two new searches (BRP4 and FGRP1) to issue.
* We're still sorting out issues with OpenCL support in BOINC. In fact, we are actually funding one BOINC developer to get this done. I don't know what exactly Aqua@home does, it couldn't be more than a workaround until there is official OpenCL support in BOINC.
* OpenCL is meant to be an open, generic, vendor-independent standard, meaning that you can write code for e.g. NVidia and ATI devices using OpenCL. However an OpenCL program that runs efficiently on one device may not run as efficient (or, if using certain extensions, run at all) on a different architecture. ATI cards have fewer but 'wider' cores than NVidia, you have to use ATI-specific code (extensions I think) to harvest the theoretically larger computing power of ATI cards. As the first version of OpenCL apps will probably be a plain port of the CUDA app and won't run as good on ATI as on NVidia, I suspect that for quite some time the CUDA app will continue to be the most efficient app on E@H. It's certainly not worth to buy an ATI card in the near future to crunch for E@H.
BM
BM
AQUA started with CUDA code
)
AQUA started with CUDA code quite early, but found that for the particular science they require, multi-threaded CPU applications were more appropriate and more efficient. The fact that AQUA still appears on BOINC's list of GPU projects is merely a vestige of those early experiments - they haven't generated any CUDA workunits for at least 18 months.
They did try re-building their GPU app as OpenCL, and it got as far as off-line testing a year ago (May 2010), but it turned out to be even less efficient than CUDA for their purposes, so they didn't persue it.
I appreciate the responses. I
)
I appreciate the responses. I didn't want to be a message board gorilla, but some times you have to kick the hornet's nest in order to evoke responses. Also, having previously worked in academia myself, I am painfully aware of resource/people/funding issues and sincerely meant no disrespect.
Thanks for the updates.
Edited to Add:
GPU computing is on the horizon as AMD is gearing up to fuse (Llano APU), then integrate, GPU cores into their CPUs. Soon GPU computing won't be done on just graphics cards, but the APU also.
RE: I don't know what
)
AQUA has decided to stop development of GPU apps because what they want to run is not suitable for high-parallel apps currently.
OpenCL is currently not much more than a plan. AMD is not putting in enough effort, NVIDIA is not even half-harted because CUDA is working quite well and INTEL - oh well. even if they would - it will need a long time to develope compilers which get good performance out of all those different architectures.
currently i think CUDA is the one and only way to go..
RE: RE: OpenCL is meant
)
ATI's started talking about their new GPU core; and it appears that they're moving towards a fermiesque architecture. Hopefully that will result in easier convergence apps in the future.
Here's a quickie: Is BRP4
)
Here's a quickie: Is BRP4 going to run on Nvidia cards just like BRP3?
I checked the message boards but I couldn't find a definitive answer.