OpenCL the end of Stream and CUDA

[Stiganator]

..

 

[chizow]

They're all C based and "open", but I'd expect NV and AMD to continue pushing their own standard, brand and SDK with a wrapper and any necessary fixes for OpenCL compatibility. Same goes for DX11 when it arrives. For example, I'd expect NV to just write a DX11 wrapper for PhysX instead of writing code from scratch.

 

[ViRGE]

Short answer: Maybe

Long answer: OpenCL really can't be compared to anything else. The fact that the Khronos group finished the spec in record time (they never get anything done quickly) and the fact that Apple is pushing this hard implies that all parties are on board. On the flip side we know AMD needs this a lot more than NVIDIA, so I'm not sure NVIDIA is really behind it. And Microsoft is doing their own thing entirely AFAIK.

We'll see what happens next year.

 

[Wreckage]

Not really. So far CUDA is really the only thing to actually show real application.

 

[Munky]

May not be the end for now, but it's a nice alternative, and couldn't come soon enough. I hate the idea of writing code vendor-specific code.

 

[taltamir]

Originally posted by: ViRGE
Short answer: Maybe

Long answer: OpenCL really can't be compared to anything else. The fact that the Khronos group finished the spec in record time (they never get anything done quickly) and the fact that Apple is pushing this hard implies that all parties are on board. On the flip side we know AMD needs this a lot more than NVIDIA, so I'm not sure NVIDIA is really behind it. And Microsoft is doing their own thing entirely AFAIK.

We'll see what happens next year.

And then intel has their own x86 on GPU thing that they will bring with larabee (and AMD will bring that with fusion... eventually).

This is a format war with so many players that you sometimes forget to LIST all of them.

 

[cmdrdredd]

Originally posted by: taltamir

Originally posted by: ViRGE
Short answer: Maybe

Long answer: OpenCL really can't be compared to anything else. The fact that the Khronos group finished the spec in record time (they never get anything done quickly) and the fact that Apple is pushing this hard implies that all parties are on board. On the flip side we know AMD needs this a lot more than NVIDIA, so I'm not sure NVIDIA is really behind it. And Microsoft is doing their own thing entirely AFAIK.

We'll see what happens next year.

And then intel has their own x86 on GPU thing that they will bring with larabee (and AMD will bring that with fusion... eventually).

This is a format war with so many players that you sometimes forget to LIST all of them.

I'm not sure the Intel solution will matter for gamers, but system integrators yes.

It all comes down to this folks. When there's a game that is so revolutionary and so genius that you have to have physix or whatever because without it the game is just junk, that will be the time when any of this matters.

Lets face it, you can write all the APIs and wrappers you want and have the hardware available and drivers there, but if no games touch it then the value is 0.

 

[nRollo]

Originally posted by: Stiganator
With OpenCL as an official standard. Will both companies pick it up, or are they going continue down the same road. I hope we get more info on this implementation soon. It seems like it will be an important thing for every programmer to learn soon.

Given the number of universities, indutries, and scientists using CUDA, I don't think it's going away anytime soon.

Every week I get at least a few NVIDIA press releases emailed to me about some big new use of it (I believe portable Cray supercomputers was arecent one) so it seems like it's really gaining traction and acceptance across a wide variety of uses.

 

[taltamir]

Originally posted by: cmdrdredd

Originally posted by: taltamir

Originally posted by: ViRGE
Short answer: Maybe

Long answer: OpenCL really can't be compared to anything else. The fact that the Khronos group finished the spec in record time (they never get anything done quickly) and the fact that Apple is pushing this hard implies that all parties are on board. On the flip side we know AMD needs this a lot more than NVIDIA, so I'm not sure NVIDIA is really behind it. And Microsoft is doing their own thing entirely AFAIK.

We'll see what happens next year.

And then intel has their own x86 on GPU thing that they will bring with larabee (and AMD will bring that with fusion... eventually).

This is a format war with so many players that you sometimes forget to LIST all of them.

I'm not sure the Intel solution will matter for gamers, but system integrators yes.

It all comes down to this folks. When there's a game that is so revolutionary and so genius that you have to have physix or whatever because without it the game is just junk, that will be the time when any of this matters.

Lets face it, you can write all the APIs and wrappers you want and have the hardware available and drivers there, but if no games touch it then the value is 0.

there are some guys who are talking about badaboom v1... So far only one person provided benchmarks, but he is saying it cut down encoding time from 11.5 hours to 50 minutes. That is certainly impressive if you are into video encoding (which is one of the chief reason people buy quad cores).

 

[cmdrdredd]

Originally posted by: taltamir

Originally posted by: cmdrdredd

Originally posted by: taltamir

Originally posted by: ViRGE
Short answer: Maybe

Long answer: OpenCL really can't be compared to anything else. The fact that the Khronos group finished the spec in record time (they never get anything done quickly) and the fact that Apple is pushing this hard implies that all parties are on board. On the flip side we know AMD needs this a lot more than NVIDIA, so I'm not sure NVIDIA is really behind it. And Microsoft is doing their own thing entirely AFAIK.

We'll see what happens next year.

And then intel has their own x86 on GPU thing that they will bring with larabee (and AMD will bring that with fusion... eventually).

This is a format war with so many players that you sometimes forget to LIST all of them.

I'm not sure the Intel solution will matter for gamers, but system integrators yes.

It all comes down to this folks. When there's a game that is so revolutionary and so genius that you have to have physix or whatever because without it the game is just junk, that will be the time when any of this matters.

Lets face it, you can write all the APIs and wrappers you want and have the hardware available and drivers there, but if no games touch it then the value is 0.

there are some guys who are talking about badaboom v1... So far only one person provided benchmarks, but he is saying it cut down encoding time from 11.5 hours to 50 minutes. That is certainly impressive if you are into video encoding (which is one of the chief reason people buy quad cores).

Going to be written for STREAM as well which makes it universal. Again, these fancy things don't matter if you don't use it. I don't encode video so it doesn't interest me. I also have no interest in leaving my CPU on 24/7 to fold. For me it has to be valuable in games that I enjoy.

 

[thilanliyan]

Originally posted by: taltamir
there are some guys who are talking about badaboom v1... So far only one person provided benchmarks, but he is saying it cut down encoding time from 11.5 hours to 50 minutes. That is certainly impressive if you are into video encoding (which is one of the chief reason people buy quad cores).

Is that encoding or transcoding? I thought for example AVIVO did encoding and cut down the time by a lot but it turns out it was just transcoding...and fyi transcoding sucks.

EDIT: Just took a look at the badaboom website and it says it's transcoding and again encoding (or recoding) is much better. If you want a quick method of say converting a video file so you can play it on a media player transcoding is okay but quality won't be amazing.

 

[taltamir]

Originally posted by: cmdrdredd
Going to be written for STREAM as well which makes it universal. Again, these fancy things don't matter if you don't use it. I don't encode video so it doesn't interest me. I also have no interest in leaving my CPU on 24/7 to fold. For me it has to be valuable in games that I enjoy.

not really.. with AMD in the market with two technologies, x86 and their own gpu acceleration, nvidia with physics, openCL for both, DX11 from MS and intels x86 larabee there are for too many formats for anything to be universal.

 

[taltamir]

thilan29, very good point. Really not as good as I hoped it would be.