OpenCL will kill CUDA?

[RealTime]

That's what someone tried arguing to me when discussing Photoshop and the need for co-processing with a GPU to increase speed.

I don't have a strong background or thorough understanding of OpenCL or CUDA. But isn't CUDA a type of processing architecture, and OpenCL is a framework? This person seemed to imply they are interchangable. He also said OpenCL can turn any GPU into a parallel processor "just like nVidia's CUDA". He went on to say that once OpenCL gains momentum that CUDA will die because everyone's graphics processor can co-process. And that Adobe will never support CUDA in Photoshop because they will shun the ATi/AMD owners.

I am no expert, but his argument makes no sense to me.

[AntDX316]

That's what someone tried arguing to me when discussing Photoshop and the need for co-processing with a GPU to increase speed.

I don't have a strong background or thorough understanding of OpenCL or CUDA. But isn't CUDA a type of processing architecture, and OpenCL is a framework? This person seemed to imply they are interchangable. He also said OpenCL can turn any GPU into a parallel processor "just like nVidia's CUDA". He went on to say that once OpenCL gains momentum that CUDA will die because everyone's graphics processor can co-process. And that Adobe will never support CUDA in Photoshop because they will shun the ATi/AMD owners.

I am no expert, but his argument makes no sense to me.

I feel CUDA can only go so far then it hits max limit..

I feel OpenCL's max limit is beyond that of CUDA's..

the top people if it exist know that..

technology before which was inferior to today CUDA > OpenCL

but with the technology power of today or tomorrow OpenCL > CUDA

like in BC2.. that guy on youtube for NCIX with his youtube video where a crossfire setup exceeded the "100% limit".. he said it was error and took pride in it thinks he knows better..

BF3 a bench showed 2 crossfire's did exceed the "100% limit"..

so it shows there is a cap.. cause potential for 1 card is greater when that cap is lifted..

OpenCL I think is like that.. say for example low performance.. would work great with something light.. but high performance would work great with something heavy..

CUDA = light, OpenCL heavy but considering CUDA is already heavy.. OpenCL is heavier

I mean I think the faster technology or a technique is used.. CUDA shows little gain with more more performance.. but OpenCL shows more more gain with more more performance.. I think it's about the technique

[Irontiger]

I feel CUDA can only go so far then it hits max limit..

I feel OpenCL's max limit is beyond that of CUDA's..

the top people if it exist know that..

technology before which was inferior to today CUDA > OpenCL

but with the technology power of today or tomorrow OpenCL > CUDA

like in BC2.. that guy on youtube for NCIX with his youtube video where a crossfire setup exceeded the "100% limit".. he said it was error and took pride in it thinks he knows better..

BF3 a bench showed 2 crossfire's did exceed the "100% limit"..

so it shows there is a cap.. cause potential for 1 card is greater when that cap is lifted..

OpenCL I think is like that.. say for example low performance.. would work great with something light.. but high performance would work great with something heavy..

CUDA = light, OpenCL heavy but considering CUDA is already heavy.. OpenCL is heavier

I mean I think the faster technology or a technique is used.. CUDA shows little gain with more more performance.. but OpenCL shows more more gain with more more performance.. I think it's about the technique

There's one thing I don't understand. The thing I don't understand is every mother****in' word you're saying.

[jfitzw]

There's one thing I don't understand. The thing I don't understand is every mother****in' word you're saying.

I think he is trying to say that its all about how the two different techniques allow for scaling to improve, and that somehow combining two ati cards together will show greater than 100% increase in performance. I honestly dont know the difference between the two. but at the same time I feel like amd is already dead in the water. Remember when they decided to try and push their own ppu on their processors but they were so far behind agiea and later nvidia after they bought agiea had accomplished in getting their ppu out on the market and getting people using their code in games that amd didnt stand a chance. better tech/software or not doesnt always make something successful in the computing world. being first to the game with something new and pushing it out their for people to see and use and learn can cause inferior products to be a standard and become insanely successful.

[icemanchilled]

That's what someone tried arguing to me when discussing Photoshop and the need for co-processing with a GPU to increase speed.

I don't have a strong background or thorough understanding of OpenCL or CUDA. But isn't CUDA a type of processing architecture, and OpenCL is a framework? This person seemed to imply they are interchangable. He also said OpenCL can turn any GPU into a parallel processor "just like nVidia's CUDA". He went on to say that once OpenCL gains momentum that CUDA will die because everyone's graphics processor can co-process. And that Adobe will never support CUDA in Photoshop because they will shun the ATi/AMD owners.

I am no expert, but his argument makes no sense to me.

He doesn't know what he's talking about, CUDA is far more than an API and it's reach extends well beyond it's use in consumer applications. CUDA and OpenCL can happily co-exist, they may overlap at times, but that doesn't mean one will kill the other, developers will continue to use whichever methods are best for them.

[powerarmour]

CUDA is way more low level than OpenCL, so it's generally going to have better performance (on Nvidia hardware).

OpenCL trades off performance for multi-platform GPU/CPU compatibility.

[icemanchilled]

CUDA is way more low level than OpenCL, so it's generally going to have better performance (on Nvidia hardware).

OpenCL trades off performance for multi-platform GPU/CPU compatibility.

CUDA actually has a high level API and low level API and is also cross platform, supporting Linux, Mac and Windows environments. OpenCL is more extensive in it's cross platform support though.

[powerarmour]

CUDA actually has a high level API and low level API and is also cross platform, supporting Linux, Mac and Windows environments. OpenCL is more extensive in it's cross platform support though.

Well obviously CUDA only works on Nvidia hardware, whereas OpenCL can run on almost anything. That was my point.