OTOY Announces cross-platform CUDA...

[AEponym]

Wow, incredible. Maybe Apple is going to announce a new Mac Pro following amd's announcement for external gpu's…??

I'm just afraid Nvidia's respond will be : "let's kill all Otoy guys at sight "

If I understand correctly the term retro-engineering used in the article : Otoy tore apart one of nvidia's biggest asset in the GPU battle, Cuda, which I think is proprietary code, subject to copyright laws, to make it usable on every chip there is. Congrats!

Is Nvidia planning to buy Otoy? To keep the development of Otoy's advanced GPU technologies in-house?

If it's not that, it's a very very bold move from Otoy.

[mustardseedsg]

I was wondering about the same thing... is this endorsed by Nvidia, and if not wouldn't they need to protect their assets? CUDA basically keeps people buying Nvidia hardware. If CUDA is opened to all hardware, wouldn't Nvidia be losing sales? Or is Nvidia licensing CUDA tech out?

https://youtu.be/A8I9pYCl9AQ

[whersmy]

Just sit back and enjoy the ride.

Im predicting a NVOTOYDIA

Or OTIDIA

oh yeah and mark this: apple egpu. They aren`t that stupid not to come with a device like that.
Matter of time. Not an 'if'.

[kommando_2k]

hmmm is that legal to reverse engineered CUDA software and outport to AMD cards & CPU?. That would be breach EULA

[prehabitat]

? Or is Nvidia licensing CUDA tech out?

I think you nailed it mustardseedsg,

It makes more business sense to assist(maybe even fund) someone with the R&D funding & vested interest in cross-porting/compiling Cuda to everything then charge everyone a licence fee than it does to sit back and wait for the inevitable rise of openCL (or some other alternative) maturity to put competitive pressure on CUDA.

[milanm]

Regarding legality of it, I think you guys will NVOY reading this

http://www.anandtech.com/show/5238/nvid ... ce-kind-of

Regards
Milan

[prehabitat]

http://wccftech.com/amd-cuda-compilerco ... nitiative/

Doesn't read as being as good: only about 90% of the code is translated; whereas the Otoy solution seems super low level?

[Goldorak]

http://wccftech.com/amd-cuda-compilercompatibility-layer-announced-with-the-boltzmann-initiative/

Doesn't read as being as good: only about 90% of the code is translated; whereas the Otoy solution seems super low level?

Boltzmann didn't work for us and as you pointed out is only a partial translation tool. We couldn't just run: Octane compile CUDA->AMD from untouched code as we do here.