Why does Apple prefer AMD over Nvidia?

[Arthur]

Why does apple prefer Polaris, Vega and RDNA over Pascal and Turing?
All I can really think of is more Half precision performance and Double precision performance and Async compute? So I guess it better suites their Metal API? but looking at the applications that their devices are targeting, i.e. Video editing, it seems that those workloads don't really benefit from half and double precision performance. Wouldn't the Nvenc encoder in Pascal and Turing benefit Video editing more? I would like to know your input on this topic, Thanks!
Sorry if this is the wrong place to ask but I'm just really curious.

[Nabster]

It is because of $$$, whoever can do the same thing cheaper wins

[Arthur]

Originally posted by Nabster:

It is because of $$$, whoever can do the same thing cheaper wins

So ur telling its cheaper to put a vega core with hbm2 rather than a 1060 or 1070 max-q?

[The HopelessGamer™]

Its none of the above.

Apple had a big falling out with Nvidia in regards to drivers not being made along with a whole other load of issues so eventually they parted ways.

https://gizmodo.com/apple-and-nvidia-are-over-1840015246

[mtono]

Originally posted by Nabster:

It is because of $$$, whoever can do the same thing cheaper wins

thumbs up!
it is all a business thing and how it came to this, an enduser doesnt know why in any case.

[Omega]

Because Nvidia is an absolutely horrible company to work with.

[max]

In a nutshell, because AMD doesn't have secrets worth withholding from Apple.

The latter probably want full source code access to Metal driver stack, something NVIDIA is very unlikely to provide to Apple, while AMD has done plenty in the past.

Other than this, NVIDIA has CUDA, which is enabled in pytorch, so it will render Apple's deep learning framework useless (on Macs). Unless again, Apple has source code access to CUDA implementation and can use it as a backend for their framework, probably not going to happen (as far as NVIDIA is concerned).

Obviously AMD were a temporary solution for Apple, and they're being ditched as these lines are written. I'm quite sure NVIDIA would fare no better (ignoring the other difficulties), with the end result of Apple retaining access good deep learning frameworks (because now they would have support for Apple's layers), and NVIDIA coming out as the loser.

So it's not really what Apple prefers, but how much NVIDIA is willing to concede/reveal.

[Arthur]

Originally posted by max:

In a nutshell, because AMD doesn't secrets worth withholding from Apple.

The latter probably want full source code access to Metal driver stack, something NVIDIA is very unlikely to provide to Apple, while AMD has done plenty in the past.

Other than this, NVIDIA has CUDA, which is enabled in pytorch, so it will render Apple's deep learning framework useless (on Macs). Unless again, Apple has source code access to CUDA implementation and can use it as a backend for their framework, probably not going to happen (as far as NVIDIA is concerned).

Obviously AMD were a temporary solution for Apple, and they're being ditched as these lines are written. I'm quite sure NVIDIA would fare no better (ignoring the other difficulties), with the end result of Apple retaining access good deep learning frameworks (because now they would have support for Apple's layers), and NVIDIA coming out as the loser.

So it's not really what Apple prefers, but how much NVIDIA is willing to concede/reveal.

Thanks for your insight! I didn’t know there was so much bad blood between the 2. And now I know why Cuda isn’t supported on MacOS and driver support stopped around mojave

[Arthur]

Originally posted by The HopelessGamer™:

Its none of the above.

Apple had a big falling out with Nvidia in regards to drivers not being made along with a whole other load of issues so eventually they parted ways.

https://gizmodo.com/apple-and-nvidia-are-over-1840015246

Thanks for ur input, i had no idea they had beef haha, always thought it was because of metal

[max]

Originally posted by Arthur:

Thanks for your insight! I didn’t know there was so much bad blood between the 2. And now I know why Cuda is supported on MacOS and driver support stopped around mojave

I wouldn't call it bad blood. It's more about similar ideologies. Imagine the reverse situation, would Apple share trade secrets with NVIDIA (or with any other company for that matter)? The only companies they share secrets with are the manufacturers, and even that comes with a stack of NDAs.
NVIDIA, in this sense, is very much like Apple, they're extremely protective of their intellectual property, because they value it like that.

Indeed, in the past there was a fallout between Apple and NVIDIA, but in business, very few people care, it's more about the bottom line and future outcome. Future outcome for Apple from such support would be immensely favorable, so if it was entirely up to them I think it would've happened already. However, from NVIDIA's point of view this would be an incredibly bad deal, so I don't think they'll do it.

I can name a few more companies that operate in a similar way. For example, Tesla. And, the surprise, there's no carplay in Tesla. And while there was some talk about Musk saying this and that, bottom line is that Tesla doesn't want Apple anywhere near their technology, so that the latter will not siphon trade secrets for their own car project.

[Arthur]

Originally posted by max:

Originally posted by Arthur:

Thanks for your insight! I didn’t know there was so much bad blood between the 2. And now I know why Cuda is supported on MacOS and driver support stopped around mojave

I wouldn't call it bad blood. It's more about similar ideologies. Imagine the reverse situation, would Apple share trade secrets with NVIDIA (or with any other company for that matter)? The only companies they share secrets with are the manufacturers, and even that comes with a stack of NDAs.
NVIDIA, in this sense, is very much like Apple, they're extremely protective of their intellectual property, because they value it like that.

Indeed, in the past there was a fallout between Apple and NVIDIA, but in business, very few people care, it's more about the bottom line and future outcome. Future outcome for Apple from such support would be immensely favorable, so if it was entirely up to them I think it would've happened already. However, from NVIDIA's point of view this would be an incredibly bad deal, so I don't think they'll do it.

I can name a few more companies that operate in a similar way. For example, Tesla. And, the surprise, there's no carplay in Tesla. And while there was some talk about Musk saying this and that, bottom line is that Tesla doesn't want Apple anywhere near their technology, so that the latter will not siphon trade secrets for their own car project.

I understand now, thanks again for your great insight! Just a shame they couldn’t make it work.