Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
CUDA is dying other than hobbyist ML. The big cloud players are all moving to ASIC (e.g. Google TPU). CUDA was a half-decent interim step but making a GPU do ML/AI tasks won't compete with purpose built silicon.

That's true for bulk Google-scale production, not for developmental and research work. You can't afford to spin an new ASIC, or even learn to code for a proprietary ASIC, to explore new ML methods.

You wouldn't do production ML on a desktop computer either. You do development work. Enough stuff to then deploy it on a cluster.

There's also a ton of other scientific computing on CUDA that's not ML.
 
  • Like
Reactions: 09872738
didn't the original retina MacBook Pro have an Nvidia card too?

Apple List of OpenGL support outlines the GPUs used . https://support.apple.com/en-us/HT202823

The Nvidia GPUs disappear about early 2014 ( 2013 models limping into 2014 ) .
The MBP 2013 models are already on the Vintage list . https://support.apple.com/en-us/HT201624
They are schedules to fall onto full Obsolete in 2020. Extremely likely the 2014 system will transition from Vintage to Obsolete by 2021.

The Mac Pro (2012) will be full Obsolete status next year. Cards from that time will be in same boat.


More than likely Apple and Nvidia have a tech support contract that lasts the lifecycle of those systems about to age out. these are also the older Metal feature families . So basically just coating until hit macOS cut list Keyed off the Obsolete stasus.

Stuff that is completely decouple from Apple use as an embedded or official explicit support add In card has been adrift for over a year.
 
I wonder how long until Apple takes ownership of the built in Nvidia drivers and cuts off the support relationship. Apple still has some Nvidia Macs they’d want to support (like early rMBPs), but the chipsets aren’t changing and neither are the supported features.

At some point Apple could just take the driver source and maintain it themselves. Just have to make sure nothing breaks and that things are kept aligned with the internal driver ABI.

If the Metal ABI becomes stable enough to not even have to touch the nVidia drivers even better.
 
Apple has their own drivers for all NVIDIA products they have issued. Nothing breaks with them.
 
Apple has their own drivers for all NVIDIA products they have issued. Nothing breaks with them.

Just because Apple is distributing something doesn't necessarily mean Apple wrote it or owns it outright. For example, Apple distributed Rosetta to ease the PPC -> x86 transition. At no point though did Apple own it. Nor were they doing the low level coding there.

It is quite likely that the drivers Apple is distributing are wholly "Apple" drivers. Substantive parts there are likely licensed from Nvidia. ( and likely no embedded GPU vendor gets a design win for a product unless they commit the necessary driver support for the lifecycle of the product). So Apple can keep it afloat with some incremental macOS changes but they aren't doing this with Nvidia utterly out of the loop. They "own" and distribute a branch/fork of the Nvidia driver, but to label them "Apple drivers" is probably a stretch in ownership implication. .
 
Apple has their own drivers for all NVIDIA products they have issued. Nothing breaks with them.

Apple doesn't write the built in drivers. Nvidia does.

There has been a lot of leaking around how unhappy Apple is with the situation. They don't like how long Nvidia takes to turn fixes and improvements around.

They don't write the AMD or Intel GPU drivers either, but both those companies have engineers located on Apple's campus, which dramatically reduces the time it takes to get fixes done.

A myth got started on these forums that Apple writes the built in drivers. I'm not sure where that comes from, it's not true.
 
but with the next system release after catalina, all macs
with nvidia GPUs (books and imacs) are obsolet.
and then there will no nvidia driver systemwide
 
Apple was selling 15" MacBookPro's with GeForce GT 750M (Mid-2014 model) through at least mid-2015. The 15" MacBookPro with Radeon R9 M370X models were not introduced until mid-2015.

Should be at least one more macOS version compatible with Kepler GPUs.
 
CUDA is dying other than hobbyist ML. The big cloud players are all moving to ASIC (e.g. Google TPU). CUDA was a half-decent interim step but making a GPU do ML/AI tasks won't compete with purpose built silicon.

Not really. Not everything can be run cloud native. The automotive industry, heavy machines, construction industry, and many other industrial applications use nvidia solutions for their AI tasks. Most smaller engineering firms and teams also use nvidia on premise. Robotic applications use nvidia. Scientists use nvidia, e.g. bioinformaticians and physical/chemical simulation applications. All hobbyists use nvidia. So I would say that nvidia is dominating the scientific applications/AI field today. Things may change of course.
 
Basically the Mac platform is confirmed dead for many scientific users, particularly in machine learning/AI.

The advantage of CUDA is the existing software base and portability from desktops to datacenter. Metal simply isn't available on servers or on cloud computing providers, so it's not an option.

Windows has nicely taken over, with Microsoft's Windows 10 push for Linux and command line support, plus far better workstation and hardware support.

Apple has decided the only true pro user they care about is video production.

This is so true.
Currently kicking along with my 4,1 and 5,1 Mac Pro’s on 10.13.6

Running Ubuntu natively on my 4,1 with a GTX1080 for scientific CUDA compute.

Next machine will be threadripper and Ubuntu.
 
  • Like
Reactions: AidenShaw
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.