Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
I still don't understand the choice of GPU in the 2016 and 2017 MBP, the radeon pro 455-560 are honestly so rubbish. A mobile 1060 would have been such a hugely better option in my eyes. So the fact that people (like myself) are buying eGPU's is saying something about apples current line up. Definitely is disappointing that Nvidia cards are still not natively supported and more so that thunderbolt 1/2 have been rendered useless in 10.13.4...


Nvidia GTX 1070 vs AMD Radeon Pro 555
Userbenchmark Effective 3D Gaming GPU Speed: 99.7% vs 19.6%
Nvidia GTX 1070 - 15th / 579
AMD Radeon Pro 555 - 125th / 579
Even a 1060 Max-Q has about a 70W TDP. The Radeon 560 in the 2017 MacBook Pro is supposed to be 35W. AMD actually has pretty powerful stuff for low wattage applications.
 
  • Like
Reactions: sd70mac
now imagine:

2018 macmini with quad core and tb3
tb3 sonnet puck amd 570 blessed by apple
multiple tb3 SSDs and HDDs
buy an empty Macpro (tower) or Powermac G5 case off ebay
mount everything inside the empty MacPro case
(solder wires from MacPro power button to MacMini power button?)

boom you have a desktop Mac with easily swappable GPU and drives for the first time in years

when the next “puck” sized eGPU is available you can easily upgrade
 
  • Like
Reactions: sd70mac
I don't get it. What is it with Apples bias towards AMD? Nvidia kills AMD in GPU rendering and 3D partical simulations. Cinema4D integrated pro render and frankly it just sucks.

Oh well, another Apple mystery.

I love Apple but I hate them for not going all the way with this. There stupid reason (whatever it may be) for ignoring Nvidia in ALL their products is ridiculosly consumer-hostile and also really hurts the ability to use their damn machines with anything other than their own software.

side note concerning pro render: of the 3 times I tried pro render it took ages to "compile shaders" and whatever it has to do to even start rendering and and some point it crashed every time
 
No NVIDIA support?? What were they thinking?

They will sell 80% less because they don't support the major grahics cards company.
 
  • Like
Reactions: sd70mac
Yes it does. Not officially supported by apple of course.

I have the eGPU dev kit from apple, the one that come with the Pulse 580.

To make it work I bought apple TB2 to TB3 adapter, which is bidirectional. I plugged the male end of the adapter in the eGPU breakout box, then used a standard thunderbolt cable from my Mac mini 2012 to the female end of the adapter.

and it still worked on macOS 10.13.4??? Have same Mac mini so would like to know if I can keep using it
 
Does this mean I could play some AAA games (Through Windows) with a GTX 1080ti using my 2016 matchbook pro?

Edit: NM...now Nvidia support :(

You can still play them. I play games on my 2014 5K iMac in Windows with a sonnet eGPU case with a 1070. You might have to read forum guides and fiddle around with the setup at first, but it can be made to work.

Windows doesn't care that Apple/macOS doesn't support nVidia cards...
 
The Pro 555 is in the 2017 MBP not the 2016. Is it not just space? Look at the 12" MacBook, the tiered battery design Maximises its design minimising wasted space. Nvidia state the 1050 mobile is 53 watts, in addition to the standard apple 'under clock' you are looking at 40-45 watts tops. Im finding it hard to believe that you are arguing FOR there pro 555. It just simply doesn't live up to the 'pro' spec anywhere. The CPU's in the line up are great! But the GPU, just lets the laptop down...
Look, for the 2018 variant i'm hoping we might see the mx150 (/successor) or the 8th gen intel w/ RX Vega M. For a professional product these would be welcomed.

Batteries take up both physical space and add the most weight of any component.

Also, I don’t really care what they use. I can only try to understand why. And I can only speculate as to that. I’m assuming it’s really because of a myriad of things. NVidia is probably a crap partner who wants to shove CUDA everywhere. Apple wants to go with something else. AMD has proven they’re willing to work with anyone to make custom-ish solutions for them.

Would you work with the company that tells you what you’re going to use or the company that works with you?
[doublepost=1522515002][/doublepost]
How about Apple opensource Metal? Or ditch Metal, and rather use Vulkan, which is open source?
Apple with their proprietary Metal API isn't any better than Nvidia with their proprietary Cuda technology.

I’d rather they stick with Metal and not screw over everyone who has already moved their stuff over to Metal.
 
Yea... without Nvidia support, what's the point?

Why does Apple hate gamers so much?


I mean, gaming on a mac is hardly ideal...And on a MBPs chip you wont push an nVidia GPU to it's max unless it's mid range and that's where AMD sits.
 
  • Like
Reactions: Regime2008
My DIY eGPU from 2014 with Thunderbolt 2 and GTX 970 still works fine. I'm not sure why you need to have Apple support?
Not on 10.13.14 it doesn't. It will kill any functionality you have now.

Does this work with thunderbolt 1 or 2?
not on 10.13.4. It used to on 10.13.3 and earlier.
[doublepost=1522516708][/doublepost]
Yes it does. Not officially supported by apple of course.

I have the eGPU dev kit from apple, the one that come with the Pulse 580.

To make it work I bought apple TB2 to TB3 adapter, which is bidirectional. I plugged the male end of the adapter in the eGPU breakout box, then used a standard thunderbolt cable from my Mac mini 2012 to the female end of the adapter.
NO it does not. 10.13.4 will kill any functionality you have. Your solution works ONLY on 10.13.3 and earlier.

I'm pissed as I have an EGPU setup right now that works great and I am stuck on 10.13.3 for the time being
[doublepost=1522516773][/doublepost]
why would I buy eGPU if I can get Internet GPU. Streaming games for a 2 months now.
well some people use gpus for things other than games
[doublepost=1522516848][/doublepost]
A lot of wrong information given in this thread. Please be aware that
  • TB 1/2 did work but does no longer work
  • Bootcamp support would need an EFI update which does not exist. Booting Windows often results in a hard to solve Windows error 12 (not enough resources for the PCI devices)
  • Nvidia drivers exist, but they do not support eGPU out of the box
  • only those TB3 PCI boxes work, which are explicitely sold with eGPU support for Mac
Research egpu.io for all the problems and tutorials. I did a lot of trial and error. If you want it to work, please just use exactly the things that Apple recommends.
Exactly.

Its unnerving.

However, I have read that a TB2 enclosure with the adapter can work on TB3 machines still
[doublepost=1522516892][/doublepost]
To elaborate. eGPU will work natively in MacOS 10.13.1 and 10.13.2. My Mac Mini 2012 quad i7 is a very useful machine after adding an eGPU last year. Now it is stuck on 10.13.2 forever because Apple specifically removed the functionality of the eGPU for Thunderbolt 1 & 2. They could have left the capability in and just not supported it, but chose to cripple older machines instead. This really bothers me, intentionally removing the ability to upgrade a machine after allowing the upgrade previously. If you can remain on 10.13.2 for the life of the machine, then eGPU will work on TB 1 & 2.
You should be able to have it work on 10.13.3 as well. That is how I am running my setup on my 2012 mbp
[doublepost=1522516930][/doublepost]
I have been using an eGPU with my 2015 MBP, which is of course TB2. Does this release stop support for that? It has worked up to this point with the unofficial eGPU support in High Sierra. I want to know before upgrading.
Yes, it kill the unofficial support
[doublepost=1522516988][/doublepost]
Not officially. But you can go to eGPU.io and there are guides for TB1 and TB2 and how to get them working.
Not for 10.13.4! this update kills any functionality that had existed prior.
[doublepost=1522517072][/doublepost]
Can one still hack as usual to get TB1 and TB2 support?
That is the hope. I have quite the expensive workflow that offing Apple just killed for NO reason
[doublepost=1522517157][/doublepost]
This is unacceptable.
Especially since it was working in 10.13.3 and earlier
 
No, it's Metal vs CUDA and also Metal vs DirectX. Closed source or otherwise restricted platforms on both sides. Mainly cause whoever is behind OpenCL/GL held things back for so long that both sides gave up.

Errr not really. "whoever is behind". The Khronos Group is behind the OpenCL/GL. The Khronos is basically an umbrella group for a committee of implementers and users. That group includes Apple, Nvidia, and (to a lesser extent) Microsoft.

https://www.khronos.org/members/list Apple, Nvidia are in the Promotoer level of sponsorship. Microsoft is down in the Contributor levels.

So you have folks who have their own proprietary competitors to OpenCL/GL members of the committee that gets to vote and approve of the feature list of the open solutions they are competing with.

Case in point the recent press release for OpenCL 2.2

" ... “By finalizing OpenCL 2.2, Khronos has delivered on its promise to make C++ a first-class kernel language in the OpenCL standard,” said Neil Trevett, OpenCL chair and Khronos president. ..."
https://www.khronos.org/news/press/khronos-releases-opencl-2.2-with-spir-v-1.2

Hmmm, let's go look and see who is Trevett's "day job" employer.....

https://en.wikipedia.org/wiki/Neil_Trevett " ... Vice president of Mobile Ecosystem,[1] NVIDIA Corporation "

half joking I could say "fox guarding the hen house " at this point.

Honestly, I think the guy probably separates the two as much as he can, but Nvidia (as a whole and especially the CUDA group) doesn't want OpenCL to evolve faster than CUDA. Nvidia has dragged their feet on implementation. I'm not sure they have fully implemented OpenCL 2.0 yet. When Nvidia bought the Portland Compiler group they mothballed some OpenCL aspects of those products ( in favor of surprise CUDA). There were a ton of moves Nvidia made in 2013-2015 timeframe where the company decided to fully monetize CUDA. as strategic area.

Open Standards are a tricky thing. If you try to push for a standard before a critical mass of people are ready for it, then it usually stalls in these committees. There are lots of arguments about corner cases and the focus just isn't there. Too late and most folks have jumped into their own proprietary solutions they are reluctant to let go of. So there is a bit of Goldilock moment where they work well.

OpenGL started out somewhat related to Silicon Graphics backed standard. Not only involved to gatheing a group they were proactively pushing out an implementation the discussion could evolve around.

"openGL next" turned into Vulkan. Similar thing jump started that into motion past getting bogged down in committee. AMD just opened their whole Mantle implementation. "take it and mutate to what is best".

OpenCL didn't work out as well. Partially it was Apple's reaction to AMD and Nvidia wanting to go down the proprietary path with GPGPU. Microsoft was off on the side cooking up something separate on their own as usual. It was late getting started for CUDA had some traction and AMD was somewhat failing around. There was a tight coupling to LLVM (compiler) and Apple largely controlled that too. LLVM was also evolving. At some point it appears that there was kind of a vacuum of leadership as far as implementation. Apple started backpedaling. Nvidia was not in a hurry to get to OpenCL 2.0 ( shared/mapped memory). Mobile GPU vendors didn't really have much of a shared memory problem (since pragmatically all integrated graphics).


I suspect Apple got tired of herding cats and said this is too strategic, so we are doing our own and took up Metal. ( also I think there were some rumblings they didn't like OpenCL getting friendly with C++ also. I doubt Apple is the major driver of the LLVM C++ front end. ) Metal is more of a combo of Vulkan and a subset of OpenCL compute concepts narrowed down to primary what Apple needs. No committee means Apple could move fast. Essentially the joined Nvidia and Microsoft with conflicted goals inside of Khronos.


No way Apple can win that war on the desktop, but with mobile probably, so that will be their bargaining chip.

Microsoft is winning on their "desktop" just fine. Apple has enough critical mass to win in there subset also. They have the money and resources to do it. Metal is going to be optimized for iOS and macOS second but overtime it will evolve on both.


[quote ]
It's like the CPU wars except that there's nothing analogous to cross-compiling. It sucks.[/QUOTE]

There is cross compiling. OpenCL is cranking up the integration with SPIR ( and SYSCL)

OCL21_Eco_575px.png

https://www.anandtech.com/show/9039/khronos-announces-opencl-21-c-comes-to-opencl

As more platforms pick up Vulkan the compilation tools for OpenCL will improve too.
( Vulkan is a different committee working at a different pace. These two standards are not trying to tightly couple because that brings too many voters together with probably conflicting viewpoints but they aren't 100% decoupled either. )


On the macOS platform I suspect it will continue to lag as long as Apple doesn't help with Vulkan, OpenCL, or OpenGL. I think the neglect they are doing now will bite in the butt later, but for now it is working for them.
 
  • Like
Reactions: sd70mac and fairuz
Seems like a lot of people are being fussy over the lack of official support for Nvidia GPUs. CUDA drivers are available for macOS. If it’s really that bothersome then why not build a Hackintosh? That’ll way, you’ll be able to install whatever precious Nvidia GPU you want.

You will also get more in terms of customization (heck, even Ryzen CPUs work in macOS).
 
  • Like
Reactions: MarkC426
Yes agreed.
Anyone that has a serious use for Cuda probably has a macpro anyway, where the nvidia options are available :)
 
  • Like
Reactions: sd70mac
I had begun to dislike Apple because of iOS, user-unfriendly aggressive strategies, and their exponentially growing access to private data. However, I kept believing their hardware was better. With this step in not supporting Nvidia, I don't feel attracted to their hardware anymore. And those of you claiming that Nvidia doesn't support standards, please stop it. It's Apple who not only bashed OpenGL and OpenCL in favor of that custom nonsense called Metal, but even won't care to support Vulkan (Nvidia does).

The only Apple hardware I might be interested at this time would be a 15inch MacBook with the same weight as a current 13inch MBA (big display with maximum portability, not interested in a discrete GPU for that MB). But I don't hold my breath, I think Apple won't beat the LG Gram either: They fail in every end now: At the Pro, at the low-end, at the high-end... everywhere.

A shame.
[doublepost=1522524041][/doublepost]
Seems like a lot of people are being fussy over the lack of official support for Nvidia GPUs. CUDA drivers are available for macOS. If it’s really that bothersome then why not build a Hackintosh?
The Hackintosh is not the solution unless Apple allows them in the MacOS license.
[doublepost=1522525068][/doublepost]
And Apple wants to use OpenCL instead of CUDA so they go AMD.
Really? Then why MacOS is the only OS that doesn't have an OpenCL debugger? Apple has moved away from being a supporter of standards. They've been moving more and more into proprietary APIs all these years since Tim Cook is in charge. They care about Metal. They have zero interest in OpenGL and OpenCL (they mention them in this document just because they know that eGPU potential customers use standards rather than Metal, but Apple has no real interest on standards anymore). Note that they don't mention Vulkan, because they hope the potential customers of eGPU solutions are not using Vulkan yet.
 
Last edited:
It's a shame that Apple keeps thunderbolt as a moving target. No one is going to adopt it if they keep coming out with a view version and plugs. Reminds me of FireWire. They kept changing it and there was few manufacturers that supported it.
Try to go down to Best Buy and buy a thunderbolt 2 hard drive. Nope.
Next year when thunderbolt 4 is released with a new plug all those 3 devices will need a dongle and 10.14 will drop support for thunderbolt 3 for newer tech.
That's why sata and USB are so widely available.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.