Apple didn't have eGPUs on the table, even experimentally, until early this year. How much yowling would all you be doing if they had waited THAT long to release the 2016 MacBook Pro
You'd only need that if your use case was to run 2 5k displays plus the built-in LCD, pretty niche and the end result is that they are still WAY behind in GPU performance.
I get it, your view is it's ok that the GPU performance across the Mac range is poor so long as you can run 2 5k monitors + the internal display, pretty serious apologist view there and still no excuse for how crappy it is on the Mac Pro which has multiple GPUs.
and HOW much whining about "you even have to use an eGPU" would all you people who whine INCESSANTLY about even the most SIMPLE, PASSIVE, USB adapter be doing?!?
To run multiple 5k displays plus the built in LCD? Seems pretty reasonable to me.
Oh, and although the current version of NVidia GPUs is definitely ahead in performance, NVidia has already warned us that those gigantic performance games (for them, at least) are OVER.
They never said any such thing, all they said is the next Geforce GPUs won't be out for a long time, which is of course that the Turing architecture GPUs will be out toward the end of this year. I'm not sure why you're focussing on games, for a professional machine I'm not really that interested in gaming though I'm sure some people are.
So then what?
If what you say were actually true (it isn't) then when the next AMD GPUs come out and Apple integrates them we'll finally have Macs with GPU performance rivalling systems available in 2016. Hooray.
My only question is if it will be a discrete graphics card or a SOC like now?
Having a discrete GPU means you have the issue of non-unified memory, being able to create buffer that is both host and device accessible rather than having 2 copies and having to transfer over a bus to synchronize them (and choosing appropriate points to synchronize) is hugely advantageous. It's one of the big issues on PC architectures when you want to offload compute (graphics or general purpose) to the GPU. Consoles and most embedded and mobile systems don't suffer this but PCs do - there is of course the CPU/GPU dies that Intel and AMD offer to the PC market that alleviate this but those are relatively low performance compared to the discrete solutions.
And the net result is that they are years behind in GPU performance. Whatever you accept for their reasoning behind the decision the result is that the best offering from Apple is a LONG way from state of the art in terms of GPU performance and this isn't just laptops but also desktops.
It's just odd how you can't seem to acknowledge how far they are behind in GPU performance.
But Apple wanted to drive TWO 5K displays, PLUS the internal laptop display. NVidia couldn't do it. AMD could.
eGPU is one solution to that, multiple GPUs (like they already have in the Mac Pro) is another. But instead we get sub-par GPU performance on the Mac across the board, and by a wide margin too.
Well yes, they ignored the Mac Pro for many years, then came out with the trashcan replacement - in priciniple wasn't that bad of an idea - which they never updated.
If you think the NVidia reps weren't hanging around the Apple labs at the same time, you're mistaken. And if you think that Apple isn't smart enough to pick the best GPU, ALL things considered, you are sadly mistaken.
They certainly picked the slowest by a WIDE margin.
Apple chose the AMD GPUs because they weren't building a gaming machine
They weren't building anything that requires significant GPU compute capability at all, if they were they wouldn't have picked such a poor performing part relative to the rest of the market.
the AMD GPUs provided more multiple display capability than the NVidia ones did. That's not speculation, that's according to Apple.
I run dual 4k display displays off an nvidia GPU with no problem, you can do more than that too. So I'm not sure what you're talking about but ultimately the performance of the GPU Apple chose is a LONG way from "state of the art".
But honestly, the iMac Pro isn't exactly a weenie computer. Yes, there are more powerful; but it isn't laughable...
The problem with Apple's current strategy is there's no upgrade option, it's a "throw it all out and start again" situation. I bought into the Mac Pro trashcan thing because - despite the idiotic marketing of the rotating base (ever wonder why they never show that with anything, including the power cord, actually plugged in?!) - the concept was that this core bit would be replaceable and all your peripherals and mass storage would hang off it on high speed interconnect. MacOS has always had great support for transitioning user data between systems when upgrading so it didn't seem like a problem. But, like they did to us with the old Mac Pro, they abandoned it and screwed us over again.
The AMD RX Vega Series GPUs were released on August 14, 2017.
And the top end part there is still slower - by a WIDE margin - than the GTX 1080, TitanX or P6000 (if you want to go highend professional) GPUs that were released back in mid-late 2016! If your definition of "state of the art" is simply the time that it's released then sure but if you're looking at performance then it's WAY off.
So what do people use portable workstations like this - which have been around for decades - for? If not CAD/CAM/CAE, video editing, audio editing, scientific computing, etc... then what are they used for?
Professional gaming.
On these portable workstations with Quadro GPUs? No.
This. My wife's 2010 MacBook Pro was starting to act flakey and she wanted a new one, but all the new ones are a step backwards. No MagSafe port, no multiple USB boards, no built-in CDROM reader, no headphone jack, the new keyboards are not fun to type on/keep clean, and really not much improvement in RAM, CPU or graphics either.
That's precisely why Apple still sell the old 2015 Macbook Pro (see the third option here). The new one has been absolutely woeful, so bad in fact that they still sell the last decent one from 3 years ago. Sure it lacks the CD/DVD drive but at least you can plug in any USB one and don't need a USB-A to USB-C converter plug to make it work.
So what do people use portable workstations like this - which have been around for decades - for? If not CAD/CAM/CAE, video editing, audio editing, scientific computing, etc... then what are they used for?
Also I was doing CAD on workstations with 8gb of ram. You do not need 128gb of ram to run cad programs.
There were a lot of people doing CAD on workstations with 1mb of RAM too, therefore 8gb is massive overkill? You'd think that 640k Bill Gates quote has had enough exposure that people would have got the point of it by now, obviously not.
Needing power sucking CPUs and multiple GPUs, this laptop does not solve that problem.
They use desktop-grade CPUs rather than low-power portable ones and if you really need it you can expand the GPU capability with an eGPU for those times that you need it.
So again, what's the point?
Oh no you can buy a laptop with 128GB of RAM, what a terrible thing! What's the point of complaining about it? If you don't need it don't buy it, if nobody needs it nobody will buy and it will go away and you can stop whining about the existence of something you don't want or need.
How many computers are out there that can run OpenGL 4.6 in a meaningful way today, Apple or not?
A lot of them:
Hardware support: Nvidia GeForce 400 series and newer, Intel Haswell and newer, AMD Radeon HD 5000 Series and newer Driver support:
Mesa on Linux mostly supports OpenGL 4.6 by Mesa 18.0 for Intel Haswell+, AMD Radeon GCN, Nvidia Kepler+
NVIDIA GeForce 397.31 Graphics Driver on Windows 7, 8, 10
AMD Adrenalin 18.4.1 Graphics Driver on Windows 7 SP1, 10/p>
But that's 4.6, released only last year. macOS is stuck on 4.1 which was released in 2010 (even though if you install Windows or Linux on a capable Mac you can get much more recent OpenGL (and Vulkan) support.
But what if you did? That's the question I'm asking, obviously the answer to your question is that the gap between what you pay for perpetual and what you pay for subscription would be narrower the less often you update.
I'd be willing to bet that Adobe based the subscription price on what the average user was willing to spend to stay mostly up to date.
Probably, so the result is the average customer is no worse off financially but they are kept up to date.
Tech that has proven to be reliable and working over many years is not "very old", it is called VERY UBIQUITOUS. Just like your bicycle.
It can be both. OpenGL 4.1 is very old, all the platforms that support more recent versions of OpenGL also support 4.1 (even a Macbook pro from 2015 with a Radeon 370 that only supports OpenGL 4.1 on macOS can support OpenGL 4.3 if you run Windows or Linux on it). So if that's what you want to target that's fine but Apple does not support more recent versions, the most recent it supports is 4.1 (which is very old and doesn't support many of the new and more modern features of later versions of OpenGL).
This argument could be applied to directx too but microsoft still shipped directx with windows.
Of course they do, they are the vendor of DirectX and the platforms on which it is supported so it makes sense for them to ship it, the actual hardware implementation is still supplied by the hardware vendor though.
It certainly doesn't run better for machine learning.
If you want performance for machine learning then now the best approach is really to exploit volta tensor cores using CUDA, so target platforms are Windows or Linux (locally or cloud). Metal is certainly not the best approach for machine learning performance unless you're limited to macOS with mac supported hardware.
I remember when Microsoft dropped OpenGL from the Windows platform.
Well the implementation generally came in the form of an MCD anyway and the obvious progression was to move to an ICD model (the same as is done on Linux) especially as hardware became more complex so vendors could implement the OpenGL spec for their own hardware.
It being Apple, they could throw a bigger fit and forbid it, but at least it's possible that OS dropping support may mean nothing in practice.
Apple doesn't use an ICD model for OpenGL, that's why the best OpenGL support you can get on a Mac (even before the 2015 introduction of Metal on the Mac) is 4.1 from 2010. Apple always lagged in OpenGL support, it's never been very good.
Not really, the Volta architecture is available and while it's a bit difficult you can certainly get a hold of a TitanV the issue really is that it's performance for gaming isn't really mindblowing when compared to a current TitanX so there's little reason to upgrade, its performance for workloads that can leverage the tensor core is incredible but very few games are going to be able to do much with that.
People doing machine learning (or things that leverage fast inferencing) didn't care about gaming-focussed things like single-pass stereo that Pascal introduced and gamers aren't interested in the tensor cores that Volta introduced. Turing may have something in store for gamers that other markets aren't interested in and that will drive a new generation of Geforces. Everyone would be complaining if they released a Volta-based Geforce refresh that did nothing for gamers so what's the point? Wait until there's some innovation.
No, hardware vendors have to re-implement OpenGL in their custom drivers because Microsoft refused to do it
Of course they have to do that, how do you expect Microsoft to write an implementation of OpenGL for every vendor's hardware? It's exactly the same on Linux, the vendor provides the OpenGL driver which is the implementation of OpenGL for their hardware. This is also why the free software drivers lag the proprietary ones, because the free software devs are trying to write an OpenGL implementation for hardware that they don't have the full specifications for.
Well vendors supply support for OpenGL and Vulkan in their drivers for Windows, Apple restricts that which is why OpenGL support is stuck at the 2010 release (5 years before Metal was even available on MacOS) and why Vulkan is only available as an abstraction over Metal and not as a native driver library.
My take on it: Apple is setting itself up to eat crow a year down the road and humbly slither back into the Vulkan/OpenCL camp.
I would say developers are likely to target Vulkan via MoltenVK so they don't need to write Apple-specific code when Apple might simply abandon Metal anyway. Apple has had poor support for GPU acceleration even before Metal came out on the desktop, the latest supported version was OpenGL 4.1 in mid-2010 and Metal didn't come to Macs until 5 years later in mid-2015. Apple has never been very good at GPU support.
And you stay up to date, what was the cost to do that on the perpetual license system? Even assuming you bought the full version and then upgraded every year? It was a lot more than that.
Also helps line developer's pockets, as Adobe CS4 is only 32 bit and when it stops being supported by OSX, you pretty much have to pay Adobe whatever it asks for the current version (I believe that take $1000 cash or monthly installments on par with a compact car lease payments)
No that's part of the benefit of the subscription model, like it or not if the underlying platform change forces developers into a more modern architecture (personally my view is that would be Vulkan - MoltenVK on MacOS - rather than Metal) then the software users aren't hit with a big bill to upgrade all their software.
Slashdot Mirror
Apple didn't have eGPUs on the table, even experimentally, until early this year. How much yowling would all you be doing if they had waited THAT long to release the 2016 MacBook Pro
You'd only need that if your use case was to run 2 5k displays plus the built-in LCD, pretty niche and the end result is that they are still WAY behind in GPU performance.
I get it, your view is it's ok that the GPU performance across the Mac range is poor so long as you can run 2 5k monitors + the internal display, pretty serious apologist view there and still no excuse for how crappy it is on the Mac Pro which has multiple GPUs.
and HOW much whining about "you even have to use an eGPU" would all you people who whine INCESSANTLY about even the most SIMPLE, PASSIVE, USB adapter be doing?!?
To run multiple 5k displays plus the built in LCD? Seems pretty reasonable to me.
Oh, and although the current version of NVidia GPUs is definitely ahead in performance, NVidia has already warned us that those gigantic performance games (for them, at least) are OVER.
They never said any such thing, all they said is the next Geforce GPUs won't be out for a long time, which is of course that the Turing architecture GPUs will be out toward the end of this year. I'm not sure why you're focussing on games, for a professional machine I'm not really that interested in gaming though I'm sure some people are.
So then what?
If what you say were actually true (it isn't) then when the next AMD GPUs come out and Apple integrates them we'll finally have Macs with GPU performance rivalling systems available in 2016. Hooray.
My only question is if it will be a discrete graphics card or a SOC like now?
Having a discrete GPU means you have the issue of non-unified memory, being able to create buffer that is both host and device accessible rather than having 2 copies and having to transfer over a bus to synchronize them (and choosing appropriate points to synchronize) is hugely advantageous. It's one of the big issues on PC architectures when you want to offload compute (graphics or general purpose) to the GPU. Consoles and most embedded and mobile systems don't suffer this but PCs do - there is of course the CPU/GPU dies that Intel and AMD offer to the PC market that alleviate this but those are relatively low performance compared to the discrete solutions.
And the net result is that they are years behind in GPU performance. Whatever you accept for their reasoning behind the decision the result is that the best offering from Apple is a LONG way from state of the art in terms of GPU performance and this isn't just laptops but also desktops.
It's just odd how you can't seem to acknowledge how far they are behind in GPU performance.
But Apple wanted to drive TWO 5K displays, PLUS the internal laptop display. NVidia couldn't do it. AMD could.
eGPU is one solution to that, multiple GPUs (like they already have in the Mac Pro) is another. But instead we get sub-par GPU performance on the Mac across the board, and by a wide margin too.
Well yes, they ignored the Mac Pro for many years, then came out with the trashcan replacement - in priciniple wasn't that bad of an idea - which they never updated.
If you think the NVidia reps weren't hanging around the Apple labs at the same time, you're mistaken. And if you think that Apple isn't smart enough to pick the best GPU, ALL things considered, you are sadly mistaken.
They certainly picked the slowest by a WIDE margin.
Apple chose the AMD GPUs because they weren't building a gaming machine
They weren't building anything that requires significant GPU compute capability at all, if they were they wouldn't have picked such a poor performing part relative to the rest of the market.
the AMD GPUs provided more multiple display capability than the NVidia ones did. That's not speculation, that's according to Apple.
I run dual 4k display displays off an nvidia GPU with no problem, you can do more than that too. So I'm not sure what you're talking about but ultimately the performance of the GPU Apple chose is a LONG way from "state of the art".
I don't, but they've screwed us over twice now...not a great record.
But honestly, the iMac Pro isn't exactly a weenie computer. Yes, there are more powerful; but it isn't laughable...
The problem with Apple's current strategy is there's no upgrade option, it's a "throw it all out and start again" situation. I bought into the Mac Pro trashcan thing because - despite the idiotic marketing of the rotating base (ever wonder why they never show that with anything, including the power cord, actually plugged in?!) - the concept was that this core bit would be replaceable and all your peripherals and mass storage would hang off it on high speed interconnect. MacOS has always had great support for transitioning user data between systems when upgrading so it didn't seem like a problem. But, like they did to us with the old Mac Pro, they abandoned it and screwed us over again.
The AMD RX Vega Series GPUs were released on August 14, 2017.
And the top end part there is still slower - by a WIDE margin - than the GTX 1080, TitanX or P6000 (if you want to go highend professional) GPUs that were released back in mid-late 2016! If your definition of "state of the art" is simply the time that it's released then sure but if you're looking at performance then it's WAY off.
So what do people use portable workstations like this - which have been around for decades - for? If not CAD/CAM/CAE, video editing, audio editing, scientific computing, etc... then what are they used for?
Professional gaming.
On these portable workstations with Quadro GPUs? No.
This. My wife's 2010 MacBook Pro was starting to act flakey and she wanted a new one, but all the new ones are a step backwards. No MagSafe port, no multiple USB boards, no built-in CDROM reader, no headphone jack, the new keyboards are not fun to type on/keep clean, and really not much improvement in RAM, CPU or graphics either.
That's precisely why Apple still sell the old 2015 Macbook Pro (see the third option here). The new one has been absolutely woeful, so bad in fact that they still sell the last decent one from 3 years ago. Sure it lacks the CD/DVD drive but at least you can plug in any USB one and don't need a USB-A to USB-C converter plug to make it work.
So what do people use portable workstations like this - which have been around for decades - for? If not CAD/CAM/CAE, video editing, audio editing, scientific computing, etc... then what are they used for?
This also serves to hide the fact their employers *expect* them to work anywhere, all the time.
Sounds like you're working for the wrong company.
Why are people doing CAD, etc on laptops?
Because they want/need portable workstations.
Also I was doing CAD on workstations with 8gb of ram. You do not need 128gb of ram to run cad programs.
There were a lot of people doing CAD on workstations with 1mb of RAM too, therefore 8gb is massive overkill? You'd think that 640k Bill Gates quote has had enough exposure that people would have got the point of it by now, obviously not.
Needing power sucking CPUs and multiple GPUs, this laptop does not solve that problem.
They use desktop-grade CPUs rather than low-power portable ones and if you really need it you can expand the GPU capability with an eGPU for those times that you need it.
So again, what's the point?
Oh no you can buy a laptop with 128GB of RAM, what a terrible thing! What's the point of complaining about it? If you don't need it don't buy it, if nobody needs it nobody will buy and it will go away and you can stop whining about the existence of something you don't want or need.
How many computers are out there that can run OpenGL 4.6 in a meaningful way today, Apple or not?
A lot of them:
Hardware support: Nvidia GeForce 400 series and newer, Intel Haswell and newer, AMD Radeon HD 5000 Series and newer
Driver support:
Mesa on Linux mostly supports OpenGL 4.6 by Mesa 18.0 for Intel Haswell+, AMD Radeon GCN, Nvidia Kepler+
NVIDIA GeForce 397.31 Graphics Driver on Windows 7, 8, 10
AMD Adrenalin 18.4.1 Graphics Driver on Windows 7 SP1, 10/p>
But that's 4.6, released only last year. macOS is stuck on 4.1 which was released in 2010 (even though if you install Windows or Linux on a capable Mac you can get much more recent OpenGL (and Vulkan) support.
But what if you didn't fork out for every update?
But what if you did? That's the question I'm asking, obviously the answer to your question is that the gap between what you pay for perpetual and what you pay for subscription would be narrower the less often you update.
I'd be willing to bet that Adobe based the subscription price on what the average user was willing to spend to stay mostly up to date.
Probably, so the result is the average customer is no worse off financially but they are kept up to date.
Tech that has proven to be reliable and working over many years is not "very old", it is called VERY UBIQUITOUS. Just like your bicycle.
It can be both. OpenGL 4.1 is very old, all the platforms that support more recent versions of OpenGL also support 4.1 (even a Macbook pro from 2015 with a Radeon 370 that only supports OpenGL 4.1 on macOS can support OpenGL 4.3 if you run Windows or Linux on it). So if that's what you want to target that's fine but Apple does not support more recent versions, the most recent it supports is 4.1 (which is very old and doesn't support many of the new and more modern features of later versions of OpenGL).
This argument could be applied to directx too but microsoft still shipped directx with windows.
Of course they do, they are the vendor of DirectX and the platforms on which it is supported so it makes sense for them to ship it, the actual hardware implementation is still supplied by the hardware vendor though.
It certainly doesn't run better for machine learning.
If you want performance for machine learning then now the best approach is really to exploit volta tensor cores using CUDA, so target platforms are Windows or Linux (locally or cloud). Metal is certainly not the best approach for machine learning performance unless you're limited to macOS with mac supported hardware.
I remember when Microsoft dropped OpenGL from the Windows platform.
Well the implementation generally came in the form of an MCD anyway and the obvious progression was to move to an ICD model (the same as is done on Linux) especially as hardware became more complex so vendors could implement the OpenGL spec for their own hardware.
It being Apple, they could throw a bigger fit and forbid it, but at least it's possible that OS dropping support may mean nothing in practice.
Apple doesn't use an ICD model for OpenGL, that's why the best OpenGL support you can get on a Mac (even before the 2015 introduction of Metal on the Mac) is 4.1 from 2010. Apple always lagged in OpenGL support, it's never been very good.
Not really, the Volta architecture is available and while it's a bit difficult you can certainly get a hold of a TitanV the issue really is that it's performance for gaming isn't really mindblowing when compared to a current TitanX so there's little reason to upgrade, its performance for workloads that can leverage the tensor core is incredible but very few games are going to be able to do much with that.
People doing machine learning (or things that leverage fast inferencing) didn't care about gaming-focussed things like single-pass stereo that Pascal introduced and gamers aren't interested in the tensor cores that Volta introduced. Turing may have something in store for gamers that other markets aren't interested in and that will drive a new generation of Geforces. Everyone would be complaining if they released a Volta-based Geforce refresh that did nothing for gamers so what's the point? Wait until there's some innovation.
No, hardware vendors have to re-implement OpenGL in their custom drivers because Microsoft refused to do it
Of course they have to do that, how do you expect Microsoft to write an implementation of OpenGL for every vendor's hardware? It's exactly the same on Linux, the vendor provides the OpenGL driver which is the implementation of OpenGL for their hardware. This is also why the free software drivers lag the proprietary ones, because the free software devs are trying to write an OpenGL implementation for hardware that they don't have the full specifications for.
Well vendors supply support for OpenGL and Vulkan in their drivers for Windows, Apple restricts that which is why OpenGL support is stuck at the 2010 release (5 years before Metal was even available on MacOS) and why Vulkan is only available as an abstraction over Metal and not as a native driver library.
My take on it: Apple is setting itself up to eat crow a year down the road and humbly slither back into the Vulkan/OpenCL camp.
I would say developers are likely to target Vulkan via MoltenVK so they don't need to write Apple-specific code when Apple might simply abandon Metal anyway. Apple has had poor support for GPU acceleration even before Metal came out on the desktop, the latest supported version was OpenGL 4.1 in mid-2010 and Metal didn't come to Macs until 5 years later in mid-2015. Apple has never been very good at GPU support.
That's $600 per year. Every year.
And you stay up to date, what was the cost to do that on the perpetual license system? Even assuming you bought the full version and then upgraded every year? It was a lot more than that.
Also helps line developer's pockets, as Adobe CS4 is only 32 bit and when it stops being supported by OSX, you pretty much have to pay Adobe whatever it asks for the current version (I believe that take $1000 cash or monthly installments on par with a compact car lease payments)
No that's part of the benefit of the subscription model, like it or not if the underlying platform change forces developers into a more modern architecture (personally my view is that would be Vulkan - MoltenVK on MacOS - rather than Metal) then the software users aren't hit with a big bill to upgrade all their software.