Performance between MBPr-750m ± eGPU ± Win7

[TrumanLA]

Adobe CC Performance benefit

- L-13 / M-14 MBPr (750m) + eGPU for Adobe CC (Video editing, Premier) ​

• If you have some sort of expertise that allows you to speak confidently - please do....

• If you have any metrics of previous GPU tests, or, speed changes for rendering that'd be amazing:


(Premise: Everything is to describe the performance of a 750m MBPr + 1080 Ti as eGPU)

750m MBPr
750m MBPr vs 750m MBPr + eGPU in OS X
750m MBPr + eGPU in OS X vs 750m MBPr + eGPU in Windows 7​

Trying to assist a friend with the most precise opinion possible.
I'm hoping one of you has done this -- otherwise, I will have to set up the experiment
Thanks you guys. Appreciate any time used towards this!

 

[Yebubbleman]

The 750M rMBP will have Thunderbolt 2. Apple only officially supports eGPUs via Thunderbolt 3. Given the bandwidth differential between Thunderbolt 2 and Thunderbolt 3, this makes sense as Thunderbolt 3 is really the first time there's been enough of a throughput to make an eGPU really worth doing as you're otherwise limiting your bus speed by a considerable amount.

So, that being said, I wouldn't try your 750M-based rMBP with an eGPU as far as macOS is concerned. You MIGHT have better luck on the Windows 7 side of things, though I'd say that even then you're probably going to fight an uphill battle as far as driver support is concerned. With Windows 10, you might still have a better shot; but then the trick is to find a good eGPU solution that supports Thunderbolt 2 (of which there really weren't that many [most were proprietary like the one Razer made for its notebooks at that time]) and again, due to the throughput limitations of Thunderbolt 2, you probably won't find something that does what you want it to do well. It might serve nicely as a proof of concept, but to actually use, it will probably not work or suck.

If you really want to do the whole Mac+eGPU combo, I'd get a 13" MacBook Pro from 2016 or newer and then get one of Apple's recommended expansion chassis and one of their recommended graphics cards to throw into it. Or, if you want more flexibility and/or to only use the eGPU in Windows 10 on said 13" MacBook Pro from 2016 or newer, you can probably use any Thunderbolt 3 chassis and any graphics card and have it be fine so long as you install the right drivers for it.

 

[TrumanLA]

The 750M rMBP will have Thunderbolt 2. Apple only officially supports eGPUs via Thunderbolt 3. Given the bandwidth differential between Thunderbolt 2 and Thunderbolt 3, this makes sense as Thunderbolt 3 is really the first time there's been enough of a throughput to make an eGPU really worth doing as you're otherwise limiting your bus speed by a considerable amount.

So, that being said, I wouldn't try your 750M-based rMBP with an eGPU as far as macOS is concerned. You MIGHT have better luck on the Windows 7 side of things, though I'd say that even then you're probably going to fight an uphill battle as far as driver support is concerned. With Windows 10, you might still have a better shot; but then the trick is to find a good eGPU solution that supports Thunderbolt 2 (of which there really weren't that many [most were proprietary like the one Razer made for its notebooks at that time]) and again, due to the throughput limitations of Thunderbolt 2, you probably won't find something that does what you want it to do well. It might serve nicely as a proof of concept, but to actually use, it will probably not work or suck.

If you really want to do the whole Mac+eGPU combo, I'd get a 13" MacBook Pro from 2016 or newer and then get one of Apple's recommended expansion chassis and one of their recommended graphics cards to throw into it. Or, if you want more flexibility and/or to only use the eGPU in Windows 10 on said 13" MacBook Pro from 2016 or newer, you can probably use any Thunderbolt 3 chassis and any graphics card and have it be fine so long as you install the right drivers for it.

Two things that must be said:

1. Thank you! Beyond thank you. Ultra thank you. For your time - and for sharing your expertise which is beyond dispute, impeccable logic.

2. You are very knowledgeable. Usually I'll see an error or product knowledge in any hardware hypothesis. Nope. I'd say pretty much precise.


That said, for a thought experiment, let's consider the Cubix products and the use of (though I believe even 1080 Ti s would be far better than each quantity less than the total it could hold) -- the card which transfers the data (looks like a SAS interface) is far slower than the x16 Gen 3 vs Gen 2, which the Mac Pro's PCI bus is ... poses some thoughts that call in to question the necessity of every component having the same throughput as the device doing the work.

x16 Gen 2
x16 Gen 3

Basically, there's no way a SINGLE pci slot can transfer as much as 4 could.

In the simplest words, if I sent stephen hawking a text message that fit within 160 characters requesting he did some [integral differential calculus describing perturbation theory, effects and proof - for any combination of 4 local celestial bodies] ... I doubt his answer would be as small as mine.

Just as giving an uber driver your location and where you want to go can't be described as concisely as that compacted formulae. And I can do those ad infinitum; and if all a GPU needed was that ability to keep up with the speed that data flowed through the bus, GPU's would be as cheap as the plastic interface that connected them.

Anyway, I know I'm being absurd. I am exactly deploying 'reductio ad absurdum' -- because as the progenitor of that philosophical device understood in regarding it useful enough to deserve a name... it's a good method of evaluating thoughts.

 

[Yebubbleman]

Two things that must be said:

1. Thank you! Beyond thank you. Ultra thank you. For your time - and for sharing your expertise which is beyond dispute, impeccable logic.

2. You are very knowledgeable. Usually I'll see an error or product knowledge in any hardware hypothesis. Nope. I'd say pretty much precise.


That said, for a thought experiment, let's consider the Cubix products and the use of (though I believe even 1080 Ti s would be far better than each quantity less than the total it could hold) -- the card which transfers the data (looks like a SAS interface) is far slower than the x16 Gen 3 vs Gen 2, which the Mac Pro's PCI bus is ... poses some thoughts that call in to question the necessity of every component having the same throughput as the device doing the work.

x16 Gen 2
x16 Gen 3

Basically, there's no way a SINGLE pci slot can transfer as much as 4 could.

In the simplest words, if I sent stephen hawking a text message that fit within 160 characters requesting he did some [integral differential calculus describing perturbation theory, effects and proof - for any combination of 4 local celestial bodies] ... I doubt his answer would be as small as mine.

Just as giving an uber driver your location and where you want to go can't be described as concisely as that compacted formulae. And I can do those ad infinitum; and if all a GPU needed was that ability to keep up with the speed that data flowed through the bus, GPU's would be as cheap as the plastic interface that connected them.

Anyway, I know I'm being absurd. I am exactly deploying 'reductio ad absurdum' -- because as the progenitor of that philosophical device understood in regarding it useful enough to deserve a name... it's a good method of evaluating thoughts.

I'm not really sure what you were trying to say here. You're welcome, I guess?