2013 15" Macbook Pro + GTX780Ti@16Gbps Thunderbolt2 eGPU implementation

[thundersteele]

So I'm in a bit of a pickle.

I'm going to pick up an rMBP on Friday, but now I'm unsure what I want to go with. I was going to get the higher end model solely because I want to occasionally boot into Windows and play PC games, but the higher end model is ~$700 more than the base.

Could I just take that $700 and invest it into a set up like this? The card the guy in the video is using is obviously going to run me more than that, but if I do something like a 780 or even 760 I could probably squeeze that into my budget.

Obviously just getting the upgrade will be a bit simpler, but if it results in me getting much better performance when I'm sitting at my desk anyway, I'd rather go with the set up like in the video.

Anyone have an opinion?

Yes. I think it is a good idea, but you should only do it if you are reasonably tech savvy. The technology is rather fresh and not too well tested, so you might have to overcome some hurdles here and there (e.g. OSX might not support all GPUs). Also the cheaper enclosures require some extra work (see post #3 in the T|I thread).

Furthermore yjchua already pointed out that the $600 buys you more than just a faster GPU. Some of those upgrades you might want to get anyways (e.g. the RAM or larger (and faster) SSD).

Finally the external solution is still very expensive. Imho if you go for this then you want full flexibility, since that is the big advantage of the external GPU. But that would probably also mean going for the $979 enclosure so that you are not limited by size or power. From there of course you can start with a cheap 750 TI and upgrade to a nice Maxwell GPU in a year, but the upfront cost is still substantial.

I can't make the decision for you... just giving you some points to think about.

 

[nando4]

So I'm in a bit of a pickle.

I'm going to pick up an rMBP on Friday, but now I'm unsure what I want to go with. I was going to get the higher end model solely because I want to occasionally boot into Windows and play PC games, but the higher end model is ~$700 more than the base.

Could I just take that $700 and invest it into a set up like this? The card the guy in the video is using is obviously going to run me more than that, but if I do something like a 780 or even 760 I could probably squeeze that into my budget.

Obviously just getting the upgrade will be a bit simpler, but if it results in me getting much better performance when I'm sitting at my desk anyway, I'd rather go with the set up like in the video.

Anyone have an opinion?

That $700 can get you quite a lot of Thunderbolt eGPU goodness. However, the $979 Sonnet Echo Express III-D used in this demo is too pricey to make it viable for most people. It can be replaced by a US$399-16Gbps or US$199-10Gbps alternative. The details covered at http://forum.techinferno.com/diy-e-...sonnet-echo-express-iii-d-win8.html#post91184 sectioned as:

• How to use a US$399 16Gbps Thunderbolt 2 enclosure instead (Sonnet Echo Express SEL)
• How to use a US$199 10Gbps Thunderbolt 1 enclosure instead (Firmtek Thundertek/PX)
• Iris Pro 15" Macbook is preferred to the GT750M model for NVidia eGPU purposes
• A GTX750Ti eGPU beats a GT750M and scales up to a Titan - recommend iTX GTX670/760 or GTX770

 

[arnamak]

That $700 can get you quite a lot of Thunderbolt eGPU goodness. However, the $979 Sonnet Echo Express III-D used in this demo is too pricey to make it viable for most people. It can be replaced by a US$499, US$399 or even US$199 alternative, each giving less bandwidth. The details covered at http://forum.techinferno.com/diy-e-...sonnet-echo-express-iii-d-win8.html#post91184 sectioned as:

• How to use a US$499 20Gbps Thunderbolt 2 enclosure instead (Sonnet Echo Express SE II)
• How to use a US$399 16Gbps Thunderbolt 2 enclosure instead (Sonnet Echo Express SEL)
• How to use a US$199 10Gbps Thunderbolt 1 enclosure instead (Firmtek Thundertek/PX)
• Iris Pro 15" Macbook is preferred to the GT750M model for NVidia eGPU purposes

I just want to make sure I'm understanding this at a basic level. For now, let's assume I'm referring to the $499 alternative listed here, and that any time I have the eGPU plugged in I would be running Windows. From what I understand, I put a GPU into the enclosure, connect it to power, and connect it to my macbook via a thunderbolt cable, correct?

Now lets say I'm using an external display, I don't see an HDMI port on the back of this enclosure...does that mean I'd have the eGPU connected to my MBP via thunderbolt and then a display plugged into my MBP as well? Wouldn't it make more sense to plug a display directly into the GPU?

Is a 750Ti going to get me drastically different performance via these means vs a 750m? When I say drastically, I mean will I be able to push year old games in 1080p at high settings and 60FPS?

Sorry I'm directing all of my questions at you, but you seem to have experience with a similar setup.

Probably the most important question I have, since I'm almost positive I'll go with a set up like this. At what price point does an upgraded GPU hit a ceiling? Meaning, Thunderbolt is limited in how much data can pass through it...So I'm assuming a GTX 780 wouldn't be any better than a 760 due to those limitations, but I don't know if a 760 is also overkill in that regard. I guess my question is, how high end of a GPU can I get?

 

[nando4]

I just want to make sure I'm understanding this at a basic level. For now, let's assume I'm referring to the $499 alternative listed here, and that any time I have the eGPU plugged in I would be running Windows. From what I understand, I put a GPU into the enclosure, connect it to power, and connect it to my macbook via a thunderbolt cable, correct?

Now lets say I'm using an external display, I don't see an HDMI port on the back of this enclosure...does that mean I'd have the eGPU connected to my MBP via thunderbolt and then a display plugged into my MBP as well? Wouldn't it make more sense to plug a display directly into the GPU?

Is a 750Ti going to get me drastically different performance via these means vs a 750m? When I say drastically, I mean will I be able to push year old games in 1080p at high settings and 60FPS?

Sorry I'm directing all of my questions at you, but you seem to have experience with a similar setup.

Probably the most important question I have, since I'm almost positive I'll go with a set up like this. At what price point does an upgraded GPU hit a ceiling? Meaning, Thunderbolt is limited in how much data can pass through it...So I'm assuming a GTX 780 wouldn't be any better than a 760 due to those limitations, but I don't know if a 760 is also overkill in that regard. I guess my question is, how high end of a GPU can I get?

A GT750M is a GK107 chip (384 shaders) is clocked at up to 967 MHz plus boost. A GTX750Ti (Keplar) is also 384 shaders clocked up to 1058Mhz. We can say that it's cousin, a GTX750Ti (Maxwell), as an eGPU is a slightly faster 75W external alternative to the GT750M dGPU. Good to have a basis for comparison looking to purchase a Iris Pro only 13"/15" MBP.

While a GTX750Ti is already ahead of the game, why stop there? You can see that that at 20Gbps, the pcie bandwidth approaches desktop-levels having minimal impact on gaming. Higher-end NVidia GPUs that will scale up in performance.

For example, the GTX780Ti used in this article gives premium performance at a premium price. However, I'm reluctant to recommend that as a solution for anybody looking for the sweet spot for value.

Instead, I'd suggest a mini ITX GTX670 (Asus) or GTX760 (Asus/MSI) to give a more compact configuration when being used with the $499 Sonnet Echo Express SE II enclosure + external 8x-to-x16 pcie riser. Or if wanting higher end performance, a GTX770 is a great value choice.

 

[Jmx]

MacBook Pro running an NVIDIA GTX 780 Ti over Thunderbolt 2

There's hope for connecting external GPU's over Thunderbolt after all.
In this example he uses Win8 on his MBP probably due to drivers but maybe it already works in OS X or at least in time it will.
For gaming he doesn't seem to get 100% GPU, probably TB2 is the bottleneck transferring data back to the computer. But for 3D GPU rendering i guess that won't be any problem. Personally i just want the external GPU's for 3D rendering (CUDA) so this gave me hope for the future

 

[goMac]

It's a very impressive demo. Still has some technical rough edges, but definitely shows that it's workable and very usable.

The next rev of Thunderbolt is supposed to get even faster, so if Thunderbolt is the bottle neck here, it should be entirely a non-issue next revision of Thunderbolt.

The amazing thing here is that this shows the nMP may have GPU expandability far beyond the cMP. Four way Crossfire is certainly on the table at this point.

 

[mcnallym]

http://www.eoshd.com/content/10908/thunderbolt-external-nvidia-graphics-comes-to-macbook-air-sort-of

Has been around since July last year

Has only ever been that worked with Windows ( Apple doesn't support with OSX ) and so far have only seen properly with Nvidia GPU's.

At the moment I believe is that Intel don't support GPU's on TB, which doesn't mean that it won't work, however I suspect that Apple won't support in OSX until Intel does.

 

[MH01]

This does show what he potential of TB might be in the future.

----------

It's a very impressive demo. Still has some technical rough edges, but definitely shows that it's workable and very usable.

The next rev of Thunderbolt is supposed to get even faster, so if Thunderbolt is the bottle neck here, it should be entirely a non-issue next revision of Thunderbolt.

The amazing thing here is that this shows the nMP may have GPU expandability far beyond the cMP. Four way Crossfire is certainly on the table at this point.

Ahh the irony that mac fans looked down on PC users who built 4xcrossfire / 4x sli systems as ridiculous

 

[arnamak]

A GT750M is a GK107 chip (384 shaders) is clocked at up to 967 MHz plus boost. A GTX750Ti (Keplar) is also 384 shaders clocked up to 1058Mhz. We can say that it's cousin, a GTX750Ti (Maxwell), as an eGPU is a slightly faster 75W external alternative to the GT750M dGPU. Good to have a basis for comparison looking to purchase a Iris Pro only 13"/15" MBP.

While a GTX750Ti is already ahead of the game, why stop there? You can see that that at 20Gbps, the pcie bandwidth approaches desktop-levels having minimal impact on gaming. Higher-end NVidia GPUs that will scale up in performance.

For example, the GTX780Ti used in this article gives premium performance at a premium price. However, I'm reluctant to recommend that as a solution for anybody looking for the sweet spot for value.

Instead, I'd suggest a mini ITX GTX670 (Asus) or GTX760 (Asus/MSI) to give a more compact configuration when being used with the $499 Sonnet Echo Express SE II enclosure + external 8x-to-x16 pcie riser. Or if wanting higher end performance, a GTX770 is a great value choice.

Much appreciated, thank you.

 

[MacVidCards]

The next rev of Thunderbolt is supposed to get even faster, so if Thunderbolt is the bottle neck here, it should be entirely a non-issue next revision of Thunderbolt.

The amazing thing here is that this shows the nMP may have GPU expandability far beyond the cMP. Four way Crossfire is certainly on the table at this point.

The truly amazing thing is how long everyone has been kept happy with "the GOOD version of Thunderbolt is RIGHT AROUND THE CORNER"

Been going on since day one.

And here comes Godot !!!!

 

[N19h7m4r3]

Ahh the irony that mac fans looked down on PC users who built 4xcrossfire / 4x sli systems as ridiculous

They do? Why would anyone bother with that? I say go for it. Especially now that you can hack Crysis to run in 8K.

 

[joe-h2o]

The truly amazing thing is how long everyone has been kept happy with "the GOOD version of Thunderbolt is RIGHT AROUND THE CORNER"

Been going on since day one.

And here comes Godot !!!!

Look, we're sorry your business model is drying up because of the design of the nMP, but acting like a petulant teenager about it at every opportunity just makes you look foolish.

 

[MH01]

The truly amazing thing is how long everyone has been kept happy with "the GOOD version of Thunderbolt is RIGHT AROUND THE CORNER"

Been going on since day one.

And here comes Godot !!!!

While TB is limited at the moment to take advantage of external GPU's, it does show what the potential brings.

Frankly the main reason I upgrade my Macbook Pro or Alienware is the GPU, now having the Ability to run full desktop GPUs on my laptop while I "dock" is damn awesome! Not many games are CPU limited these days. A Macbook pro retina for travel and work during the day, and high FPS gaming via a monitor at night would be awesome.

 

[N19h7m4r3]

While TB is limited at the moment to take advantage of external GPU's, it does show what the potential brings.

Frankly the main reason I upgrade my Macbook Pro or Alienware is the GPU, now having the Ability to run full desktop GPUs on my laptop while I "dock" is damn awesome! Not many games are CPU limited these days. A Macbook pro retina for travel and work during the day, and high FPS gaming via a monitor at night would be awesome.

Aye, also the next version is rated for 40Gb/s. That's a hell of a lot for a little cable. I've only wanted it for mass storage and RAIDS, and it's fantastic. Mix in display port 1.3 and imagine also the bandwidth for single stream linked 4K displays.

The potential keeps expanding, and there's constantly more to it. I don't see why anyone would want to disregard helpful technology that'll expand people's options and productivity more.

I only wish Intel would push it more, along with other motherboard manufacturers. Having even 2 current TB ports would be very beneficial.

 

[CASLondon]

Look, we're sorry your business model is drying up because of the design of the nMP, but acting like a petulant teenager about it at every opportunity just makes you look foolish.

We're sorry, but in the meantime we will clutter up a Mac Pro forum with trolling posts about MacBook Pros and thunderbolt

 

[joe-h2o]

We're sorry, but in the meantime we will clutter up a Mac Pro forum with trolling posts about MacBook Pros and thunderbolt

I'm sorry, I wasn't aware that a single proof of concept story that demonstrated a 780Ti running at 85-90% performance over thunderbolt in a forum that contains a lot of talk about issues relating to the problem of GPU expansion for the Mac Pro in the future constitutes "cluttering up".

Given that the MBP in question has Thunderbolt 2, which the nMP also has, I assume the original poster assumed that people might be interested that it was possible to do this.

Those who are not interested in it, or who are annoyed that Apple have decided to go with a design that doesn't dovetail with his business model, are free to ignore it rather than attempt to point out to everyone how much it all sucks.

MacVidCards used to be a useful source of information, but lately has just been posting what effectively amounts to trolling because he dislikes the nMP and thunderbolt and boy does he want you to know it!

We get it, you think thunderbolt sucks and think the nMP should have been a giant box with a bunch of PCI-e slots, it doesn't mean you have to keep harping on about it in every thread.

Edit, also I'd like you to quote the specific part of the OP's post that constitutes "trolling", since you called it a "trolling post". I can wait.

 

[CASLondon]

Edit, also I'd like you to quote the specific part of the OP's post that constitutes "trolling", since you called it a "trolling post". I can wait.

Sure, genius, since the part I'm calling a "trolling post" is not the original post but yours - the hint is that I quoted your own words casting assumptions about a poster's motivation/agenda well beyond that justified by the words he actually wrote. Calling someone a "petulant teenager" for raising an eyebrow or making a skeptical quip is not really justified, and is why I called it a troll post.

I could give two craps about thunderbolt being the second coming or not, since I own both Tbolt MBP and cMP. I'm not in that debate

I could also give two craps about taking sides in some imaginary debate between Team FutureVIsion and Team PathSkeptic

Most people have been aware of external thunderbolt expansion chassis like those made by Magma, etc, for some time. If the MP forum is going to get taking over by this silliness, well, its a shame.

 

[whooleytoo]

While TB is limited at the moment to take advantage of external GPU's, it does show what the potential brings.

Frankly the main reason I upgrade my Macbook Pro or Alienware is the GPU, now having the Ability to run full desktop GPUs on my laptop while I "dock" is damn awesome! Not many games are CPU limited these days. A Macbook pro retina for travel and work during the day, and high FPS gaming via a monitor at night would be awesome.

Yeah. My biggest issue with Macs these days is the relatively wimpy & non-upgradable GPUs. I'd LOVE it if Apple (or anyone) would make a decent dock with a GPU TB slot. It'd really help to make a no-compromise laptop. A nice, light laptop with low power requirements on-the-go, and a high-performance gaming machine at home.

 

[EpicBlob]

I've had my Mac Mini connected to a 670 GPU via thunderbolt and ViDock for over a year now, working in both Windows 8 and OS X. This setup in the article would work in OS X as well (just edit a few kext files and its 100% working).

This is definitely something I hope continues on. As laptops become more portable and desktops slowly dying off, it only seems logical to have a laptop paired with a dock as your complete system, doing easy work on the road and plugging it in at home for a massive performance boost.

 

[joe-h2o]

Sure, genius, since the part I'm calling a "trolling post" is not the original post but yours - the hint is that I quoted your own words casting assumptions about a poster's motivation/agenda well beyond that justified by the words he actually wrote. Calling someone a "petulant teenager" for raising an eyebrow or making a skeptical quip is not really justified, and is why I called it a troll post.

I could give two craps about thunderbolt being the second coming or not, since I own both Tbolt MBP and cMP. I'm not in that debate

I could also give two craps about taking sides in some imaginary debate between Team FutureVIsion and Team PathSkeptic

Most people have been aware of external thunderbolt expansion chassis like those made by Magma, etc, for some time. If the MP forum is going to get taking over by this silliness, well, its a shame.

Still waiting on that quote.

You're claiming that the OP's "trolling post" about Macbook Pros and Thunderbolt "cluttered up" the Mac Pro forum, and yet you saw fit to open it (despite thinking it clutters up the forum and is trolling) to do what exactly? For someone "not taking sides" you sure seem very eager to get involved, as was MacVidCards. On that front, my comments on his character are based on more than a single post, but that's getting into some pretty advanced comprehension techniques like "remembering what he has written in other posts" and "putting things in context".

What are the two of you doing in this thread if not to let everyone know how pointless and rubbish thunderbolt (and all technology decisions involving it) is?

My calling MVC a petulant teenager for a hilariously ill-informed (for a supposed technology expert) comment about how we're all being kept sated by promises about how "good" thunderbolt is coming despite the knowledge of how this was going to play out from Intel's description of the roadmap is entirely justified. Not to start a "he started it" tit-for-tat, but he did come blundering in here with his axe to grind, determined to start huffing and puffing as usual. The fact that he then also poorly used a Waiting for Godot reference to try and sound smart is even more amusing in a thread where actual results and demonstrated working solutions have been posted - the joke being, of course, that Godot never arrives.

If he didn't want to face criticism then he should have stayed out of it, as should you. For someone so concerned about the "clutter" that this short thread has brought to your precious nMP forum, you sure seem eager to wade in rather than simply ignoring it.

Sorry, next time we'll keep the discussion off the discussion forum. Wouldn't want to have any sort of discussion cluttering up the place, would we?! Cats and dogs living together! Mass hysteria!

 

[slughead]

It's a very impressive demo. Still has some technical rough edges, but definitely shows that it's workable and very usable.

The next rev of Thunderbolt is supposed to get even faster, so if Thunderbolt is the bottle neck here, it should be entirely a non-issue next revision of Thunderbolt.

The amazing thing here is that this shows the nMP may have GPU expandability far beyond the cMP. Four way Crossfire is certainly on the table at this point.

I'm not sure how well 4 way crossfire will run over 6GBps (a lot less if you have any 4K displays or any other thunderbolt devices). Probably about 60-80% performance drop using next years graphics cards. Sad to pay all that money for expensive video cards and PCIe chassis and get such a huge and unnecessary bottleneck.

If and when Alpine Ridge (TB3) comes out, nMP users will likely have to dump their iTubes off to exchange them for the next model, as I'm pretty sure TB3 devices will not work over TB2, even with an adapter.

I've heard also that rendering requires even more bandwidth than gaming, therefore TB2 is an even greater bottleneck--I've not seen anyone do benchmarks on it yet though.

 

[MacVidCards]

OK, let me get this straight.

There is someone in the world who thinks that moving a bunch of PCIE cards into a PCIE enclosure so that you can then rely on a tiny cable to connect them back to the PCIE bus on your computer is a good use of resources?

Sort of like moving your car engine into a trailer where it runs a hydraulic pump to pressurize fluid that flows through a small hose into the car where another unit transfers this back into forward momentum. Lots of lost energy and bother when you could just put the engine under the hood and be done with it.

And all of this "The NEXT TB is going to come even closer to that 100 Gbit/s that we promised in 2009" stuff.

Well, we got 1/5 of the promised speed, but now we need to switch connectors to get more.

Guess how much demand is going to exist for all of the TB1 and TB2 stuff in a year when TB3 comes out with it's different plug?

 

[dmccloud]

OK, let me get this straight.

There is someone in the world who thinks that moving a bunch of PCIE cards into a PCIE enclosure so that you can then rely on a tiny cable to connect them back to the PCIE bus on your computer is a good use of resources?

Sort of like moving your car engine into a trailer where it runs a hydraulic pump to pressurize fluid that flows through a small hose into the car where another unit transfers this back into forward momentum. Lots of lost energy and bother when you could just put the engine under the hood and be done with it.

And all of this "The NEXT TB is going to come even closer to that 100 Gbit/s that we promised in 2009" stuff.

Well, we got 1/5 of the promised speed, but now we need to switch connectors to get more.

Guess how much demand is going to exist for all of the TB1 and TB2 stuff in a year when TB3 comes out with it's different plug?

The physical size of the TB cable is irrelevant - it's the bandwidth that matters. Also, when Intel debuted LightPeak (Thunderbolt), they were using optical cables for that throughput. Finally, there are legitimate reasons to have an external PCIe enclosure over Thunderbolt (your car analogy is so wrong it's almost comical, because you equate an engine to a GPU when it's actually equivalent to a CPU). For machines like notebooks, trying to cram multiple GPUs inside would only be possible by sacrificing weight, thermals, and battery life, none of which are good compromises for a portable machine. You'd most likely have giant "laptops" with loud fans to not only fit multiple GPUs into the machine, but to provide enough airflow to cool down the system. And portability is completely out the window in that scenario. In comparison, a PCIe enclosure would add no weight to the machine, nor would it require giant fans to cool everything down. Even if you took the enclosure with you, the machine would still be significantly more portable than trying to cram multiple GPUs into a laptop.

 

[yjchua95]

I'm not sure how well 4 way crossfire will run over 6GBps (a lot less if you have any 4K displays or any other thunderbolt devices). Probably about 60-80% performance drop using next years graphics cards. Sad to pay all that money for expensive video cards and PCIe chassis and get such a huge and unnecessary bottleneck.

If and when Alpine Ridge (TB3) comes out, nMP users will likely have to dump their iTubes off to exchange them for the next model, as I'm pretty sure TB3 devices will not work over TB2, even with an adapter.

I've heard also that rendering requires even more bandwidth than gaming, therefore TB2 is an even greater bottleneck--I've not seen anyone do benchmarks on it yet though.

TB2 is a 20 Gb/s connection, so it's somewhere between PCIe 2 (16 Gb/s) and PCIe 3 (32Gb/s). The original eGPU post in the Tech Inferno forums showed that the TB2 setup showed at least 90% of the performance when compared to being connected to a PCIe slot in a desktop motherboard.

When TB3 comes out, I'm very sure that the connectors will still be the same. After all, TB1 and TB2 shared the same connectors and TB1 peripherals works over a TB2 port (at TB1 speeds), and TB2 peripherals work over a TB1 port (but at TB1 speeds).

So all the talk about 'needing an adaptor' for TB3 - it's all hogwash to me.

 

[Jmx]

OK, let me get this straight.

There is someone in the world who thinks that moving a bunch of PCIE cards into a PCIE enclosure so that you can then rely on a tiny cable to connect them back to the PCIE bus on your computer is a good use of resources?

Sort of like moving your car engine into a trailer where it runs a hydraulic pump to pressurize fluid that flows through a small hose into the car where another unit transfers this back into forward momentum. Lots of lost energy and bother when you could just put the engine under the hood and be done with it.

And all of this "The NEXT TB is going to come even closer to that 100 Gbit/s that we promised in 2009" stuff.

Well, we got 1/5 of the promised speed, but now we need to switch connectors to get more.

Guess how much demand is going to exist for all of the TB1 and TB2 stuff in a year when TB3 comes out with it's different plug?

For 3D rendering on GPU this solution looks highly interesting.
The job is loaded and processed on the GPU and there's not much data being sent back hence no need for more speed than TB can handle.
It's pretty much the render preview being sent back (a few frames/sec) and finally the render result when the job is finished.

And since the nMP isn't available with Nvidia GPU's we can't use render engines that need CUDA.
It's also interesting since i would be able to bring the external GPU unit with me and render jobs on lots of different machines.