For awhile now people have been asking us about TB EGPU cards. And I always said I wasn't interested AND that I didn't think it would work.
And I also said that we had nothing to offer, as the accounts I read spoke of hacking some kexts. And our specialty has been re-writing EFIs to make cards work in Mac Pros. I had read accounts from people who used Mac EFI cards and they got same results that people using PC Bios (and UEFI) cards got. They had same success, and same issues. So it was 100% my belief that EFI had nothing to offer, therefore we had nothing to offer.
But then I did some digging. And I bought a TB enclosure. And realized that cMP was referencing some older tables in it's EFI. (40 vs 41, if you must know). So I updated the tables, and made a few minor other changes...and guess what?
Turns out having EFI on a TB enclosure card is actually QUITE beneficial, the trick being it had to be an EFI that hasn't existed before. A newly modded form that takes into account the changes in Macs. Cupertino has once again eschewed the UEFI that is industry standard and used one a little different, a little off. Not the Mac Pro EFI, but their own.
So, by writing a new DIFFERENT custom EFI I have been able to make a GTX680 not only run CUDA in a 2014 Mini (or 2012 rMBP), but also SHOW BOOT SCREENS. This is the "new" thing EFI brings to table. Past eGPU efforts relied on self-init feature of newer drivers. So if you had driver issues or wanted to switch boot disks you had to unplug the eGPU and plug in the Mini. Now with this capability you can just plug the Mini into another box that can sit anywhere and you have a Mini that is now sporting a GTX680. It also solves a dreaded issue with "Code 12" running bootcamp. Basically Windows didn't know the card was there since it wasn't introduced during EFI boot and wouldn't set aside any RAM for it.
I see the Mini and the nMP being the principle beneficiaries of this tech, though laptops and iMacs can also use them.
Why bother you ask? Well, what if I told you that I was able to use a GTX780 6GB in my bottom-of-the-barrel Mini and score a higher OpenGl Valley score than any currently shipping Mac tested by Barefeats.com? So, if you want high powered OpenGl, skip the nMP and get the $500 bobby-soxer Mini and wait for a TB enclosure & 780 combo? Obviously, there are other considerations, but seeing the stumbling and stuttering Mini burst into a high-powered OpenGl machine made me realize that this was more than flash in the pan.
One thing I can tell already is that there is a tremendous amount of testing to be done. (Hence this post, i'll get to it). My 2012 rMBP and 2014 Mini show different behavior. Most notably the Mini is limited to running just the screen that boot screen shows up on, and that one alone. Something tells me this is Apple's doing, the Mini is already full of artificial limitations, I feel this is yet another. And while the 2012 rMBP will work with unflashed cards (minus boot screen & with Code 12 error in Windows) and can run as many displays as the card is capable of running, the Mini will ONLY work with EFI flashed cards. This may be something Apple has added for the reason noted above, Mini with GTX780 leaves all other Macs for dead in OpenGl.
I have posted in Mini section trying to see if anyone else has gotten eGPU working in 2014 Mini as it is the obvious choice being first with TB2. The best forum for this stuff is "DIY eGPU" at Tech Inferno. Can't find any there either, but I have trouble navigating that site.
So this finally gets me to why I began this post. I have every reason to believe that this same modded EFI will work in nMP. Sadly, I don't have one. Here is my little fantasy, somewhere in LA is a Production Office that bought one to try out. The lack of CUDA was a deal breaker and now the pretty little creature at the front desk is suing it primarily for FaceBook Puppy Pictures and sending in the afternoon Latte order.
I would love to borrow one for a few hours. Failing that I would like to at least drive over with my little creation and see if boot screens work, test would take 2 minutes, provided you have a non TB monitor.
The kind soul who does this will get word to Mac world faster than my bidding on various units on Ebay is going too. I will have one here, but would love to get first answer soon.
Oddly, that DIY eGPU site doesn't seem to have any reports from nMP that I could find. It would be interesting if Apple modded the nMP to behave like Mini, i.e., only way to see anything is with an EFI card.
Either way, we will have the answer soon enough, I just have lost like 3 bids on nMP so far.
Yes, bandwidth is constrained by TB, but the Mini's Valley score with internal GPU is 3.4 FPS on Valley. With the GTX680 it becomes 34.6, LITERALLY TEN TIMES FASTER. With the 780 it gets up into 50s. Completely transforms the thing.
And a quick "Happy Birthday" to the GPUs in nMP, they just turned 3 years old. Here is a chance to sidestep them, add CUDA, or just enjoy thumbing your nose at Apple and their artificial rules and limits.
Fact of the matter is that Apple and Intel have done nothing since Day 1 but throw roadblocks in the way of eGPUs. TB stuff is great for them as long as it sells things they want to sell. Neither of them wants to sell GPUs so they have simply refused to ratify anything for eGPU except a couple of $1,000 enclosures. (and you still have to hack the drivers)
So keep an eye on this space, I look forward to announcing that nMP can be upgraded to newer Nvidia GPUs and some screen shots. Hoping to have someone in LA help me make that happen sooner than later, but either way the info will come out.
And I also said that we had nothing to offer, as the accounts I read spoke of hacking some kexts. And our specialty has been re-writing EFIs to make cards work in Mac Pros. I had read accounts from people who used Mac EFI cards and they got same results that people using PC Bios (and UEFI) cards got. They had same success, and same issues. So it was 100% my belief that EFI had nothing to offer, therefore we had nothing to offer.
But then I did some digging. And I bought a TB enclosure. And realized that cMP was referencing some older tables in it's EFI. (40 vs 41, if you must know). So I updated the tables, and made a few minor other changes...and guess what?
Turns out having EFI on a TB enclosure card is actually QUITE beneficial, the trick being it had to be an EFI that hasn't existed before. A newly modded form that takes into account the changes in Macs. Cupertino has once again eschewed the UEFI that is industry standard and used one a little different, a little off. Not the Mac Pro EFI, but their own.
So, by writing a new DIFFERENT custom EFI I have been able to make a GTX680 not only run CUDA in a 2014 Mini (or 2012 rMBP), but also SHOW BOOT SCREENS. This is the "new" thing EFI brings to table. Past eGPU efforts relied on self-init feature of newer drivers. So if you had driver issues or wanted to switch boot disks you had to unplug the eGPU and plug in the Mini. Now with this capability you can just plug the Mini into another box that can sit anywhere and you have a Mini that is now sporting a GTX680. It also solves a dreaded issue with "Code 12" running bootcamp. Basically Windows didn't know the card was there since it wasn't introduced during EFI boot and wouldn't set aside any RAM for it.
I see the Mini and the nMP being the principle beneficiaries of this tech, though laptops and iMacs can also use them.
Why bother you ask? Well, what if I told you that I was able to use a GTX780 6GB in my bottom-of-the-barrel Mini and score a higher OpenGl Valley score than any currently shipping Mac tested by Barefeats.com? So, if you want high powered OpenGl, skip the nMP and get the $500 bobby-soxer Mini and wait for a TB enclosure & 780 combo? Obviously, there are other considerations, but seeing the stumbling and stuttering Mini burst into a high-powered OpenGl machine made me realize that this was more than flash in the pan.
One thing I can tell already is that there is a tremendous amount of testing to be done. (Hence this post, i'll get to it). My 2012 rMBP and 2014 Mini show different behavior. Most notably the Mini is limited to running just the screen that boot screen shows up on, and that one alone. Something tells me this is Apple's doing, the Mini is already full of artificial limitations, I feel this is yet another. And while the 2012 rMBP will work with unflashed cards (minus boot screen & with Code 12 error in Windows) and can run as many displays as the card is capable of running, the Mini will ONLY work with EFI flashed cards. This may be something Apple has added for the reason noted above, Mini with GTX780 leaves all other Macs for dead in OpenGl.
I have posted in Mini section trying to see if anyone else has gotten eGPU working in 2014 Mini as it is the obvious choice being first with TB2. The best forum for this stuff is "DIY eGPU" at Tech Inferno. Can't find any there either, but I have trouble navigating that site.
So this finally gets me to why I began this post. I have every reason to believe that this same modded EFI will work in nMP. Sadly, I don't have one. Here is my little fantasy, somewhere in LA is a Production Office that bought one to try out. The lack of CUDA was a deal breaker and now the pretty little creature at the front desk is suing it primarily for FaceBook Puppy Pictures and sending in the afternoon Latte order.
I would love to borrow one for a few hours. Failing that I would like to at least drive over with my little creation and see if boot screens work, test would take 2 minutes, provided you have a non TB monitor.
The kind soul who does this will get word to Mac world faster than my bidding on various units on Ebay is going too. I will have one here, but would love to get first answer soon.
Oddly, that DIY eGPU site doesn't seem to have any reports from nMP that I could find. It would be interesting if Apple modded the nMP to behave like Mini, i.e., only way to see anything is with an EFI card.
Either way, we will have the answer soon enough, I just have lost like 3 bids on nMP so far.
Yes, bandwidth is constrained by TB, but the Mini's Valley score with internal GPU is 3.4 FPS on Valley. With the GTX680 it becomes 34.6, LITERALLY TEN TIMES FASTER. With the 780 it gets up into 50s. Completely transforms the thing.
And a quick "Happy Birthday" to the GPUs in nMP, they just turned 3 years old. Here is a chance to sidestep them, add CUDA, or just enjoy thumbing your nose at Apple and their artificial rules and limits.
Fact of the matter is that Apple and Intel have done nothing since Day 1 but throw roadblocks in the way of eGPUs. TB stuff is great for them as long as it sells things they want to sell. Neither of them wants to sell GPUs so they have simply refused to ratify anything for eGPU except a couple of $1,000 enclosures. (and you still have to hack the drivers)
So keep an eye on this space, I look forward to announcing that nMP can be upgraded to newer Nvidia GPUs and some screen shots. Hoping to have someone in LA help me make that happen sooner than later, but either way the info will come out.
