MVC

Nice job. I have a 2012 rMBP and have found it to be MASSIVELY easier to eGPU with then the 2013 and 2014 models I have tried. Nice job.

Hey, if you have it working, great. I apologize if my tone was wrong, I just keep reading stories about failure and trouble and they frequently mention a riser. Years back I worked on and sold parts for G4 Cubes. The VRM board was in the way to add larger GPUs so there was a VRM relocation kit that was similar to a powered riser, it had a slot and a pcie type connector. We made them at the time from PATA cables that got cut up. I always found that less current was available to the system after installing one of these. So we could put larger GPU in, but system became less stable. The other reason this issue reminds me of Cubes was at the time the mantra was "Only original Geforce 3 can be flashed for Mac" Nvidia had introduced the Ti200 and Ti500 models, and there were ones that fit in Cubes, but everyone kept repeating the mantra, and prices for the Cube sized GF3 went through the roof while the same size Ti200 andTi500 cards sold for PC type prices. I discovered a way to move a single resistor on those boards and change device id into GF3, then they flashed beautifully. A little later on, someone else discovered that you could also just change the device id in the rom file. So, a couple of edits in Hexed and you tripled the number of cards that could be flashed for Mac. I was called a liar on the Cube forums when I said it was possible, the Mantra had been repeated so often that nobody thought to question it. There may very well be powered risers that are well made, use copper wire, good solder connections with gold plated pins, etc. But the "race to the bottom" to get them to be $5 on Ebay doesn't buy those. It buys ones made of aluminum or steel wire with shoddy soldering and brass plated contacts that get corrosion and degrade signals. There are so many barriers already to making these work on Macs that adding something that increases the false issues is bad, and just adds to the noise. Lots of people come here with hope, reading about a guy who has burned up 2 GPUs and has RMA'd them and happily waiting for 3rd one to plug into his riser is a BAD thing. When half of a circuit is powered and other half isn't, power runs wrong way, many components aren't designed to handle that and go "poof". Leaving the riser out eliminates it as a source of trouble. And thus increases your odds of spending time enjoying your eGPU instead of troubleshooting it.

Pick a number and color, you'll get your pics. Endeavors like this don't move forward by endlessly repeating myths.

I can and will post the pic. I do have what I claimed to have. So if I do....who will be the asshole then? Recent posts make it quite clear that the "get a powered riser" mantra has been repeated for no reason whatsoever. Somebody just posted about burning up a 2nd GPU that is getting RMA'd....using a riser...we all pay for those needlessly burned up GPUs. And the whole eGPU thing gets held back by people burning up eGPUs because nobody wants to put their foot down. I got a couple risers, powered and not. They ONLY MADE THINGS WORSE. Have a look at the guides, fewer Macs running eGPUs. Fewer OS X by a long shot. Need to weed out the nonsense and get to the root issues. Risers only make things worse

I am using Akitio TB2 chassis. Working on ways to use eEFI to solve Mac TB2 issues. In past we have been able to allow boot screens on 2014 Mini and 2013 nMP via eGPU and display output in OS X. Nobody else has done this, as far as I am aware. Have moved forward, now able to have fully active progress bar during boot and multiple displays on TB2 Macs in OS X. And the big one, I am typing this on a Maxwell equipped nMP that is running 2 displays via 750Ti. Doesn't always boot via eEFI, but can. Next up are GTX980 and Titan-X. They are proving more problematic but at least we have first Maxwell running as internal card. Have made some eEFI discoveries, but also some OS issues.

They don't help. They make the likelihood of black screens and dead Akitios and bogus GPU RMAs MUCH, MUCH higher. It has been proven to be a BAD idea yet newbies are still buying them and burning up cards because they read old guides from before this was figured out. They don't work, are a bad idea, etc. If you want a long card, bend the case. Look how many pins there are. Think about it. Bend the case.

I have 12 Akitios At least a dozen Macs I have probably 100+ GPUs Maybe I know what I'm typing about. Riser is a moronic idea that does NOTHING but make problems and lead to black screens, dead cards and bogus RMAs. Someone needs to put a giant red "DON'T BOTHER WITH RISER CABLES UNLESS YOU LIKE BURNING UP PARTS". Bend the end of the case and be done with it. I have worked on 1,000s of Mac GPUs. Take my word or the word of someone who has made 1 work. I have been playing with eGPUs for 6 months or so. Powered riser is MORONIC, STUPID, WORTHLESS, DOESN'T HELP. If people want actual progress here, it's time to take an active step in preventing this electronic carnage. And I am going to ask the seasoned "experts" here to stop posting theoreticals, we really need actual "what works" not "what some book on electronics says should work". I really feel that there is too much inertia of BS here to make progress anymore. Macs are getting trickier. If we don't shed the repetitive nonsense that everyone repeats because everyone else is repeating it, NOTHING WILL EVER GET DONE. RISERS ARE NO LONGER PART OF RECOMMENDED BUILD UNLESS YOU LIKE FAILURE.

I tried to flash nvidia and nvflash stopped when in eGPU Did the guy with dead card say how he "turned on PSU"? Without a paperclip or SWEX mentioned I am curious. Flipping the switch on a PC PSU does nothing more than let it wait to be told to turn on. And hearing "RMA again" is painful. If he has in fact blown up 2 cards then he is definitely doing something wrong, and EVGA is picking up the tab. Powered riser is begging for trouble. Not having swex or paperclip is begging for trouble. When power only gets to half of a card, it can flow in wrong direction, burning things up. TB2 actually has less bandwidth than PCIE 2.0 x4. Can be shown quite easily with PCIE SSD. Read up at Barefeats.com. He got same results I did, the blade SSDs are faster in x4 slot in cMP then in a TB2 chassis.

Just had a look, isn't this an error: <code style="box-sizing: border-box; font-family: Consolas, 'Liberation Mono', Menlo, Courier, monospace; padding: 0px; margin: 0px; background-color: transparent; border-top-left-radius: 3px; border-top-right-radius: 3px; border-bottom-right-radius: 3px; border-bottom-left-radius: 3px; word-break: normal; border: 0px; display: inline; line-height: inherit; word-wrap: normal;">sudo nvram boot-args="kext-dev-mode=1" Would kill the Nvidia drivers Should be: <code style="box-sizing: border-box; font-family: Consolas, 'Liberation Mono', Menlo, Courier, monospace; padding: 0px; margin: 0px; background-color: transparent; border-top-left-radius: 3px; border-top-right-radius: 3px; border-bottom-right-radius: 3px; border-bottom-left-radius: 3px; word-break: normal; border: 0px; display: inline; line-height: inherit; word-wrap: normal;">sudo nvram boot-args="</code>kext-dev-mode=1 nvda_drv=1"</code></pre>

newon, Look and see that the 3 kexts you modded all loaded, but also list "loadable - no" The System profiler means nothing, the fact that it shows up in the 2nd screen is good enough. Make it primary display and maybe that will change, but that isn't the issue causing KPs. Check the kexts and if they are OK, I imagine it may be your powered riser. I never found them to help, only make things more unstable. And I imagine that you will have trouble getting 4K output anyhow, as 2012 rMBP on blacklist for 4K via MST. Might work on SST, I will plug my machine in with 750Ti tonight when done working and see what I get. I have same 2012 rMBP and found it very easy to eGPU with.

No idea what you are asking. Please include details. Are those screen shots yours? Can't help by guessing

Have you tried a few times just plugging the TB cable in AFTER the boot chime? And I mean right after. Also, i have noticed two versions of the Error 12. The better one is when there is a line at bottom that says "The changes you have made to hardware require a restart" Usually I will have a good result on next one, even though I still need to plug in TB after chime most of the time. On 2014 Mini I can get by Error 12 with a reboot or two. On 2013 nMP I have yet to find a way, haven't dug into DSDT but instead mucked with mmio in EFI. A lot of people very curious about your test as things are worrisome with Apple making harder with each new model.

Thank you, haven't tested yet but if this works, could really help the newbs who might have buggered it otherwise.

I don't understand why he has a riser at all. He has the Akitio barenaked, he could just plug the card in and be done. Just needs to power the Akitio barrel via the 12V he has on riser. Also, I find that Windows works much better with TB being plugged in AFTER boot chime. What Nando suggest should work. I think you just need to delay the plug in. I have seen the UEFI on cards freak out the Mac EFI before, plugging in after the chime means it doesn't get read.

DING DING DING, and we have a winner !!! You have somehow borked that driver. It shouldn't be loading that 2nd kext. It should be loading the one that ends with "Web" The one it is loading hasn't got an entry for Maxwell cards, and thus says "Nvidia Chip Model" Your problem is right there. EDIT: This gets even weirder. There shouldn't be a NVDAGM100HAL that doesn't have "Web" in it's name. Did you ever load a pre-release Beta or something? I literally mean there has been no such kext that I am aware of. Hopefully new driver fixes things. Here are kexts loaded in my rMBP: 115 2 0xffffff7f80e49000 0x2d8000 0x2d8000 com.nvidia.web.NVDAResmanWeb (10.0.3) <87 82 78 12 7 5 4 3 1> 116 0 0xffffff7f82564000 0xd00000 0xd00000 com.nvidia.web.GeForceWeb (10.0.3) <115 87 83 78 12 7 5 4 3 1> 117 0 0xffffff7f81342000 0x1af000 0x1af000 com.nvidia.web.NVDAGK100HalWeb (10.0.3) <115 12 4 3>

I don't think stability is often the problem. Any time I want to fire up my Mac Mini and play games it gets right to desktop in Windows 8.1. And I mean like clockwork. And it almost NEVER crashes. Gaming on classic Mac Pros have same number of crashes. Only real issue is ability to boot in the first place. On MBPs there is something that prevents OS X from display screen output on Nvidia cards. But it is quite likely that we (MacVidCards) will soon be offering cards that get around this on some TB2 Macs. We have been able to have these eEFI cards show boot screens, just as if output was from internal card. We have been able to make this work on every TB2 Mac we have tested. There is some concern on the 2015 13" MBP, hopefully we'll get to test soon. The only combo that has given us trouble is nMP and Windows. And this has more to do with the bewildering array of PCI to PCI bridges that weave 6 exterior TB ports with 3 internal ones that are also connected to other things. Anyhow, there is reason to believe that we will be able to offer flashed cards that break some or most barriers. And also possible that some experimenters on here will figure a way to do a similar thing in EFI shell.

OK, and if you click in upper right corner on Nvidia, do you see this? If using it in Windows it can play a game or run a benchmark we can rule out power as the issue. Also, have you tried plugging in right after boot chime like what you do for Windows? I would echo goalque, there really is no precedent for a TB1 Mac to have these issues. It really feels like some other outside issue.

The 2013 and 2014 TB2 Macs got more difficult, will be sad if 2015 13" is impossible. If anyone in LA area wants to find out, I can test with Akitio and standard PC cards as well as eEFI cards. Would take maybe 30 minutes, tops.

Mark, I believe that you and Izard have completely different issues. His machine type has not been noted for issues in the past. TB1 Macs typically are rather easy for OSX and a little tougher for Windows, if they are on MBR. Your machine is known for being one of the few TB2 machines that can achieve screen output from eGPU in OSX. The kexts I posted had an error in the HDA part, but I don't think most people use the audio out to their monitors anyway, as it typically requires yet another hack. You should be able to have your Titan show up in System Profiler with right name and 6143 Meg's of RAM. It is only screen output that should be tricky and there was a recent discovery with that gfx switch app that may help you. I have 4 TB Macs here that I test with. None of them have acted any differently in 10.10.3 then in past versions. One is a 2014 rMBP (iris only). I test on a Dell 5K and an Asus PQ321 4K. In OSX resolutions are limited via a blacklist on certain machines, but yours shouldn't be on that list. I will fix the broken HDA kext and post them again. If your Titan is a standard first gem Titan it doesn't even require the Web Drivers. I have one of each Titan here so I can verify.

IIRC, I found that adding 2nd card was great but anything beyond 2 hit the point of diminishing returns. The bandwidth can only go so far in TB2, especially on a Mini. The nMP may have more potential. I am down to 12 Akitios as I sent #13 to my resident genius. He will have in a week or two, expect some progress. This community has been in a rut for awhile. Many people seem to have stopped posting once their particular situation is solved. When I first posted about TB2 Macs having issues it looked like I was trying to drum up demand for my own cards. But I think it is pretty established now that most TB2 Macs do in fact have issues, usually with OSX but at least with nMP they can also have Windows issues. Expect some changes coming. - - - Updated - - - Well, next step is to find someone else with same machine and see if they have same issue. The drivers support GTX970. So issue is either with entire line (i.e., everyone with your machine will have trouble) or only with your specific machine. (i.e., bad hardware or software) It is one or the other.

Honestly, who cares? As I stated, anything I can do in EFI, you can do with mmio commands in EFI shell. So if you or Netstor can't rest until you get 3 running, start there. I did it fairly easily but I'm not going to drag all the wires into the same room and do it again. Again, if I could do it with EFI loaded on a card, you can do the exact same thing in the EFI shell. That is how EFI works. Point being there isn't some hard and fast lock, there just needs to be more EFI testing done. If we made genuine progress with EFI commands, I wouldn't end up having to write a gazillion different EFIs for every card on the planet. I would be thrilled, I spend FAR TOO MUCH time staring at Hexedit screens. Nobody is going to make any progress with these things unless they get scientific and write things down, record what works and doesn't. The other half of my team is joining eGPU fray in next 2 weeks. We will be further altering the face of this hobby, hopefully turning it a little more mainstream. General OS things we find will be made public knowledge. While trying to get a handle on it, many people get locked into a mode of thinking that HURTS the effort. All of the "must have a powered riser" stuff was hooey. Yet it was repeated like a mantra because one person said it first. I hear a dog bark. Someone gets off the elevator. Did the person get off the elevator because the dog barked or did the dog bark because he heard the elevator? Assuming either is dangerous and unsupported. Maybe the dog barked because his owner held a biscuit in the air and the two events had NOTHING WHATSOEVER TO DO WITH EACH OTHER. Whether I used EFI or Fairy Dust to make 3 run matters little. If I did it, it is doable and anyone else can do it. But nobody cares. Most people on this board would be THRILLED to get one eGPU working and have no intention or desire to have 3 running so what good does it do to exhume and beat that horse? You said 2 was the limit, I proved 3 could work. Done. - - - Updated - - - Very glad I was able to help. This is what I wish those who have had success could do. Help the newbies. I am still hopeful that your OS X issues can be solved. Did you try the "nvram -p" command to check that both needed boot args are loading?

My goal right this second is to help Odyseus get his setup running. I already posted screenshots of Mini running 3 external GPUs. Anything I can do with EFI can be also done by others with mmio in EFI shell. So, Mini can run at least 3 Nvidia eGPU is a proven fact. Done. As far as Odyseus issue, I think that it is important to note that the "TB2 Macs no Nvidia output" issue is something that can only be confirmed ONCE THE CARD SHOWS UP IN SYSTEM PROFILER. With all 3 TB2 Macs I have with the issue the cards looks fine and dandy in the system Profiler. It has right name and the kexts are all loaded. If I install CUDA I can run CUDA-Z and get numbers moving. The "TB2 Macs" issue is that even having it look perfectly happy in profiler it won't EVER output via one of it's ports. I believe that him seeing the "Your Nvidia Driver blah, blah, blah........... using OS X Default" means he installed the wrong one for his OS. So the NVStartup kext doesn't load and BINGO, nothing. If he shows back up in this thread I'll toss a GTX260 in and see what I get. But KPs and boot loops are a different issue from the TB2 no-output issue. And I really think Windows issue is also separate. WHEN the TB plug goes in has big influence on success in Windows. Someone in last few days followed my timing advice and got right into Windows. I am concerned that newbies stumbling on these pages get dragged into the myriad issues and have no clear path, especially when the issues get all muddled together. Many people don't understand that those plugs on the GPU are mandatory. If they are there, they need to have power run into them. Many people don't understand that just plugging a PC PSU into the wall and flipping the switch to "On" is not the same as "turning it on". I firmly believe that MANY people do the mods wrong or don't repair permissions afterward or get into kext dev mode. Many people don't understand that many early cards don't need the web drivers at all. Many people don't understand that CUDA isn't needed WHATSOEVER unless they have an App that uses CUDA. (Adobe Premiere, DaVinci Resolve, etc) I find it a useful debug tool, but otherwise the guy who wants to play Crysis on his MacBook has no need of CUDA. And I think many get hung up on "must install nvidia web driver" and then use the package puffer-upper and mod the plists, etc. And, like Odyseus end up with having modded the wrong driver, creating lots of work and getting no benefit. I am concerned that many people read all the tales of woe and decide that a PS4 or a PC gaming rig isn't such a bad investment to avoid the hassles. And I have to say that my 2014 Mini is a great gaming setup with eGPU. Doesn't need any fancy eEFI or modded kexts to get into Windows. I just have to pay attention to timing and then I am off for hours of Far Cry 4, skinning leopards, etc.In 3 months of game playing it has crashed maybe twice, it is rock solid stable and crashes no more than my PS4 or X-Box 1.

Odysseus, each build of OSX requires a specific NVIDIA driver. The message you got indicates you used wrong one. A little understood secret is that you should be able to use the 260 to allow install of correct web driver. If you go to our site we gave download links for all NVIDIA drivers listed by OSX build number. Also have a blog post about how to find your current build number. With my 2014 Mini I am able to reliably get into Windows, the key being to plug TB cable in within seconds of boot chime. I cal also run more than 2 eGPUS but that has nothing to do with your problem. Does the Cube power the GTX260? Power plugs on GPUs aren't optional. Macvidcards.com Look in FAQ section for driver list

Great thread ! I think the title has an error in it, isn't 2013 rMBP a TB2 machine?

If you type "Nvram -p" in terminal do both of the needed boot args show up?