r00x

Hi @Kazeki, it does seem easier now yeah. However I'm quickly learning it matters which GPU you're trying to use. I just tried the exact same thing with a GTX 970 and it didn't work at all. Running 56.25GB endpoint PCI compaction on it as above (eGPU+dGPU with eGPU forced 32 bit) failed with an error like "no solution found". When I eventually found something that worked it just hung on booting Windows. In fact for the GTX 970 it seems the way to get it to work was, after the DSDT override, don't use eGPU Setup at all and just boot straight into Windows. It worked right away then. Weird. You're using a 650Ti, which I don't have here to test... hmm. How far have you gotten with yours then? Have you got to the point where it shows "Large Memory" in Device Manager? Note that on Windows 8/10 you have to do an extra step to enable DSDT override - enable test signing by opening Command Prompt as admin and using: bcdedit -set TESTSIGNING ON Also check this thread, it seemed a little easier for me:

Ok it's fine with rebooting. Although it should be noted the internal LCD and external display remain dead after chainloading begins - both only wake up once Windows finishes booting (at first I thought the system had hung, but I guess it was just sat at the Windows bootloader countdown screen thing and I couldn't see it).

IT WORKS. GT650m and GTX 560 Ti both working in Device Manager. Now if only I can remember how... So I think after the DSDT override I ran PCI compaction, 56.25GB endpoint, but targeting only eGPU+dGPU, then when asked to force a GPU to 32bit, I picked the eGPU. Previously I was running compaction on "ALL except (sata, etc etc)" which was the recommended setting I'd seen in previous Macbook guides. Am now scared to reboot.

Hmm, so forcing the dGPU to 32bit during compaction (DSDT overridden, compaction with eGPU setup 1.20 with 56.25GB endpoint, pci_alloc_valid == 512MB) gets me the same, error 12 on dGPU, but at least the internal LCD works again. I am now basically where I was in the first place, where eGPU would work so long as the dGPU was disabled in control panel, but internal LCD is software-driven. The tinkering continues!

It didn't work for me, alas. I got as far as getting the "large memory" shown in the Resources view of Device Manager, and using PCI compaction with the 56.25GB endpoint shows pci_alloc_valid == 512MB (previously it would only ever say 256MB). If I then chainload Windows it lets it boot and the eGPU works, but the dGPU is throwing error 12 now. Still have some stuff to try like forcing one of the GPUs to 32bit during compaction, will report back...

Hi @Kazeki, If you can see the card in Device Manager that's good news. Hmm, it shows as Microsoft Basic Display Adapter - usually at this point it's a case of getting the drivers installed for the eGPU, I think? Once you disabled the GT650m, did you restart? Have you installed up-to-date drivers from NVIDIA? The default Microsoft ones might not be good enough (I might be wrong, but I think the error you describe is it complaining that it expected to work only with a mobile GPU?). Maybe try this: Download latest NVIDIA drivers from their site (use your eGPU when they ask what card it's for) Then uninstall your existing drivers (from Programs and Features, not device manager) Then try to limit Windows automatic driver downloads (go to your System Properties (right click in bottom-left corner of screen and click System), then Hardware tab, then Device Installation Settings and set it to "No".) Then restart Then install the new drivers you downloaded (use the clean installation option). It should, ideally, show both GPUs now, but the eGPU will have an error (maybe 43, like with mine) Then disable your gt650m and restart - ideally now your eGPU will pop into life! You do need an external monitor plugged into the eGPU and turned on, I think.

Hi @Tech Inferno Fan, thanks for the suggestions! I did have a look at the other implementations but didn't see any really mentioning editing the pci.bat file. Most seemed to just mention running PCI compaction which is what I did - in fact one I recall explicitly said they didn't edit any files. Even looking now, perhaps I am bad at searching, but I can't see any advice on what to edit into the pci.bat file (and I thought, perhaps erroneously, that was only necessary if you didn't want to manually run compaction and boot with the test function, anyway?). I did get the "yes" for all things being correctly allocated but the behaviour in Windows 10 was no different to if I did not run eGPU Setup 1.20 at all (in other words the eGPU still threw error 43 unless the GT650m was disabled first). So for now it's been easier to just forgo all that and disable the dGPU but of course it's not an ideal solution... I am suspicious of the drivers, though. Can error 43 be thrown by drivers? The GT650m is a mobile chip and the GTX560Ti I was testing as eGPU is an discreet desktop card. Could it be the drivers don't like the two running side-by-side?

Hi Kazeki, what switch settings have you used on the PE4C? I have the exact same laptop (assuming you have Retina model? Anyway not much different except for screen really) and I've found it worked best with SW1 set to "2" and SW2 set to "1". Set up like that, I can get eGPU running in OS X or Windows with no hotplugging required. However for me, to get eGPU to work in Windows, I did have to first open Device Manager and disable the GT650m. If I do that, then restart, the eGPU works and you can still use the normal laptop display (but you might need to reset the resolution and scaling, and it won't be hardware accelerated like the external monitor). eGPU setup 1.20 didn't seem to work for me, it made no difference using PCI compaction so I actually don't use it at all - as long as I disable the GT650m in device manager, the eGPU works, and this is good enough for playing games on an external monitor. eGPU setup 1.30 couldn't even complete PCI compaction without hanging/crashing. If I don't disable the GT650m then Device Manager will show the eGPU is detected but it says Code 43 error. Hope this helps!! Also let me know if eGPU setup worked for you and you somehow can have the GT650m AND the eGPU running because I'd love to know how to do that.

Hi willex - what happens when you run stuff fullscreen? For instance if you maximise the Finder/browser window/YouTube? I've done a little more testing - it seems fullscreening anything will get rid of the judder on the external monitor. For example, fullscreening Finder will let it run smoothly, whereas when it's just a window on the OS X desktop, it exhibits the same skipping/juddering every second or so. Having said that, since I switched from the Nvidia web drivers back to OS X standard ones (using the menu bar icon/Nvidia driver manager thing) it seemed to run a bit better too. Certainly more games work than before, barely any games launched (fullscreen, anyway) with the Nvidia web drivers but plenty worked fine on the original OS X ones. I've since found it's a lot better an experience on Windows, but there I have a separate issue, which is I have to disable the internal GT650m dGPU in Device Manager before the eGPU can activate, I guess because for some reason PCI compaction isn't working.

Hi there... the journey continues!! I've managed to get my eGPU running on Windows 10 (installed via bootcamp, BIOS mode) however I have strange behaviour from the system compared to what others with similar setups have. Without using eGPU setup at all, I get code 43 on the eGPU in Device Manager (it's correctly detected and even tried to install drivers but it won't start). When I used eGPU Setup 1.20 however, it's the same!!! ( I use 1.20 as 1.30 hanged when compacting PCI). I boot eGPU Setup 1.20 into the menu mode (option 2) and use Compact PCI with standard settings (exclude all the SATA/etc etc stuff). On hitting Windows nothing has changed. I've uninstalled all the drivers, disabled as best I can windows 10 pulling drivers automatically from Windows Update, and then disabled the GT650m dGPU and installed GeForce drivers manually. This, in combination with keeping the dGPU disabled afterwards and rebooting, and using eGPU setup again, seems to be the only way to get the eGPU to work. It's not the end of the world since on this model I can't use Optimus anyway and with an accelerated external monitor I'm not really interested in doing much on the internal panel. Even with the dGPU disabled I can use the internal LCD and change its resolution back to native 2880x1800. It's just an irritating extra step though that nobody else seems to have encountered? At least I haven't come across it, has anyone dealt with this before?

Ok, so I tried switching from "NVIDIA Web Driver" to "OS X Default Graphics Driver" using the NVIDIA Driver Manager and that has improved things tremendously with many titles I try now launching. Still no War For the Overworld though, alas...!

Hi there, thanks to the help of this forum I've got as far as a sort-of-working eGPU setup on my 2012 rMPB (GT650m model) and am currently using a Zotac GTX 560ti 448 core for testing, with a PE4C V3.0 and a Sonnet Echo Pro Thunderbolt to Expresscard adapter. It works in OS X and I can drive an external monitor. I have set the external monitor as primary and can run Unigine Heaven on the eGPU (and it does seem to utilise the eGPU, now). However! Most games I try to launch crash out immediately (the icon appears in the bar then disappears unceremoniously). For example, War For the Overworld and Habitat both do this. Others like Kerbal Space Program just hang on a blank screen, or Half Life 2 which just sits as an icon on the bar. It seems that anything that tries to run fullscreen just implodes; though some windowed stuff won't work either, I do seem to have slightly better luck with those. I haven't seen anything to suggest that this is normal... other people talk as if their eGPU implementations are working just fine? Is it maybe just the wrong graphics card to use?

Ah, I see!! I disabling mirroring and dragged the menu bar to the external monitor in Display Arrangement (ostensibly setting it as the primary?) and now it works. I can hear the Vregs in the eGPU whine under load and the performance is much greater in Unigine! Thanks very much for the tip. It seems I can also drag the window back to the internal LCD and it still uses the eGPU, however, I have a new problem... The performance seems to skip a lot... as in every second or so it pauses for about 100ms? This is true even in the normal OS X desktop - for instance if I drag a window around, it moves smoothly but skips/judders every other second. Or for instance playing a YouTube video, it seems to judder at regular intervals instead of playing smoothly. (side note: if I play YouTube fullscreen, and leave it in focus, it plays smoothly...) is this a common issue?

Hi Dschijn, I did try fiddling with the display arrangements. I could change the mirroring settings between no mirroring (extend displays - GT650m did all the rendering), mirror the internal display (GT650m does everything again) and mirror the external display on the eGPU (in which case, it *said* the eGPU was doing all the work). However, it only claimed the eGPU was doing anything. It actually seemed to be idle. So by mirroring the external display on the eGPU, it made About This Mac show each screen as being driven by the eGPU - however, when trying something like Unigine Heaven it was clear the internal dGPU was still doing the rendering. Not sure how to proceed, the eGPU is powering up and detected and working perfectly well to drive the monitor but I can't get it to actually do any rendering even by changing the mirroring settings!! Is this what you meant by setting it to be the primary display? I'm not aware of any other option to control primary/secondary displays in OS X... perhaps if I drag the menu bar to the eGPU display in the Display Arrangement window maybe?

Hi there, lots of useful info on here! Still I need to ask a few questions as my setup isn't quite working! I've a 2012 rMBP on El Capitan 10.11.2 and I used the automate_eGPU.sh script to quickly sort out the modification of files and installation of NVIDIA web drivers. I'm testing it with an NVIDIA GTS450 which is connected via a Sonnet Echo Pro and PE4C 3.0. SW1 is set to 2 (PERST# delay 9.8sec), SW2 to 1 (CLK no delay). It's picked up fine in OS X and I can drive an external monitor with it no problem. However, it doesn't appear to be doing any rendering - all the legwork is still being done by the internal GT650M!!! It shows each screen as being driven by the GT650M in About This Mac, and testing with games seems to corroborate this. What gives? I can't find any examples online of other people who've had this problem (at least in OS X). The GTS450 is correctly identified in System Information. A second question... I know there isn't a way to get the output on the internal LCD on Windows on this Macbook due to it having a dGPU and Apple's rubbish firmware, but is that the case in OS X? The Intel HD4000 is accessible there, does that mean I could funnel the graphics back to the internal LCD or is it just impossible on all counts?