Izzard

"No."? I think you misunderstood me: I was suggesting a method that uses what an ATiKiO owner already has, rather than having to buy anything, that's all. I'm sure your method works just great

I had a similar experience to others: bought screw-terminal (no-solder) barrel plugs on eBay and when they arrived, even though they were the correct spec, they didn't fit as the centre pin of the AKiTiO did not fit into the barrel plug. I recommend just chopping the plug off the original AKiTiO PSU (assuming you're modifying the AKiTiO anyway) and wiring that up to your ATX PSU!

Ah, just realised you haven't reposted the files / link.

Glad I’m not losing my sanity just yet! And this is practically a virgin install of 10.10.3 as of a few days ago, from a freshly downloaded install image (never upgraded from an earlier version). I will try again before I change from Asus STRIX to MSI Gaming. If either of those things cracks it, it’ll be interesting!

Thanks MVC for posting your kexts. I tried them, and still had the same issues ("NVIDIA Chip Model" in System Information, and kernel panic if I try to boot OS X with the eGPU connected). By the way, using your kexts I got this in the log: Driver "AppleHDAController" needs "IOPCITunnelCompatible" key in plist

Thanks, MVC. So it's either the fact it's a 2011 mini, or it (hopefully) just a gotcha with the Asus STRIX. MSI Gaming Edition arrives tomorrow... **fingers crossed**. One other thing I'm wondering about: I'm unable to use Physx in Windows because the Nvidia drivers detect the Radeon in my Mac mini (regardless of whether I disable it in Device Manager and/or uninstall the driver), and therefore disables PhysX, forcing it to run on the CPU. If nothing else, this murders my 3D Mark score because even though the "graphics" benchmark is very good, the combined test is crushed by running Physx on the 2.7 GHz dual-core i7 instead of the GTX 970. This makes me sad. Do you know any solution to this? Should I consider getting nando's Setup software and experimenting? Thanks again!

Hi Mark, Glad you got it working. This is interesting to me because it confirms my latest theory: that 10.10.3 (14D136) and/or 346.01.02f02 simply does not work, at least not for me with my GTX 970 and 2011 Mac mini. I have tried "everything", like you, even totally from scratch and in different orders multiple times. So, for now I just have to give up and - like you - hope that the next revision of the OS / Web drivers fixes it. In the meantime, I've ordered an MSI GTX 970 to try instead of my Asus one.. .but I don't have high hopes. - - - Updated - - - Well I'm not making a bigger mess because I'm careful to make sure I can roll back after my experimenting (which I did when I realised it wasn't a true solution). We have to dabble to learn, you know! NVStartup is definitely modded correctly, permissions are definitely correctly; this latest update from Mark has convinced me this version of the OS / drivers just doesn't work for this card. The nearest similar setup I've seen is a 2012 mini (not 2011) and with a Sonnet case instead of an AKiTiO. Tomorrow I'm going to try with a different brand of GTX 970 (an MSI) and I'll report back. When I finally get it working I'll make a guide to help my fellow 2011 mini owners.

Those two things won't change even when it DOES load. It will show "loaded" in spite of being "Loadable: No". So I think the question is: is it being requested to load? Does the card show up under Graphics in System Information? Have you eliminated all remaining requests for IOPCITunnelCompatible in the system log?

OK, in case anyone is following my plight... Windows: as mentioned above, I've honed my timing and now I have the GTX 970 working beautifully in Boot Camp / Legacy/BIOS mode Windows 10 Preview just by switching on the eGPU about 1-2 seconds after the Mac mini. Starting it at any other time doesn't work. This is golden! OS X: not much progress with OS X yet (still showing as "NVIDIA Chip Model" and not actually usable even for CUDA, despite drivers/kexts loading fine). However, I have found a way to stop it kernel panicking if I boot with the eGPU on... Following my hunch about it hanging on IONDRVSupport, I experimented in the plist; specifically, the third block: I've changed the string value for key PCIClassMatch from: 0x03000000&0xff000000 …to... 0x10de0000&0xff000000 I did this because I took the original version to mean "Only accept AMD GPUs for…[something]", with 0300 being the AMD vendor id and "FF000000" being a mask. I replaced the left hand side with 10de because that's the vendor ID for NVIDIA. And now I can happily boot into Yosemite with the eGPU connected. However, I won't celebrate just yet because I don't really know what I'm doing and it's possible I've simply gone around the issue instead of actually curing it. Can anyone understand why this has stopped the kernel panics? Anyway - I'm no further forward really as all I've achieved is being able to boot into the same result as I was getting previously by hot-plugging after booting. Interesting, though…? Update: I get the same result by removing the IOPCITunnelCompatible modification from that block instead of changing the vendor ID.

Could someone please tell me if this looks correct for my hardware (2011 Mac mini, i.e. TB1)? Also, I seem to be missing Physx and CUDA... and there's no option in the NVIDIA control panel. PCI-E 2.0 x 16 @ x4 2.0

Right! I've worked in tech support and I sense the frustration of being drip-fed information when you're trying to help someone I appreciate the comment... you can imagine what my implementation guide is going to look like when I get around to it. I have some particularly detailed photos of cracking open the AKiTiO case neatly that I will share. I've been doing a *lot* of reading before posting. As you have mentioned, it takes a lot of reading around the subject to understand the process from beginning to end and how to troubleshoot it. I guess a wiki would be useful to curate nuggets from these forums and weed out the red herrings... but these are some of the most neatly organised forums I've ever seen so it's a close second Yup, it's a short wire shunt i.e. paperclip trick. I start the PSU using the rocker switch on the back of it. I'm sure you're onto something here. The main breakthrough for me was when I was reading the thread-of-dubious-usefulness "how-often-do-you-boot-black-screen?" and came across a comment to that effect from you. So last night I concentrated my efforts on Windows. EFI mode is a real uphill struggle on this machine, even if I nobble the troublesome igdkmd64.sys file, so it's back to legacy/BIOS mode. I'm still refining the timing but I have success most of the time in Windows now if I cold-boot the mini and then flick the rocker on the eGPU PSU just as the mini's power LED illuminates, i.e somewhere between 0.5 seconds and 3 seconds after powering up, when the display fires up but before the boot option screen if I'm holding the Option key. A fraction too early and Windows goes black instead of booting. A fraction too late and Windows boots but it's oblivious to the eGPU! So as far as my setup is concerned for Windows it is all about timing, timing, timing! For me, I put this new-found knowledge on a par with the realization by nando and co. about not needing a powered riser. I have shared my modded files via pastebin, linked in my post, but it really seems to be IONDRVSupport.kext which is one of the triggers. I get the feeling the crash comes when it hits that part of booting. I have. There are a couple of 2011 Mac mini builds and they seem to work fairly straightforwardly. However this might be the first 2011 mini with a 970 (though I get the impression you will have tried it at some point) and the 970 seems to give more trouble compared to the 960 and 980 for some reason. All three show as "Loadable: No" and "Loaded" but interestingly NVDAStartup seems to *unload* after sometime. I haven't quite got my around that yet.

Hi all! TL;DR: If I start up my AKiTiO/GTX970 eGPU and then turn on my Mac mini, Yosemite boots most of the way then - instead of displaying the login screen - the computer restarts. This repeats indefinitely. If I power up the eGPU with OS X running, System Information shows it as "NVIDIA Chip Model" instead of "NVIDIA GeForce GTX 970". The kernel panic while booting only happens once I've modified IONDRVSupport.kext. The rest of this post will be me explaining what I've done and what results I get. I'd be over the moon if someone can help me get this thing started up! (Incidentally, I have Windows 10 Insider Preview installed in legacy/BIOS mode on a second SSD and this same hardware worked at least once giving me fantastic benchmark results, etc. But generally, Windows also fails to boot if the eGPU is powered on.) Hardware: Mac mini (Mid 2011) Processor Name:Intel Core i7 Processor Speed:2.7 GHz Number of Processors:1 Total Number of Cores:2 L2 Cache (per Core):256 KB L3 Cache:4 MB Memory:16 GB Boot ROM Version:MM51.0077.B10 SMC Version (system):1.75f0 [*]AKiTiO Thunder2 PCIe box [*]PSUs tried: CiT 400W 12cm Silent Fan Dual Rail Corsair Builder Series CX 430 Watt ATX/EPS [*]GPUs tried: MSI AMD 280X Graphic Card (3GB, 384-Bit, GDDR5, DL-DVI-I/DVI-D, HDMI, DL Mini-DP, TF FAN, PCI-E) Asus Nvidia GeForce GTX 970 Strix Graphics Card (4GB, GDDR5, PCI Express 3.0) [*]2 displays - one connected the Mac mini by HDMI, one connected to the eGPU by HDMI. (I added the "server" SATA cable to my Mac mini and it has two identical SSDs in it, so I can have one totally dedicated to OS X and one totally dedicated to Windows. I really like this setup when it comes to formatting and rebuilding new OS versions!) Software: OS X Yosemite 10.10.3 (14D136) CUDA Driver Version: 7.0.36 GPU Driver Version: 10.3.7 (346.01.02f02) [*]Windows 10 Insider Preview build 10074 As mentioned, I can "hot-plug" the eGPU in OS X. This is what appears in the console: [FONT=courier new]03/05/2015 19:58:15.000 kernel[0]: AppleThunderboltGenericHAL::earlyWake - complete - took 0 milliseconds 03/05/2015 19:58:15.000 kernel[0]: IOThunderboltSwitch<0xffffff8031861000>(0x3)::listenerCallback - Thunderbolt HPD packet for route = 0x3 port = 11 unplug = 0 03/05/2015 19:58:15.000 kernel[0]: IOThunderboltSwitch<0xffffff8031861000>(0x3)::listenerCallback - Thunderbolt HPD packet for route = 0x3 port = 2 unplug = 0 03/05/2015 19:58:15.000 kernel[0]: IOThunderboltSwitch<0xffffff8031861000>(0x3)::listenerCallback - Thunderbolt HPD packet for route = 0x3 port = 12 unplug = 0 03/05/2015 19:58:15.000 kernel[0]: considerRebuildOfPrelinkedKernel prebuild rebuild has expired 03/05/2015 19:58:17.000 kernel[0]: [ PCI configuration begin ] 03/05/2015 19:58:17.000 kernel[0]: [ PCI configuration end, bridges 17, devices 20 ] 03/05/2015 19:58:17.360 com.apple.kextd[40]: kext-dev-mode allowing invalid signature -67030 0xFFFFFFFFFFFEFA2A for kext "/System/Library/Extensions/NVDAStartup.kext" 03/05/2015 19:58:17.000 ker[/FONT][FONT=courier new]nel[0]: NVDAStartup: Web 03/05/2015 19:58:17.000 kernel[0]: NVDAGM100HAL loaded and registered 03/05/2015 19:58:17.000 kernel[0]: [AGPM Controller] unknownPlatform[/FONT] Is that part about AppleGraphicsPowerManagement Controller significant? Notice there are no complaints about requiring IOPCITunnelCompatible so I'm pretty sure I've edited the Kexts correctly. The three Kexts in question (when using the Nvidia card) are AppleHDA, IONDRVSupport and NVDAStartup. I've tried swapping in and out edited/original versions of these in different combinations and it seems that AppleHDA and NVDAStartup - when modified - don't cause the issue. If I use the original IONDRVSupport then it doesn't kernel panic on boot, only if the ePGU is connected and IONDRVSupport is modified. Hardware combinations I've tried: I started with the cheap CIT PSU and the MSI AMD R9 280X. I adapted the PSU to power the AKiTiO via a barrel plug. I also tried using a powered PCIe riser. It made no difference to the symptoms. I couldn't get any display out of the 280X (of course some of the kexts required are different) so I went out and bought a better PSU and GPU. (The much nicer Corsair and the Asus Nvidia GTX 970). I tried this eGPU powered by barrel plug and PCIe riser. Still no joy - extremely similar symptoms, just Nvidia-flavoured now. I've also tried the CIT PSU with the GTX 970 and the Corsair with the 280X, just for completeness but these don't seem to be what's causing the problem. I have since returned the 280X to concentrate on getting the GTX970 working (because it worked great in Windows on one of my attempts). Attempts with OS X Yosemite: If I try booting with the eGPU powered on, OS X's loading progress bar gets to the point where it would normally display the login screen / Desktop and the machine reboots, which I'm calling a kernel panic. On the next boot attempt it displays the dreaded "Your computer restarted because of a problem…" If I boot without the eGPU and then hot-plug it, everything looks like it wants to work… it just doesn't. Here's what I get with the Nvidia card: In System Information under Graphics/Displays, I see the AMD Radeon HD 6630M and its display as normal, plus "NVIDIA Chip Model" where I'd expect to see it identify as a GTX 970... Using NVIDIA Web Driver: Using the OS X default driver: I'll try to prove that I've followed the right steps to get here! Boot arguments: [FONT=Menlo]sudo nvram boot-args[/FONT] [FONT=Menlo]Password:[/FONT] [FONT=Menlo]boot-args kext-dev-mode=1 nvda_drv=1[/FONT] My NVDAStartup plist (pastebin.com) My AppleHDAController plist (pastebin.com) My IONDRVSupport plist (pastebin.com) Kext cache - no errors: Kexts loaded OK (hot-plugging): NVIDIA Driver Manager: CUDA Preference pane: CUDA-Z: About This Mac: I've even tried using only 1 x 8GB stick of RAM, then the other one. Makes no difference. I'm not sure what else is useful to provide for troubleshooting but I'm ready to try anything! The hardware looks quite happy in my opened-up AKiTiO box. The card uses one 8-pin PCIe connector, which I've voltage tested and which is illuminating the white LED next to it (it's a red LED if the PCIe slot is powered but the 8-pin connector isn't). I've checked my barrel connector and it's supplying 12 volts. The AKiTiO is happily illuminated at both ends (on-board blue and green LEDs are steady and cable-attached blue LED is steady). However, by this point, it's in danger of being short-circuited by saltwater as I'm close to crying into it. :'(

Akitio Thunder2 stocks dwindling? Hi all. I've been lurking for a few weeks, reading **tonnes** - which I think has paid off because I if I had dived in too early I would have ended up with a Sonnet ECHOPRO-based solution for my Mac mini. It's been really interesting watching the developments, especially the very recent change of direction following helpful inside information from AKiTiO themselves. There are lots of really helpful people on this forum but I'd particularly like to thank Tech Inferno Fan for being so active and answering others' questions so clearly and patiently. So: I have my graphics card now and my GPU. While waiting for my AKiTiO I have also obtained a powered PCIe riser (which I now know I'm not going to use unless I really need to experiment down that path) and I've made a molex-barrel-plug cable (which I now know I'm not going to use because it'll be better to attach my barrel plug to the spare non-molex power line on my PSU which has two yellow leads and should share the load better, etc.). The main thing I wanted to say is that I think I detect a bit of a supply issue with the AKiTiO Thunder2 units. As far as I can see there is only one reseller here in the UK - scan.co.uk. I originally ordered from them earlier in April but the AKiTiO Thunder2 was already on back-order - overdue since 31st of March. I'm not sure whether they're getting a trickle in from time to time and filling orders on a first-come, first-served basis or what, but after a couple of weeks of being thoroughly messed around by them (telling me they had one in stock, that they would send it out, then someone else saying that was a stock error, etc.) I never did get to the bottom of it and I cancelled the order in frustration. They still say they're out of stock. Next I tried a French reseller (Macway.com), and testing my GCSE French I managed to place an order just fine. But after a bit of a wait and making an enquiry with them, it transpired they, too, were waiting for stock from AKiTiO and didn't know when they would be arriving. So again I cancelled my order. Finally, begrudgingly, I ordered from the only place that seemed to actually have some stock: a US company selling via eBay's Global Shipping Program: Akitio 155191 Case T2pc-tia-aktu Thunder2 Pcie Box Without Pci Express Card | eBay In case anyone is interested, the price is $229 / £161. After adding the shipping and handling / customs fees ($79 / £55) , the total I paid was $308 / £216 (The cost of my cancelled order from Scan.co.uk was £183... interestingly I see they are now quoting £199 even though they still don't have them in stock.) So, OK - I still don't have my AKiTiO and I'm watching it crawl achingly slowly by UPS to the US-based Global Shipping centre but it turns out the cost of getting a unit here in England is going to be about £200 minimum no matter how you do it. And I'm really looking forward to putting it all together, after which I'll report back here and hope to produce a guide for others wanting a similar build.