Dschijn

220W is not enough if you want to play safe. My occed GTX 970 reached up to 240W in Furmark (worst case scenario). http://forum.techinferno.com/implementation-guides/7879-%5Bguide%5D-2013-13-macbook-pro-gtx970%4016gbps-tb2-akitio-thunder2-win8-1-osx10-10-a.html

It will probably be limited by it's bandwidth. Not sure how much your setup might offer, maybe this list will help: http://forum.techinferno.com/diy-e-gpu-projects/6578-implementations-hub-tb-ec-mpcie.html#Thunderbolt

Why should a GTX 970 throttle? It only does if it's reaching the power/temperature target. Mine is boosting all the way to it's set max.

Yeah right. Got it working, rebooted and kext files are not loaded anymore…

Tested OSX on my 13" (late 2013) MBPr and got it working but mainly on the external screen. Internal (with eGPU performance) only if I drag the animated windows into the internal screen. Rebooted…boom…eGPU is gone. That's it! No more OSX and eGPU with a GTX 970 for me. Maybe it will work with a GTX 780 better?! Not that bad for me, because I will only need the performance in Windows anyway.

Had to change it to (hd2,gpt1). Will not do the OSX testing on my SSD in the 15" (2014, Iris only)… Will try it on a test partition on an external drive. On my 13" (late 2013) the GPU is deteced now and I will do some testing now, maybe use that "test drive" on the 15".

I can do some testing tonight. Was focusing on my new 15" 2014 MBPr and my replacment card EVGA GTX 970. With the 2014 MBPr the eGPU is not working in OSX at all... nobody knows why for sure. But I can run some test with the 13" (late 2013)! I am not sure if you can use the internal screen in OSX with the power of the eGPU. Just got the Benchmark Valley working by dragging it in the window mode from the external to the internal screen.

Wow… ok. So it is a problem of the mid 2014 MBPr? That is crazy. Ok than I will just use it in Windows 8.1.

I got my system working in Windows 8.1. MacBook Pro Retina 15" (Iris only, mid 2014) with a EVGA GTX 970 FTW and a AKiTiO. But I can't get any screen detected in OSX… the card is installed an listed but no display (DVI or HDMI) is recognized. So in the system report the card is listed as a GTX 970 but any screen connected by DVI/HDMI is NOT listed and not working. Any ideas?

Like I said before. On a desktop PC you might get a good performance boost 980 vs 980Ti, but on a MacBook (limitung factor 1) with Thunderbolt 2 (limitung factor 2) and only using the internal display (limitung factor 3) you will not get such a good effect. Of course the 980Ti will be faster, but in my opinion the money is not worth the small improvement. You can play on the internal display in Windows 8.1 (100% chance). OSX I am not sure. You can play on the internal display in Windows 8.1 (100% chance). OSX I am not sure.

Awesome guide! Helped me a lot with my replaced graphics card and the new setup system! Had to do the GRUB fix Edit: Unfortunately I can't get any screen detected in OSX… the card is installed an listed but no display (DVI or HDMI) is recognized. Any ideas?

If you have the money… get a 980Ti. If you want to go for price&performance than you better get the 980, as you will lose performance by using the Retina display and the TB2 connection. The 980Ti performance boost might be very small on your setup.

Yes it is enough. Just be aware that it is just providing 1 x 6+2-Pin and you might need a splitter cable.

This is where I bought it: [PCIE] 16x - 16x PCI-Express POWERED Riser Flex

@dellcorreo: The powered riser is recommended to avoid using the AKiTiO and it's PSU beyond its specifications.

I wouldn't wait for the Ti. Two reasons: - TB2 is limiting the performance - Optimus (to play on the internal screen) is limiting as well In the end the 20% performance boost in reviews can not be achieved with your setup. Better save some money instead of a small performance boost.

Yes it has two capacitors and it's working great!

@Sachin10: exactly!

Great news and great guide. thx! But the fan noise is a no go imho…

Hehe, I was confused like you as well in the beginning. The molex powered riser can provide power to the PCIe in both directions: to the GPU and to the AKiTiO itself. I don't need to power the AKiTiO in another way.

No, I power my AKiTiO with a powered riser. No more need to power it by the barrel plug. An answer to your question regarding the barrel adapter… I think it is written somewhere in this thread.

I had my AKiTiO working with it's original PSU to do some measurements. But imho it's better to play safe. 90W should be enough… in the end you can try limiting the powe target by software (in Windows) or by a BIOS Mod (don't know if the new chips already can be flashed).

You are right, maybe I wasn't precise enough. The power switch should be before any power can reach any part of the GPU, I think the easiest would be a normal power plug with a additional switch.

Be aware that you must use an external display as well to gain all the eGPU performance in OSX 10.10. I think the internal screen can't use the eGPU, if it's used on it's own. The MB can normaly be booted without the eGPU. I wouldn't hotplug the eGPU as Thunderbolt is a PCIe connection and I think that the OS needs to detect the installed graphics hardware. So a very comfortable docking of a sleeping/running MB might be risky. I would advise to shut down, plug in the eGPU and fire it up. Best option to avoid the bully Murphy!

The power the AKiTiO is using itself is neglectable! More interessting is the power needed in total by the GTX 750 OC. As most of that cards use less than the 75W the PCIe slot is providing you might end up well powering both. Maybe you need to limit your powertarget of the card if the OC is too brutal. I would give it a go! The origin of the ground loops is mainly from bitcoing, as there are people using (sometimes) up to 4-8 cards on one Mainboard. With high end cards they ended up using mostly more than one PSU and just one of them was powering the mainboard with it's PCIe slots (mostly up to 75W). By using the power of the PCIe slot from PSU 1 and the 6/8-pin power from PSU 2 the ground was not in common. For some people that ended in fried mainboards.