DIY eGPU Macbook experiences

[euqlaog]

I did the following tests:

1) R9 280X -> a powered riser (4pin molex) -> AKiTiO <- molex-to-barrel adapter + ATX PSU

Booted into Win8.1 without an external monitor connected, attached DVI cable to the 280X -> internal and external screen works (Evo*’s method used) -> a few minutes later blank white screen

2) The same configuration as with test 1) but now booted DVI cable attached from the beginning, only the external screen recognized, started a 3DMark11 test, during the physics test (power meter showing ~50W) it crashed the 3rd time, GPU fans started to run full speed. Crashing three times in physics test consecutively cannot be a coincidence.

3) R9 280X -> the same riser but not powered from the 4pin molex -> NA211TB -> 24pin ATX connector -> ATX PSU -> passed the 3DMark11 test as many times before.

Conclusion: Stability is not a AMD Win8.1 Boot Camp driver related. AKiTiO cannot handle constantly the feeded power from a powered riser only / a powered riser + molex-to-barrel adapter combination with the R9 280X. This explains why some cards are more stable and some are not. Those cards that are less dependent on the x16 slot power might work more reliable with the AKiTiO. NA211TB shares stable electricity for all the three PCIe slots. I am not an electronic engineer, but maybe someone has an idea how to properly power the AKiTiO. Maybe it needs soldering to run 100% any GPU. This is just my conclusion.

[Morv]

@gothic860: 290X has a different chip than 280X. Did you also add IOPCITunnelCompatible true value for the “AMDHawaiiGraphicsAccelerator” block in AMDRadeonX4000.kext?

@Morv: You might have a faulty riser. I bought a couple extra of them because I got one from eBay that didn’t work. I could try to power AKiTiO with a molex-to-barrel adapter + a powered riser together from the same power source (ATX PSU), and see if that helps to pass the Physics part in 3DMark11 test.

Confirm that you have installed the latest drivers from AMD: Apple Boot Camp Software Graphics Drivers (14.301).

I thought the same and already have purchased the second riser(which didn't change a thing, btw). I'm not using the Boot Camp drivers, I'm using the usual AMD drivers, 14.12 Omega Catalyst. Same for the GTX 780, usual Nvidia drivers.

I did the following tests:

1) R9 280X -> a powered riser (4pin molex) -> AKiTiO <- molex-to-barrel adapter + ATX PSU

Booted into Win8.1 without an external monitor connected, attached DVI cable to the 280X -> internal and external screen works (Evo*’s method used) -> a few minutes later blank white screen

2) The same configuration as with test 1) but now booted DVI cable attached from the beginning, only the external screen recognized, started a 3DMark11 test, during the physics test (power meter showing ~50W) it crashed the 3rd time, GPU fans started to run full speed. Crashing three times in physics test consecutively cannot be a coincidence.

3) R9 280X -> the same riser but not powered from the 4pin molex -> NA211TB -> 24pin ATX connector -> ATX PSU -> passed the 3DMark11 test as many times before.

Conclusion: Stability is not a AMD Win8.1 Boot Camp driver related. AKiTiO cannot handle constantly the feeded power from a powered riser only / a powered riser + molex-to-barrel adapter combination with the R9 280X. This explains why some cards are more stable and some are not. Those cards that are less dependent on the x16 slot power might work more reliable with the AKiTiO. NA211TB shares stable electricity for all the three PCIe slots. I am not an electronic engineer, but maybe someone has an idea how to properly power the AKiTiO. Maybe it needs soldering to run 100% any GPU. This is just my conclusion.

This may be true. Proper powering would probably be done by soldering the molex connector directly on the Akitio's pcie slot, as shown by nando in the Akitio Thunder2 thread(photos in first thread, he shows the area with a red marker). Problem is that I would need a detailed description so I'd know where to solder which cable as I don't have knowledge of the connection.

Still, thank you for your effort.

[w4vz]

Well... After changing my psu, and installing a powered pcie riser the problems are still there.

So now I have tried absolutely everything except replacing the graphic card but from what I can read on previous post I think that EGPU is an unstable solution and we should be warmed.

I have not seen any report from anyone on this forum confirming to have a 100% stable system when used for things other than Rendering applications.

I just want to be able to play games but my system hangs COMPLETELY RANDOMLY as everyone else still. I will be starting a new thread to answer specific stability questions which in my opinion seem completely overlooked in every guides I have read... Please look for my thread on stability that I will post shortly

[Dschijn]

Well... After changing my psu, and installing a powered pcie riser the problems are still there.

So now I have tried absolutely everything except replacing the graphic card but from what I can read on previous post I think that EGPU is an unstable solution and we should be warmed.

I have not seen any report from anyone on this forum confirming to have a 100% stable system when used for things other than Rendering applications.

I just want to be able to play games but my system hangs COMPLETELY RANDOMLY as everyone else still. I will be starting a new thread to answer specific stability questions which in my opinion seem completely overlooked in every guides I have read... Please look for my thread on stability that I will post shortly

Sorry to hear that. But I have game sessions under Windows 8.1 up to 4h and no problems.

[Oscar J]

Well, thank god I didn't buy the Akitio + R9 270X then. Guess I'll have to wait for a more capable thunderbolt enclosure to be released.

[jacobsson]

Sorry to hear that. But I have game sessions under Windows 8.1 up to 4h and no problems.

Me too, been playing Far Cry 4 from beginning to end twice + borderlands 2 etc etc. =)

[Tech Inferno Fan]

Well... After changing my psu, and installing a powered pcie riser the problems are still there.

So now I have tried absolutely everything except replacing the graphic card but from what I can read on previous post I think that EGPU is an unstable solution and we should be warmed.

I have not seen any report from anyone on this forum confirming to have a 100% stable system when used for things other than Rendering applications.

I just want to be able to play games but my system hangs COMPLETELY RANDOMLY as everyone else still. I will be starting a new thread to answer specific stability questions which in my opinion seem completely overlooked in every guides I have read... Please look for my thread on stability that I will post shortly

Did you try everything? Change video card, change PSU, try a differnet power socket in case have dirty power, plug the video card into the AKiTiO and use a barrel-to-molex adapter to power the unit from the ATX PSU? Perusing the Implementation Guides you'll see users trying various different power configurations.

I'll mention too that in the expresscard side of things we have seen variation of Gen2 tolerance amongst vendors cards. On the very same system, some video cards can run Gen2 (PCIe 2.0), whereas a small number will BSOD out unless run at Gen1 (PCIe 1.x). Affected users seeing a video card swapout resolving the problem.

I believe the same issue will cause problems with Thunderbolt systems.

Well, thank god I didn't buy the Akitio + R9 270X then. Guess I'll have to wait for a more capable thunderbolt enclosure to be released.

w4vz is one of the rare users who hasn't had a straight forward successful implementation. Hardly a reason to forego the AKiTiO given the large number of users who've had a successful eGPU implementation with one. http://forum.techinferno.com/diy-e-gpu-projects/6578-implementations-hub-tb-ec-mpcie.html#Thunderbolt

[Oscar J]

Yeah, but the number of people who have tried Akitio + AMD graphics seems to be countable on the fingers of one hand. And I've heard at least two complaining about stability issues.

[Morv]

Did you try everything? Change video card, change PSU, try a differnet power socket in case have dirty power, plug the video card into the AKiTiO and use a barrel-to-molex adapter to power the unit from the ATX PSU? Perusing the Implementation Guides you'll see users trying various different power configurations.

I'll mention too that in the expresscard side of things we have seen variation of Gen2 tolerance amongst vendors cards. On the very same system, some video cards can run Gen2 (PCIe 2.0), whereas a small number will BSOD out unless run at Gen1 (PCIe 1.x). Affected users seeing a video card swapout resolving the problem.

I believe the same issue will cause problems with Thunderbolt systems.

w4vz is one of the rare users who hasn't had a straight forward successful implementation. Hardly a reason to forego the AKiTiO given the large number of users who've had a successful eGPU implementation with one. http://forum.techinferno.com/diy-e-gpu-projects/6578-implementations-hub-tb-ec-mpcie.html#Thunderbolt

I was beginning with a GTX 480 and it worked for over a month. In this time the system only crashed 1-2 times which I thought might have been the graphics card running far too hot(it was dirty...). I then upgraded to a GTX 780 and the regular crashes began. I've since then swapped the ATX PSU twice(one used from a friend, one brandnew bought, both more than capable of firing the graphics card), bought another graphics card(Powercolor R9 280 at first, got changed to a Asus R9 280 due to coil whining, both the same symptoms) and used "No riser with Akitio PSU", "Riser and Akitio PSU", "Riser and 90° angle", "Riser and 90° angle and Akitio PSU". The riser is powered of course.

I tried different power outlets with different power strips, yet it doesn't change the outcome. Everytime it crashes but the time it takes to happen varies from "at login" over "shortly after boot" to "several hours with gaming, benchmarking and high usage of the graphics card"(previously today...).

Nvidia or AMD doesn't seem to matter.

The only thing I have not swapped is the Thunderbolt cable and I don't see this being the reason. Is it possible that the cable is defect? Is it possible the Akitio hardware is defect?

I also don't see why the Barrel-to-Molex adapter is supposed to help. If the Akitio board only delivers 25W to the pcie socket by design(Akitio PSU is already 60W but they say 25W for pcie slot), how's it changing anything if the barrel plug will deliver more power?

I don't see anything I can do anymore and I will give up at this point. I'll sell the Akitio Thunder2 and probably the Macbook as well, turning back to Desktop + Laptop. It's really sad because I really like the whole concept and was happy while it worked but I want a reliable system.

[w4vz]

Did you try everything? Change video card, change PSU, try a differnet power socket in case have dirty power, plug the video card into the AKiTiO and use a barrel-to-molex adapter to power the unit from the ATX PSU? Perusing the Implementation Guides you'll see users trying various different power configurations.

I'll mention too that in the expresscard side of things we have seen variation of Gen2 tolerance amongst vendors cards. On the very same system, some video cards can run Gen2 (PCIe 2.0), whereas a small number will BSOD out unless run at Gen1 (PCIe 1.x). Affected users seeing a video card swapout resolving the problem.

I believe the same issue will cause problems with Thunderbolt systems.

I just tried the molex to barrel adaptor today, changed my PCIe cable with no result at all. Same random kernel panics..

I only have two more things to do:

Experiment with different drivers or change my graphic card...

This situation is beyond frustrating

Can someone with a working 10.10.1 system who play games tell me which driver they are using ? Im currently using WebDriver-334.01.02f02 but I have the same issue with apple ones so I am pessimistic that this is the problem..

Anyone having glitch issues in safari and ical ?

I have an EVGA GTX770 superclocked, any advice on which one to get instead? I might be able to benefit from the EVGA update program, anyone thinks I would benefit from maxwell achitecture with 970 or 980 ?

[Tech Inferno Fan]

I have an EVGA GTX770 superclocked, any advice on which one to get instead? I might be able to benefit from the EVGA update program, anyone thinks I would benefit from maxwell achitecture with 970 or 980 ?

GTX770 superclocked edition rings warning bells. Quite a few Keplar cards were released with ASICs that could not hold the factory set overclock. If you are booting Windows then recommend setting lower reference clocks with MSI Afterburner or the EVGA Precision tool and do stability tests.

If 'stepping up' then consider grabbing a EVGA GTX970 SSC: Amazon.com: EVGA EVGA GeForce GTX 970 SSC ACX 2.0 4GB GDDR5 256bit, DVI-I, DVI-D, HDMI, DP SLI Ready Graphics Card Graphics Cards 04G-P4-3975-KR: Computers & Accessories . To ensure cleanest possible PCIe electrical signal you'd need to plug this card directly into the AKiTiO chassis. To do that would require either removal of the inner circuit board or bending the right side of the chassis out to allow cards longer than 210mm to fit.

[w4vz]

GTX770 superclocked edition rings warning bells. Quite a few Keplar cards were released with ASICs that could not hold the factory set overclock. If you are booting Windows then recommend setting lower reference clocks with MSI Afterburner or the EVGA Precision tool and do stability tests.

If 'stepping up' then consider grabbing a EVGA GTX970 SSC: Amazon.com: EVGA EVGA GeForce GTX 970 SSC ACX 2.0 4GB GDDR5 256bit, DVI-I, DVI-D, HDMI, DP SLI Ready Graphics Card Graphics Cards 04G-P4-3975-KR: Computers & Accessories . To ensure cleanest possible PCIe electrical signal you'd need to plug this card directly into the AKiTiO chassis. To do that would require either removal of the inner circuit board or bending the right side of the chassis out to allow cards longer than 210mm to fit.

Im running OSX 10.10.1 so no underclocking for me..

I am already running the card inside the akitio with barel to molex adaptor, I had to bend the back so I could not even sell it If I wanted to :/ I tried with powered riser instead too but no luck at all.

Any reason why 970 and not 980 ? I also read that the 780ti is actually more powerful and it is also less expensive, any thoughts on that ?

[Tech Inferno Fan]

Im running OSX 10.10.1 so no underclocking for me..

I am already running the card inside the akitio with barel to molex adaptor, I had to bend the back so I could not even sell it If I wanted to :/ I tried with powered riser instead too but no luck at all.

Any reason why 970 and not 980 ? I also read that the 780ti is actually more powerful and it is also less expensive, any thoughts on that ?

To answer your question: the EVGA SSC is a top tier GTX970 card with it's 201W power target (+10% = 220W) and 1342Mhz boost factory overclock. You can review 'Gigabyte GTX970 G1 Gaming' card benchmarks for comparative performance. Eg: Gigabyte GeForce GTX 970 G1 Gaming review - Introduction shows such GTX970 in some cases outperforming a GTX780Ti. Consider too that the GTX970 cards can be overclocked to > 1500Mhz without breaking a sweat (+12%) where it definitely outperforms a GTX780Ti.

However, you may want to review http://forum.techinferno.com/diy-e-gpu-projects/8619-tb2-macbooks-allow-monitors-used-egpus-osx.html#post117570 to see if perhaps the GTX780Ti is a better supported card in OSX on your machine than a GTX970.

[w4vz]

However, you may want to review http://forum.techinferno.com/diy-e-gpu-projects/8619-tb2-macbooks-allow-monitors-used-egpus-osx.html#post117570 to see if perhaps the GTX780Ti is a better supported card in OSX on your machine than a GTX970.

I have a 15" early 2012 MBPRD so it is thunderbolt 1. Am I affected ? I read the link you provided but they do not have any definitive listing. They also do not list my MBPRD model so can't say. Anyone running an EVGA 970 successfully ?

The implementation hub does not have anything relevant to my setup also :/

[tamer1009]

Did you try everything? Change video card, change PSU, try a differnet power socket in case have dirty power, plug the video card into the AKiTiO and use a barrel-to-molex adapter to power the unit from the ATX PSU? Perusing the Implementation Guides you'll see users trying various different power configurations.

I'll mention too that in the expresscard side of things we have seen variation of Gen2 tolerance amongst vendors cards. On the very same system, some video cards can run Gen2 (PCIe 2.0), whereas a small number will BSOD out unless run at Gen1 (PCIe 1.x). Affected users seeing a video card swapout resolving the problem.

I believe the same issue will cause problems with Thunderbolt systems.

w4vz is one of the rare users who hasn't had a straight forward successful implementation. Hardly a reason to forego the AKiTiO given the large number of users who've had a successful eGPU implementation with one. http://forum.techinferno.com/diy-e-gpu-projects/6578-implementations-hub-tb-ec-mpcie.html#Thunderbolt

Yeah, but the number of people who have tried Akitio + AMD graphics seems to be countable on the fingers of one hand. And I've heard at least two complaining about stability issues.

I have the rMBP 2013 late with nVidia. Using the AKiTiO I was not able to eGPU any AMD GPU. So far only nVidia GPU's worked.

[w4vz]

I have the rMBP 2013 late with nVidia. Using the AKiTiO I was not able to eGPU any AMD GPU. So far only nVidia GPU's worked.

Which Nvidia GPU / make please, OS version and web driver version also please ? is it stable ? Do you game or render or both ?

[tamer1009]

I tested it with the MSi GTX 650. Windows 8.1 x64. And just the latest drivers from nVidia.

My Macbook had the GT 750M. I do both.

[lordromanov01]

RE: http://forum.techinferno.com/implementation-guides/8558-%5Bguide%5D-macbook-enabling-optimus-internal-lcd-mode.html

@Evo*, does this still require an EVGA GPU to work properly? Or would it work the same with any make?

[Morv]

RE: http://forum.techinferno.com/implementation-guides/8558-%5Bguide%5D-macbook-enabling-optimus-internal-lcd-mode.html

@Evo*, does this still require an EVGA GPU to work properly? Or would it work the same with any make?

EVGA cards are able to enable Optimus without any modifications. The guide you've linked is especially for all the other cards which would boot without Iris Pro enabled. By following the guide and applying the stuff described you get Optimus with any Nvidia card.

[Dschijn]

Is anyone using an eGPU Daisychained as the 2nd device in the chain?

E.g. using a HengeDock, which is already using a TB Port to provide extra ports. Could an eGPU be plugged into the HengeDock and still work the same way as without the Dock?

[jt_dylan]

Hi,

I'm looking for help with the following DIY eGPU config:

Config:

Akitio Thunderbolt2 PCIe

GTX 970

Yosemite (10.10.1)

GPU Driver Version: 10.1.43 (343.02.01f01)

CUDA Driver Version: 6.5.37

A NVIDIA Maxwell card with Thunderbolt on a Mac running Yosemite | Le journal du lapin

ran sudo nvram boot-args="nv_disable=1" before install of web driver

modified NVDAStartup.kext after

ran sudo nvram boot-args="kext-dev-mode=1 ndrv_drv=1"

ran sudo kextcache -system-caches

restarted, connected GTX via thunderbolt and tried HDMI (no luck) then pass thru TB with HDMI and checked System information:

found PCI section gave error

Display only showed Intel HD 4000 (which I believe means the web driver did not install correctly)

found under graphics

Intel HD 4000 (Built-in)

NVDA Chip Model (PCIe)

NVIDIA Chip Model:

Chipset Model: NVIDIA Chip Model

Type: GPU

Bus: PCIe

PCIe Lane Width: x4

VRAM (Total): 256 MB

Vendor: NVIDIA (0x10de)

Device ID: 0x13c2

Revision ID: 0x00a1

ROM Revision: preset 1.0.0

Any Idea on where I might be going wrong? I have on ordr a display port to HDMI and DVD-D to HDMI adapters but I think the model should still show up.

Any help you can provide is greatly appreciated.

Regards,

jt_dylan

[jt_dylan]

Also, note this output:

$sudo kextcache -system-prelinked-kernel

kext-dev-mode allowing invalid signature -67030 0xFFFFFFFFFFFEFA2A for kext NVDAStartup.kext

kext file:///System/Library/Extensions/MotMobileUSB.kext/Contents/PlugIns/MotMobileUSBSwch.kext/ is in hash exception list, allowing to load

kext file:///System/Library/Extensions/MotMobileUSB.kext/Contents/PlugIns/MotMobileUSBLANMerge.kext/ is in hash exception list, allowing to load

kext file:///System/Library/Extensions/MotMobileUSB.kext/Contents/PlugIns/MotMobileMTP.kext/ is in hash exception list, allowing to load

kext file:///System/Library/Extensions/MotMobileUSB.kext/Contents/PlugIns/MotMobileMS.kext/ is in hash exception list, allowing to load

kext file:///System/Library/Extensions/JMicronATA.kext/ is in hash exception list, allowing to load

kext-dev-mode allowing invalid signature -67030 0xFFFFFFFFFFFEFA2A for kext IONDRVSupport.kext

kext-dev-mode allowing invalid signature -67030 0xFFFFFFFFFFFEFA2A for kext AppleHDAController.kext

Invalid signature -67030 for kext <oskext 0x7fa70d8d3700="" [0x7fff76834cf0]=""> { URL = "NVDAStartup.kext/ -- file:///System/Library/Extensions/", ID = "com.nvidia.NVDAStartup" }</oskext>

[w4vz]

ran sudo nvram boot-args="kext-dev-mode=1 ndrv_drv=1"

ran sudo kextcache -system-caches

Try running sudo nvram boot-args to check that your boot args contain the line kext-dev-mode=1 nvda_drv=1

I had to do mine multiple times

[jt_dylan]

Try running sudo nvram boot-args to check that your boot args contain the line kext-dev-mode=1 ndrv_drv=1

I had to do mine multiple times

Yep, just tried:

$sudo nvram boot-args

boot-args kext-dev-mode=1 ndrv_drv=1

- - - Updated - - -

Yep, just tried:

$sudo nvram boot-args

boot-args kext-dev-mode=1 ndrv_drv=1

Could this be the issue:

Invalid signature -67030 for kext <oskext 0x7fa70d8d3700="" [0x7fff76834cf0]="" style="color: rgb(51, 51, 51); font-size: 13.63636302948px; line-height: 15.6545448303223px;">{ URL = "NVDAStartup.kext/ -- file:///System/Library/Extensions/", ID = "com.nvidia.NVDAStartup" }

top shows:

kext-dev-mode allowing invalid signature -67030 0xFFFFFFFFFFFEFA2A for kext NVDAStartup.kext</oskext>

[euqlaog]

@jt_dylan: Does "grep IOPCITunnelCompatible /var/log/system.log" output anything on the screen? Maxwell architecture cards don't properly support HDMI (not sure of current Web drivers). DVI or DP -monitor should work. The card should be listed with the correct name under system report. If not, something went wrong.

EDIT: It should say "nvda_drv=1", not "ndrv_drv=1"