[Solved] 10.12.6 Fully Working Dual GPU Build, Third GPU -> Black Screen Again?

[futurus]

Fresh Sierra install here.

Quick background - I used this identical hardware configuration up to 10.11.5 and had the same three discrete nVidia GPUs, alongside Intel HD 530 onboard graphics via DP@60hz for a year or so with little issue. Consistent work done in CUDA accelerated programs (Adobe, Blender, etc.) with all three cards chugging away.

That said, the most recent Sierra nVidia Web Drivers work perfectly with a GTX980 Ti FTX Gaming ACX 2.0+ and GTX 970 G1 Gaming, with the 970 G1 utilizing a Display Port output for main monitor. Acceleration works across CUDA programs, no artifacts noticed so far. Bios initial display output is set to PCI (Gtx 970 G1).

The one issue I'm encountering is trying to install my third card, a GTX 980 Ti Kingpin. Upon installation, the black screen boot problem returns (currently fixed via Lily method). I've attempted the 3 GPU install in different PCI slots, different cables, different monitors... all to no avail. Have also attempted just using the 980tis with the same result. I can, however, boot the 3 GPU setup with nv_disable=1, with no acceleration and low-res output.

Internal graphics on my Mobo are disabled. I've been looking at SSDT patching via this thread: SSDT GPU Injection (not extremely comfortable doing this myself and don't want to completely break this build).

Seems to me it has to be something in the way Sierra is interpreting my PCI slots....

Any help/insight/ideas would be appreciated!

current config.plist is attached for reference


UPDATE 9/13/2017: Replaced the 980 Ti Kingpin with a new Gigabyte 1080 G1 Gaming. Booted up with no issues immediately. From what I've been seeing elsewhere, probably best to avoid the Kingpin... I bought it as it was on sale and wanted another 980Ti, but it's proving to be a finicky one. Loving my 1080 replacement as it's slimmer, way fewer PCI power cords, and solid CUDA performance in 10.12.6 so far.

 

[beep]

Nice setup

Have you tried Octane benchmark?
https://render.otoy.com/octanebench/

Did you encounter any on screen artifacts after heavy using Blender CUDA rendering?

 

[futurus]

Nice setup

Have you tried Octane benchmark?
https://render.otoy.com/octanebench/

Did you encounter any on screen artifacts after heavy using Blender CUDA rendering?

Nothing at the moment @beep. Granted I've really dug my heels in on cooling this build and I imagine that has something to do with. The biggest issue I have with the GA-170X-UD5 is the proximity of the second and third PCI lanes. While they support 3 cards, the bottom two slots are RIGHT on top of one another. I've found a pretty decent DIY - I took two Hearoes Ear Plugs and flatted them out, placed them gently in the corners between the two cards while slotted. It looks awful, but that extra breathing room took a few degrees out of my GPU temps.

I've run all three cards in Blender 2.78RC2 and the most recent 2.79 build for several hours each. Everything has been flawless.

For reference, Octabench results attached. Looks like an average score of ~380.5. I've never used this to test GPU setups so I'm not certain what to make of the result, other than I guess it means I'm faster than a 980 which they used for the 100 value reference score (I should hope so!)

Also, my geekbench CUDA tests:

View media item 189418

 

[beep]

Thanks for the info @futurus. Your Octane benchmark looks OK! Mine two GTX980Ti's score 250 pts.