Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

bax2003

Cancelled
Original poster
Dec 25, 2011
947
203
I have few questions about Late 2013 Mac Pro....and its GPU connections:

- I have two Thunderbolt Displays connected to it, and in whichever Thunderbolt port I connect them (1 through 6), in System Information it says that both monitors are connected to AMD Fire Pro D300 (Slot-2). Is this normal ? There is never display connected to Fire Pro in Slot-1.

16308898054_9668fa7457_b.jpg


- Is there any logic in OS X display numbering ? For example monitor connected to Thunderbolt Bus 1, Port 1 is Thunderbolt Display (2), and other one to Thunderbolt Bus 2, Port 2 is Thunderbolt Display (1)... which to me does not make no sense.
 
Last edited:
Someone can correct me, but simplistically, while the 2013 has dual graphics cards, only one of them is ever used to drive displays under OS X. The other is (only?) for OpenCL compute. So if you're not doing that... I guess it works to rid the world of evil, evil electricity.

Actually that's something I was wondering about - do you get double the number of possible screens when booting in Windows?
 

Excactly, because of that diagram, there is no sense that monitor plugged in to Thunderbolt Bus 1, Port 1 is labeled Thunderbolt Display (2).

Someone can correct me, but simplistically, while the 2013 has dual graphics cards, only one of them is ever used to drive displays under OS X. The other is (only?) for OpenCL compute. So if you're not doing that... I guess it works to rid the world of evil, evil electricity.

Actually that's something I was wondering about - do you get double the number of possible screens when booting in Windows?

I have not tried windows on it yet, but there is one more interesting this: TV connected to native HDMI 1.4 port displays picture but does not show under Displays -> Arrangement.
 
Someone can correct me, but simplistically, while the 2013 has dual graphics cards, only one of them is ever used to drive displays under OS X. The other is (only?) for OpenCL compute. So if you're not doing that... I guess it works to rid the world of evil, evil electricity.

Actually that's something I was wondering about - do you get double the number of possible screens when booting in Windows?

You are correct. The nMP is setup so one GPU drives the displays and one GPU is an offline compute GPU. This offline GPU is not physically connected to the thunderbolt/display output controller. So it is impossible to drive a display with the offline GPU.

The one exception is in Windows you can use crossfire at the driver level. Unfortunately, Mac OS does not have crossfire. You can however emulate it from software, but it will never be as fast.
 
You are correct. The nMP is setup so one GPU drives the displays and one GPU is an offline compute GPU. This offline GPU is not physically connected to the thunderbolt/display output controller. So it is impossible to drive a display with the offline GPU.

The one exception is in Windows you can use crossfire at the driver level. Unfortunately, Mac OS does not have crossfire. You can however emulate it from software, but it will never be as fast.

I doubt that first card does not have anything to do with video outputs....just for driving 6 Thunderbolt Displays (6 x 2560x1440) one Fire Pro D300 (Fire Pro W7000 with 2GB VRAM) can't do that.
How do you 100% know that first GPU is not connected to Thunderbolt (DisplayPort) ? If you have reliable source, PDF or anything, please share it :)
 
I doubt that first card does not have anything to do with video outputs....just for driving 6 Thunderbolt Displays (6 x 2560x1440) one Fire Pro D300 (Fire Pro W7000 with 2GB VRAM) can't do that.
How do you 100% know that first GPU is not connected to Thunderbolt (DisplayPort) ? If you have reliable source, PDF or anything, please share it :)

My source is Apple's Developer information. They released at WWDC 2014 information for utilizing OpenGL/OpenCL on the nMP and how all the internals work. It specifically says in the information they provided the main GPU is the only one connected to the display output.

You might need a developer acc to access the pdf.

http://devstreaming.apple.com/video...f_the_mac_pro_with_opengl_and_opencl.pdf?dl=1
 
My source is Apple's Developer information. They released at WWDC 2014 information for utilizing OpenGL/OpenCL on the nMP and how all the internals work. It specifically says in the information they provided the main GPU is the only one connected to the display output.

You might need a developer acc to access the pdf.

http://devstreaming.apple.com/video...f_the_mac_pro_with_opengl_and_opencl.pdf?dl=1

Thanks for sharing. You are referring to:

What Can You Do Slide:
Use the primary GPU for display
Add support for the secondary GPU
• OpenCL
• Off-screen rendering...

I also understand that GPUs can be online or offline and you can find out with "Finding the Offline Renderer ID". This explains that GPU can be unused or offline, but when you connect 6 displays i dont believe that anything can idle at that time. Interesting part is that System Profiler say that card in Slot-2 has displays connected to it.
 
Thanks for sharing. You are referring to:

What Can You Do Slide:
Use the primary GPU for display
Add support for the secondary GPU
• OpenCL
• Off-screen rendering...

I also understand that GPUs can be online or offline and you can find out with "Finding the Offline Renderer ID". This explains that GPU can be unused or offline, but when you connect 6 displays i dont believe that anything can idle at that time. Interesting part is that System Profiler say that card in Slot-2 has displays connected to it.

Yes, Apple uses the offline gpu paradigm. You can use the offline GPU to render offscreen frames for an application, but those frames have to be copied back to the main gpu for displaying on the screen. This process is slow because you signal this from the software stack and it does not occur at the driver level like crossfire. I don't really know why Slot-2 has the displays connected to it, but it might just be the way the cards were detected. If you get Apple's OpenGL profiler tool you can very easily see what GPUs you have and their setup if they are offline or connected to displays.
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.