nMP dongle video trick

[iPadPublisher]

I'm sorry for posting something that is seemingly already on the forum here somewhere, but I searched around for awhile and just couldn't find it.

I thought I had read somewhere on here that single monitor users only get the extra GPU boost on the nMP if they "trick" the system into thinking a second monitor is attached with some kind of dongle or something.

Can someone refresh me on what that was all about and/or if I just made all that up. Thanks.

 

[CH12671]

I personally have no idea what you are talking about. Search for "headless mac" and you may find what you think you have been reading....

 

[brand]

I'm sorry for posting something that is seemingly already on the forum here somewhere, but I searched around for awhile and just couldn't find it.

I thought I had read somewhere on here that single monitor users only get the extra GPU boost on the nMP if they "trick" the system into thinking a second monitor is attached with some kind of dongle or something.

Can someone refresh me on what that was all about and/or if I just made all that up. Thanks.

Sounds made up as the second GPU is only used for data processing and not graphical processing. At least that is my understanding.

 

[goMac]

Sounds made up as the second GPU is only used for data processing and not graphical processing. At least that is my understanding.

Yeah. Attaching a dongle would do nothing for the first display as far as graphics anyway.

 

[orph]

also the 2 GPU's dont double the vram, thats a common mistake.
they have the same data on both cards 2x 1gvram =1gb vram not 2gb vram.

 

[lupinglade]

also the 2 GPU's dont double the vram, thats a common mistake.
they have the same data on both cards 2x 1gvram =1gb vram not 2gb vram.

I don't think that is correct. D700 2x6GB system shows 12GB IIRC.

 

[Cubemmal]

Sounds made up as the second GPU is only used for data processing and not graphical processing. At least that is my understanding.

No, that's surely not correct. One mystery that nobody has figured out yet AFAIK, including Anandtech, is how the video cards multiplex out through the TB ports. Personally I doubt that a single GPU will drive three 4k monitors over three TB busses, and I doubt it's as simple as one GPU for graphics, and the other for OpenCL.

 

[brand]

No, that's surely not correct. One mystery that nobody has figured out yet AFAIK, including Anandtech, is how the video cards multiplex out through the TB ports. Personally I doubt that a single GPU will drive three 4k monitors over three TB busses, and I doubt it's as simple as one GPU for graphics, and the other for OpenCL.

It is entirely possible that my understanding is not correct.

 

[orph]

well im assuming it's like CrossFireX (tho that is a gaming/windows setup) where two cards mirror the vram.

maybe there able to act in asymmetric operations vram wise ?
im not relay up on that, it's up to the application i gess.
i know blackmagic DaVinci Resolve can use one GPU for display and one for rendering but for most uses you tend to see identical card setups for rendering but as i mentioned im not relay up on that.

so dose it show up as two D700 cards with total vram of 12GB or as two cards each with 6Gb vram (ie showing up as shared memory totalling 12GB or as two independent units with 6Gb vram each?).

ps i dont have a nmp so cant check myself

 

[CH12671]

No, that's surely not correct. One mystery that nobody has figured out yet AFAIK, including Anandtech, is how the video cards multiplex out through the TB ports. Personally I doubt that a single GPU will drive three 4k monitors over three TB busses, and I doubt it's as simple as one GPU for graphics, and the other for OpenCL.

I think you're right. What would be the purpose of Apple going with two video cards, and one is dedicated to helping the CPU and only one is for driving the screen(s)? In that case, they might as well have just had two CPU's and one GPU, and kept improving on CUDA.

 

[Cubemmal]

I think you're right. What would be the purpose of Apple going with two video cards, and one is dedicated to helping the CPU and only one is for driving the screen(s)? In that case, they might as well have just had two CPU's and one GPU, and kept improving on CUDA.

Yes it would be artificial and frankly somewhat difficult to do right, and would be guaranteed to piss off half your user base. I've thought about some experiments to test this - basically by writing some GPU code and testing for "offline rendering engines" in Apple's nomenclature. However since Apple likes to abstract the hardware so much I think it would be difficult to do.

Anyhow I'm quite sure you have two GPU's are your disposal and how they are partitioned is determined by the application. Now how they plumb through the ports is the only real mystery.