Dual graphics with stock cMP

[mattspace]

Hi,

Just out of curiosity (regardless of whether it's practical from a price / performance perspective), is there a current option available that can get dual graphics in a stock cMP, within its safe power envelope, that performs as well or better than the nMP dual card setup?

 

[ActionableMango]

There are many options. If you are risk adverse, most any card with a single 6-pin power connector can be doubled. Apple even sold the cMP with this option back in the day (two AMD 5770s).

Whether or not it is practical is debatable. You can make use of both cards for compute tasks like OpenCL or CUDA. OpenGL won't double up though--there no Crossfire/SLI equivalent in OS X.

I think a better approach would be to describe what you are trying to do and ask for recommendations. In many cases a single card might be faster than two that will fit in the same power envelope anyway.

 

[mattspace]

I think a better approach would be to describe what you are trying to do and ask for recommendations. In many cases a single card might be faster than two that will fit in the same power envelope anyway.

Oh it was purely a thought experiment - i'm still on the stock gt120, though the recent post about Aperture problems with the Nvidia web driver has given me pause when considering where to go with a video card. I was just wondering if there was a turnkey option for dual cards that can match the nMP version.

 

[p.l]

There is,
You can run 2x GTX 970 internally

There are a couple of users who did this.

Stick to reference design as they will stay within the 145w draw.

 

[h9826790]

I am running I dual 7950 on my cMP now. It works without any trouble. However, I did some tricks to avoid pulling too much power via each 6pin. (E.g. Lower the voltage, and set a lower hard limit of max TDP in the VBIOS)

For stress test (e.g. Fur mark), it can pull close to 100W (AFAIK, the self shut down protection kicks in at about 120W). And for real world OpenCL application (it rarely draw anything close to 75W).

For gaming under crossfire, around 85W max.

In terms of performance, it's very close to the dual D700 in Luxmark (stock frequency 800/1250). If I set the frequency to 850/1375 (same voltage and power limit), it can beat the Dual D700 by a tiny bit.

TBH, it really depends on what you need. For FCPX, this works really well, always within the power envelope (~55W max). But if you want CUDA, then dual 970 should be a better choice.

For the cost. Each new PC 7950 (R9 280) cost about $150, most of them are easy to flash and mod to make them perform exactly the same as the 7950 Mac Edition card. Also, use native driver (this is the main reason why I stick with these cards).

For dual 970, you need OSX 10.10 or later + the a Nvidia web driver to make it work.

 

[AlexMaximus]

I am running I dual 7950 on my cMP now. It works without any trouble. However, I did some tricks to avoid pulling too much power via each 6pin. (E.g. Lower the voltage, and set a lower hard limit of max TDP in the VBIOS)

For stress test (e.g. Fur mark), it can pull close to 100W (AFAIK, the self shut down protection kicks in at about 120W). And for real world OpenCL application (it rarely draw anything close to 75W).

For gaming under crossfire, around 85W max.

In terms of performance, it's very close to the dual D700 in Luxmark (stock frequency 800/1250). If I set the frequency to 850/1375 (same voltage and power limit), it can beat the Dual D700 by a tiny bit.

TBH, it really depends on what you need. For FCPX, this works really well, always within the power envelope (~55W max). But if you want CUDA, then dual 970 should be a better choice.

For the cost. Each new PC 7950 (R9 280) cost about $150, most of them are easy to flash and mod to make them perform exactly the same as the 7950 Mac Edition card. Also, use native driver (this is the main reason why I stick with these cards).

For dual 970, you need OSX 10.10 or later + the a Nvidia web driver to make it work.

Very interesting! I would love to know how did you solve the problem with the additional two power leads that are needed to run the second AMD 7950 card? Do you use a second power supply in the CD Rom spot as some users used to do?

And most important to me, how much more GPU power do you have through crossfire in Windows for gaming purposes ?

Thanks in advance,

 

[h9826790]

Very interesting! I would love to know how did you solve the problem with the additional two power leads that are needed to run the second AMD 7950 card? Do you use a second power supply in the CD Rom spot as some users used to do?

And most important to me, how much more GPU power do you have through crossfire in Windows for gaming purposes ?

I simply run both card like this (by mini 6pin -> 2x 6pin cables).

For crossfire, it really depends on the game. The gain could be from zero to around 80%. In general, dual 7950 is good enough for 1440P 60FPS (V-Sync on) max setting without AA (applicable for MGS V, TR, GTA-V, Metro, Assassins Creed, sleeping dog...).

To simulate the power draw in Windows gaming with crossfire on. I run the Valley benchmark on one GPU and the Heaven benchmark on the other GPU in OSX (it's much easier to monitor the power draw in OSX). And I have the following result.

As you can see, after the GPU warm up. The power draw from each 6pin supply is around 85W, peak at around 90W. That's above the official limit 75W, but well below the Mac Pro's self shut down protection 120W limit. I personally happy with this power draw. Even though I won't recommend this config to anyone. All I can say is that I run like this for more than a year now, and everything still good.

Also, this is 100% continuous loading, which would not happen in real word crossfire gaming. So, the actual power should only be lower than this test result.

 

[RoastingPig]

im running a gtx 980 and a gt 120 together