EVGA GTX 980 compatibility question

[sabbott]

Hi there,

Long time reader of this great forum, and long-time tinkerer with a few cMP 4,1s.

I'm considering upgrading GPU. I'd like to stick with 6+6 power connectors: for stability, and for compatibility: I've seen 6+8 cards refuse to work with cMPs, and I don't live in the US, so it'd be difficult to change a card that doesn't work.

That means a lower-end GTX980, but I'm a bit confused as to what's available. This looks good:

EVGA GeForce GTX 980 FTW GAMING ACX 2.0 - GS
P/N: 04G-P4-2984-RX

1) Anyone here have any experience with this card?

The 04G-P4-2986-RX is certainly 8+6, 216W as listed by EVGA, but what of the 2984? There's precious little on the internet about this exact card; everything seems to point back to the 2986.

2) Or am I worrying too much and the 6+8 will be fine with internal power? Opinion seems to be quite divided on this forum...

Thanks!

 

[h9826790]

It's not about 6+6 or 6+8, but what the actual power draw is.

For a 980, I don't think it can draw too much and cause any issue.

 

[flehman]

Hi there,

Long time reader of this great forum, and long-time tinkerer with a few cMP 4,1s.

I'm considering upgrading GPU. I'd like to stick with 6+6 power connectors: for stability, and for compatibility: I've seen 6+8 cards refuse to work with cMPs, and I don't live in the US, so it'd be difficult to change a card that doesn't work.

That means a lower-end GTX980, but I'm a bit confused as to what's available. This looks good:

EVGA GeForce GTX 980 FTW GAMING ACX 2.0 - GS
P/N: 04G-P4-2984-RX

1) Anyone here have any experience with this card?

The 04G-P4-2986-RX is certainly 8+6, 216W as listed by EVGA, but what of the 2984? There's precious little on the internet about this exact card; everything seems to point back to the 2986.

2) Or am I worrying too much and the 6+8 will be fine with internal power? Opinion seems to be quite divided on this forum...

Thanks!

I own the EVGA 2981-KR, which is a reference-clocked 980 with ACX 2.0 cooling and uses 2 x 6-pin connectors. I think the only difference between mine and the "gaming" variant you mention is some overclocking. Mine works great - excellent performance, quiet, no problems.

 

[sabbott]

Hi there,

Thanks for the quick replies!

Thanks also for the details, flehman. That would probably be best for my own sense of peace, but these days there aren't that many 980 cards available at reasonable prices.

My understanding was that a 6-to-8 pin converter could result in the card drawing over 75W through a single PCI-e power connector, which is beyond the spec of the Mac Pro's motherboard... am I mistaken, or being overly cautious?

Onwards!

 

[h9826790]

Hi there,

Thanks for the quick replies!

Thanks also for the details, flehman. That would probably be best for my own sense of peace, but these days there aren't that many 980 cards available at reasonable prices.

My understanding was that a 6-to-8 pin converter could result in the card drawing over 75W through a single PCI-e power connector, which is beyond the spec of the Mac Pro's motherboard... am I mistaken, or being overly cautious?

Onwards!

That's correct, but it's been proved that a single 6 pin in cMP can deliver up to 120W.

I won't recommend anyone pull that much from a single 6pin. But a bit more than 75W seem no trouble at all. Many guys here (including me) regularly draw more then 75W from the 6pins.

This post shows you what the internal 6pins can do in real world.

https://forums.macrumors.com/thread...or-approaching-silence.1982499/#post-23120938

 

[SoyCapitanSoyCapitan]

As long as you know you will be stuck with slowish OpenGL performance (equal to a 680), CUDA/OpenCL around 20% slower than Windows, and no Metal apps yet. The Maxwell cards just don't have opportunities to shine on the Mac and the driver is still based on 346 builds for over a year. You will find some people on forums who try to deny all this, they are just online hawkers trying to empty your pockets.

In Sierra there are 367 builds but again, without apps built on decent APIs your 980 GPU is mostly untapped.

 

[pastrychef]

As long as you know you will be stuck with slowish OpenGL performance (equal to a 680),

GTX 680 OpenGL:

GTX 980 OpenGL:

CUDA/OpenCL around 20% slower than Windows,

Mac Cuda/OpenCL:


Windows Cuda/OpenCL:


The Windows result is from a test I ran in the past. I don't have time to run Lexmark in Windows at the moment... Will follow up at a later date.

 

[sabbott]

Hi again,

Thanks for the link to the thorough research on power usage.

Right, lower performance I did know about. Thanks SoyCapitan for keeping it real — I've appreciated your posts from afar before. I'm mostly looking for the CUDA performance, and a bit of future-proofing on the GPU side. (I'm living with an old GTX480 at the moment—lots of power and heat for not a lot of go.) It does look like it'll be hard to stay in Mac-OS land much longer...

Nice to 'meet' you all!

 

[pastrychef]

keeping it real

Which part did you find real? That a GTX 980 performs like a GTX 680 in OpenGL or that Cuda on a Mac is 20% slower than on Windows?

 

[Asgorath]

It's surprising that people don't understand what being CPU limited means. For GPU limited cases, like the ones you posted above, the 980 is well ahead of the 680 as expected. If you ever see cases where the 680 and 980 perform equally, the simplest answer is that the GPU is not a limiting factor and you're CPU limited (probably due to inefficient ports or the heavyweight/slow Apple OpenGL framework).