Which graphic card doesn’t require extra power?

[LEOMODE]

I was eyeing on GTX1070 for a while now and I knew that because that was the highest performing card of all within the cards that didn’t require extra power (just needing 2 6pins from cMP that is).

Now I saw that 1070Ti came out and the price is very similar to 1070.

But my question is does 1070Ti require extra power? And what is the highest/best performing card that doesn’t require extra power (as in it will suffice with cMP’s internal 2 6pins) at the moment?

I don’t care about the boot screen now since I have GTX680 and ATI 5870. But I would need to care if that card won’t run on Mac OS yet.

 

[fendersrule]

The highest performing gaming card for the Mac Pro that requires zero power modifications is the GTX 1080.

The 1070Ti is an odd card in the lineup. If you can afford a 1070Ti, you can afford a GTX 1080. Not many people, if any, have tested the 1070Ti in MacOS yet that I'm aware of.

My GTX 1070FE runs MacOS perfectly. My perception is actually that nVidia's web drivers are better than Apple's drivers.

When running an nVidia card, you should stick to the FE/reference designs for maximum potential of not having any issues.

I paid $200 brand new for my 1070 FE. But that's because it was bundled in with the HTC Vive. It's the greatest $200 card I've ever owned, and it certainly feels future proof and plentiful when it comes to 1440p performance. The fact that I can super sample @ 1.3x all VR games that I play says a lot...

 

[h9826790]

If you mean "not require extra power but still within official limit" then I also believe that should be the 1080, not 1070. The reference 1080 only need single 8pin, which rated at 150W, exactly same as 2x 6pin (75+75).

But if you mean "not require extra power, and can function normally regardless the official limit", then 1080Ti / TitanX actually can fall into this category.

I am now running a 1080Ti with just the 2x mini 6pin. So far, only one software can cause a hard shutdown, which is the CUDA-Z heavy test. But since my daily job is not running CUDA-Z (heavy), and none of any other benchmarks / real world use / gaming cause any issue (I did a lot of tests, including Furmark, OCCT, etc). So, I wouldn't worry about it.

Anyway, this is the power consumption test result.

Even though a 6pin is rated at 75W, but that's with a 15% build in buffer. That means all standard 6pin was designed to provide at least 86W. If we take this into account, apart from Furmark, my 1080Ti's power draw actually always stay within limit.

And the interesting part is, a standard 6pin is designed to deliver 7.14A max via two 12V wires. And in the cMP, all three 12V wires are powered. It seems because of this extra 12V pin available, Apple also raise the real world power limit of the mini 6pin to 86W x 150% = 129W.

In fact, real world tests show that once the power draw go above 120W, the cMP's self shutdown protection may activate at anytime. Which coincidentally very close to this 129W calculated limit.

And from the above graph, you can see that my 1080Ti can actually draw ~100W from each of the mini 6pin without any problem. Anyway, since I believe your daily work is not running Furmark or CUDA-Z. Therefore, I will say the 1080Ti / TitanX is the real most powerful card that do not require any extra power.

N.B. The above comment ONLY applicable to the reference card. There are lots of non reference card out there has very heavy factory OC, and can draw way more than a normal reference card does.

 

[fendersrule]

My bro h98, we will have to differ on this one.

The 1080ti is over the official limit, 250 vs 225w. It does require extra power if you want to avoid shutdowns in all conditions. My answer is still the GTX 1080.

That doesn't mean that some could live with it based on their tasks/workflow, but honestly, if I put in a new GPU, I don't want to have any worry if my stuff will ever shut down from what I'm doing, or what I will ever do. I want a GPU with some headroom, so maybe I could overclock or fiddle with it in the future? I don't want to use "extra bandwidth" as "all possible bandwidth". Especially when it comes to a 10 year old machine that's harder to repair as years go on. Having bandwidth (tolerance) is a good thing!

For me, even if I don't have apps that utilize Cuda-z benchmarks, I still would not put in anything more than a 1080 without power mods. Just the fact that a certain card, like the 1080ti, will shut down the Mac Pro in certain circumstances, even if they are rare and/or currently benchmark-only, is enough for me (and probably many others) to not even consider, period! And those graphs show current apps. What about future apps or games? What about Windows? What about future optimizations?

Plus, I don't want h98 to have too much faster of a GPU than me, even if it's too late.

Don't let nVidia's extensive lineup confuse you. The 1070 is a blistering card that most people probably won't even utilize the power of being that many (if not most) still play on wimpy 1080p monitors. The GTX 1080 is plain nuts and is getting into 4k, enthusiast territory. The 1080ti is a pure enthusiast and overkill card, surpassing what 99% of what people actually need, and requiring a loan to obtain. The 1070Ti? I suppose it's a great card, but it's at the same time not relevant with the current age of pascal, prices, and market conditions.

btw, I say all of this having a gutted 27" tapered iMac in front of me for repair. Power supply needs replacement, possible logic board fry from some sort of unknown power issue. Not fun, at all! Don't mess with power with these!

 

[h9826790]

My bro h98, we will have to differ on this one.

The 1080ti is over the official limit, 250 vs 225w. It does require extra power if you want to avoid shutdowns in all conditions. My answer is still the GTX 1080.

That doesn't mean that some could live with it based on their limited tasks/workflow, but honestly, if I put in a new GPU, I don't want to have any worry if my stuff will ever shut down from what I'm doing, or what I will ever do. I want a GPU with some headroom, so maybe I could overclock or fiddle with it in the future? I don't want to use "extra bandwidth" as "all possible bandwidth". Especially when it comes to a 10 year old machine that's harder to repair as years go on. Having bandwidth (tolerance) is a good thing!

For me, even if I don't have apps that utilize Cuda-z benchmarks, I still would not put in anything more than a 1080 without power mods. Just the fact that a certain card, like the 1080ti, will shut down the Mac Pro in certain circumstances, even if they are rare, is enough for me (and probably many others) to not even consider, period! And those graphs show current apps. What about future apps or games? What about Windows? What about future optimizations?

Plus, I don't want h98 to have too much faster of a GPU than me, even if it's too late.

Don't let nVidia's extensive lineup confuse you. The 1070 is a blistering card that most people probably won't even utilize the power of being that many (if not most) still play on wimpy 1080p monitors. The GTX 1080 is plain nuts and is getting into 4k, enthusiast territory. The 1080ti is a pure enthusiast and overkill card, surpassing what 99% of what people actually need, and requiring a loan to obtain. The 1070Ti? I suppose it's a great card, but it's at the same not not relevant with the current age of pascal, prices, and market conditions.

That's why I said it depends on if OP want a card that has publish TDP within the mini 6pin's official limit. OR if he is looking for a fastest GPU that can work.

But anyway, I totally agree that a reference 1080 should be the fastest 100% safe option. 1080Ti, only practically fastest in most case, definitely not always 100% safe (especially we already found a way to hard shutdown the cMP). Also, almost any 1080 can fit inside the cMP's power envelope, but 1080Ti will be very restricted to the reference design non OC card. Anything above that will greately increase the chance of triggering the shutdown protection.

Also, for gaming, a 1080Ti may be overkill. But for compute, it’s never fast enough. In fact, depend on OP’s workflow, an AMD card may work better. At the end, software optimisation play a very important part when we taking about “fastest” inside a particular app.

 

[LEOMODE]

Thanks both!

Well I use my graphic card for gaming mostly, not for productivity.

So I guess my knowledge was wrong (not 1070 but 1080) and 1080Ti can also go but can shutdown.
So for my needs, I can go with 1080 at the highest for Nvidia, but now I'm curious about AMD card.

h98, are you saying that I can install AMD card and run Mac OS and Windows 10 just fine and play games just fine? I know in Boot Camp it is possible but I'm not aware of any AMD driver post ATI5870.

Thanks!

 

[William Payne]

To really simplify things and make answers your question less complicated, your mac pro will have 2 mini 6 pins. 1 mini 6 pin can go to a 6 pin on the GPU, so if you have 2x 6 pin on GPU you will use both 6 pins on your board. If you have an 8 pin on your GPU then you will also use both 6 pins on your board going to the single 8 pin.

Yes there are people running single 6 pins to 8 pins and whatnot but what I described is the "play it safe" way to go about it, everything else is try at own risk.

 

[h9826790]

Thanks both!

Well I use my graphic card for gaming mostly, not for productivity.

So I guess my knowledge was wrong (not 1070 but 1080) and 1080Ti can also go but can shutdown.
So for my needs, I can go with 1080 at the highest for Nvidia, but now I'm curious about AMD card.

h98, are you saying that I can install AMD card and run Mac OS and Windows 10 just fine and play games just fine? I know in Boot Camp it is possible but I'm not aware of any AMD driver post ATI5870.

Thanks!

Yes, RX580 works fine in both MacOS (10.12.6 or later) and Windows. If you are looking for gaming at 1080P 60 FPS, RX580 is the simplest solution indeed. The Pascal card is powerful, but require Nvidia web driver to work in MacOS. If you didn’t do that right. The GPU won’t display anything at all. On the other hand, RX580 does not have this problem, 100% plug and play in MacOS.

 

[fendersrule]

RX580 is a fine card for $200 and yep, it's more mac OS friendly (but IMO, not to any signifcant degree). It's great if you want 1080p @ 60 FPS, any game, guaranteed framerate. It will even do VR pretty good.

But if you're after higher framerate, or after larger resolutions, the RX580 then shows its weakness.