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

This is not about delivering power (w) but enough current.
Radeon hd 5870 - 40a (example)
geforce gtx 970 - 28a

so you recommend cmp users to draw 56a from two 6-pin pcie connectors ? Very smart.

56 A?

U = R * I
I = P / U
I = 150 W / 12 W
I = 12,5 A
12,5 A * 2 = 25 A

So you have 25 A with two 970
 
I recently found the plug so today I started 970 power testing.

And I am STUNNED.

[...]

I have so far been unable to get it to use more than 140-145 Watts.

Prepare to be stunned...the power draw of 274 Watts is everything on Line 1.

Great!! Thank you for the investigation and the other thread as well. Seems dual GTX970 is the best bang for your buck at the moment if one wants to stay within the power draw manageable by a stock cMP.
 
In honesty I have been dragging my feet for a simple reason. At some point in recent past I thought I had disconnected some power cables and hadn't.

Result being I ripped the plastic plug off the logic board for one of the PCIE power cables. The pins were still there, but the plug itself gone.

I recently found the plug so today I started 970 power testing.

And I am STUNNED.

I have seen GTX285 and GTX570 draw 115-125 Watts from a plug routinely. And I use these as examples as they are both typically dual-6 cards that are considered "safe" in a Mac Pro.

GTX970s start as "reference" cards built on a GTX980 PCB, just with fewer parts soldered on them. In my opinion these were going to draw more power than the custom cards. My favorite card is the PNY GTX970 as it carries same full complement of 3 @ MDP ports. Many cheaper 970s lose 2 of the DP ports which I think people will regret in the future.

Anyhow, I decided to start testing with the reference card as I assumed it would draw more power.

I have so far been unable to get it to use more than 140-145 Watts.

I set Furmark to 5120x1440 and ran it multiple times. No matter what, the test not only finishes every time, it NEVER draws more than 97 Watts from the cable. (I have a dual 6 pin splitter cable in) Meanwhile, i never saw it grab more than 45 from the slot.

Unheard of !!! So, faster than a GTX570 while drawing 80-100 Watts less power. Incredible !!!

Going to do some additional testing with 2nd 970 in.

Note that the app "Hardware Monitor" has the "Line" and "Boost Line" categories swapped.

Moving on to dual card testing, obviously Frmark is a single card test so doubtful I can get both drawing this much.

And this really will mean that we might be able to use 2 of them internally powered.

Thanks again to Nvidia for continuing to support us despite Apple not caring.

Will start a new thread as this isn't related, but wanted to answer the question.

Prepare to be stunned...the power draw of 274 Watts is everything on Line 1.

He is right, I run a 970 from a Dell Server Powersupply and I have never seen more than 100 watts being drawn from both the PCIe power plugs ever!. Hell, I have mine connected over thunderbolt to a G46VW with a 3840QM @4ghz, I get easily over 3.5k/sec in luxmark under the default scene all while drawing <250watts from the wall (laptop and thunderbolt'ed hehe 970) It's truly amazing. I know he was using a 980 but a 970 is still a power efficient card too

S2dmCIx.png


how much more power efficient can you get?! oh wait.. until the next gen GPUs come out hehe. Sorry slightly off topic here
 
Just found some settings in FurMark I'd overlooked.

Ran it at 99% TDP most of the time (at one point hit 101%)
Completed benchmark. No shut off, no problems.

I guess they only included the 8 pin connector for overclockers?
 
56 A?

U = R * I
I = P / U
I = 150 W / 12 W
I = 12,5 A
12,5 A * 2 = 25 A

So you have 25 A with two 970

Written like someone that has never designed any hardware that uses DC/DC converters or linear regulators. I made the same error once very early in my career.
 
Written like someone that has never designed any hardware that uses DC/DC converters or linear regulators. I made the same error once very early in my career.

I made a simple calculation, no line resistance, reactive power or some other factors. But this calculation ist enough to see whats going on.
 
Elite Dangerous is finally out today on Mac.
This it is a proper native port of a great game (and demanding game).
BierFest ahem Barefoot should include it in his benchmarks instead of the awful cider ports he usually benchmarks.
 
Elite:Dangerous

Agreed, would love to see benchmarks on this. Bailing on my '08 Mac Pros with dual 7970s in order to switch to a Linux farm (password auditing). Will keep one Mac Pro for a personal workstation. The question is, upgrade to tower with 970 / 980 or trash can with D700s?

Tower seems a much better value proposition as I don't require thunderbolt. Anyway, no rush - and it's great to see more games being ported to the mac. Disappointed they never did it with the new MechWarrior Online.
 
After seeing this writeup I am considering buying a GTX 980 for my 5,1 and got the attached power specs off the web. It seems this powerful card doesn't draw much wattage at all. Does this mean that I don't have to hack the power system in any way, but simply attach two 6-pin connector power cables to the aux mini 6 pin power connectors on my 5,1 backboard?
 

Attachments

  • GeForce_GTX_980___Specifications___GeForce.jpg
    GeForce_GTX_980___Specifications___GeForce.jpg
    30 KB · Views: 214
After seeing this writeup I am considering buying a GTX 980 for my 5,1 and got the attached power specs off the web. It seems this powerful card doesn't draw much wattage at all. Does this mean that I don't have to hack the power system in any way, but simply attach two 6-pin connector power cables to the aux mini 6 pin power connectors on my 5,1 backboard?

Yes, that's right.

Running MVC gtx980 in 3,1 mp here, works great on internal power. Can even overclock it to get more fps in gta5 :D
 
http://barefeats.com/mpt5k1.html

The D700s were beaten by 100% in 3 of 6 tests. (Or you could say their score was 50% of the Titan-X)

The 8 Core D700 did manage to come from behind and win one test, a Tessmark tessellation test at 16 levels. When the heat got turned up to 32 levels, the Titan-X took a 10% lead.

I am rather surprised that the D700s managed to fare reasonably well in the few tests where they hadn't lost sight of cMP's taillights.

I imagine that either PCIE 3.0 bus speed or having a 2nd D700 at hand helped there, one way or another.

Oh, the iMac 5K was in the test too, it even beat the nMP in half of the tests.

Those D700s are really showing their age.

But when you see graphs like the attached, it makes all of the nMP apologists apologies seem especially hollow. The nMP was left for dead half way across the bar graph. So if you were running something that required OpenGl, why would you go with a machine that ran out of steam so easily?


Hi. Sorry to bump up an old thread with possibly an old topic.

MacVidCards: I see that even Furmark is running on a cMP with Titan X, on only internal power. Just for the record, are you certain that the Titan X doesn't have current spikes that exceed the 75w per auxillary connector, which limitation is because of the link through motherboard traces (don't want to melt them)? You sell the card with a 6-to-8 pin converter for one power channel. The card is rated 250w, which exceeds the 225w limit in cMP.

I will send the same communication to Rob at Barefeats.

Just want to be sure before sending you the money. :)
 
The Titan X does exceed the official 225W limit but the Mac Pro is made to stand a bit more. While in normal usage it draws about 230W, power hungry tasks like Furmark pumps as much current as possible which is strictly limited to the official 250W TDP by the card's power management. On Windows it is possible to raise the TDP to 275W but even in this case the MP is able to power it. The only drawback is that the PSU and PCI fan speeds are related to PCIe power consumption and can get very loud in this case. Modding the PSU can help to pull down the fan speeds under heavy loads.
 
TDP is one thing, load balance between PCIe slot and AUX power is another one. There is a voltage controller on each AUX power. It's factory programmed to cut off when draw from the socket will reach 10A (IIRC from its datasheet).
 
Ok...one thing...Titan X could hardly be the "double precision king" in the cMP, it hardly has any double precision capability at all. Titan, Titan Black, Titan Z are all many, many times more powerful in double precision compute. And so are lots of AMD cards, not new ones, but cards that are a year, 2 years, 3 years old. The Titan X can only manage about 220 Gflops in DP, if I recall. For comparison, Titan Z can hit about 10-11x that.

FP64 is hardly just for E-Peen, it's very useful for lots of things. It's not useful for games or for rendering, but it's very useful (and absolutely required) for any scientific applications, which lots of people use these machines for.

Just wanted to clear that up, as I saw that mentioned earlier and assumed MVC meant "single precision king".
 
I'll watch my language, if you watch the facts.



No, I haven't said that.

You claimed two GTX 970 requires extra PSU. That's false.

I claimed two GTX 970 can be fed from the PSU present in a bog standard Mac Pro, which the two setups in the previously mentioned thread proves.


You claimed that total power consumption would rise to 1200W.

In the thread I've linked to the total power consumption amount to about 400W when the GPUs are under render-load.

This is true, somewhat. A single 970 requires 145 watts of power:

http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-970/specifications

So at worst, two cards will draw about 290 watts. Then whatever else the system usually eats up (CPU, Mobo, RAM, devices, HD, SSD, etc).
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.