New OctaneRender Bench 2.17 baseline is GTX980 - Dual 970s score 155

[MacVidCards]

I replaced the 2nd power connector on my 5,1 today to do some (long overdue) tests on dual 970s.

First and foremost, as reported by a couple of other people, I have found that the power-sipping by the GTX970 allows them to be run from a 6 pin splitter.

I know that some people are going to freak out and assume that I will soon be engulfed in an inferno, but if the readings reported by Hardware Monitor are correct, Nvidia has done a superb job of giving us rendering power without using much electrcity.

I wanted to quantify this and went to see if I had latest version of Octane Render Benchmark.

Turns out I didn't and that they have recently updated to version 2.17 which run s barrage of tests and compares to a GTX980 as a baseline. Which is great since I was trying to find out whether a Dual 970 setup makes sense.

I am also going to test some old favorites like GTX570 and GTX285 as I am fairly sure that a SINGLE one of those will use nearly as much power as 2 @ 970s while offering much less in the way of render speed. (GTX285 may not even run the test)

So first up we have a pair of 970s. The power draw shown was during test. The only way I have found to get these cards to draw more is to run "Furmark", but then only one card draws power. So for overall power draw, this is max I have seen with both running.

Around 125 Watts each.

 

[MacVidCards]

GTX570 2.5GB, the former sweetheart Score 50.26

I was really proud of these when we first introduced.

Note that it was using 170 or so watts running the benchmark.

During Furmark it hit around 230 Watts though likely would have gone higher if I had been able to run full screen.

 

[netkas]

Just need to enable sli in windows then and this setup seems to be best for macpro

 

[MacVidCards]

GTX980, looks like lack of PCIE3 costs us...3% 97 Points = 97%

This is a GTX980 SC from EVGA

Interesting how there are a few places where Mac Version doesn't lose.

Guessing that many of the PC cards are running at PCIE 3.

Nvidia has won the power draw competition, hands down.

Fur mark could only pull 180 or so watts, as you can see, this benchmark pulled 135-140.

And BEAT a regular Titan pulling quite a few more.

This is why being able to change a GPU when tech advances is crucial.

Anyone want to see what happens when we use 2 @ GTX980?

 

[registudio]

Great, yep to see 2xGTX980 ;o)

@Netkas, Why 2xGTX970 would be the perfect solution for mac Pro? I was thinking myself for one GTX Titan x in my Mac pro… for GPU horsepower as well as Open GL / CL performance.


Cheers
Regis

 

[netkas]

Great, yep to see 2xGTX980 ;o)

@Netkas, Why 2xGTX970 would be the perfect solution for mac Pro? I was thinking myself for one GTX Titan x in my Mac pro… for GPU horsepower as well as Open GL / CL performance.


Cheers
Regis

Because it can be powered by internal 2 6-pin adapters only, no need for additional power it seems.

 

[registudio]

Because it can be powered by internal 2 6-pin adapters only, no need for additional power it seems.

Ah ok thank you.
I though one Titan X wouldn't need external power supply, oops my bad it seems I misunderstood it, yike!

Time very soon to change my GTX 770 4gb, but for what then? ;o)

Cheers
Regis

 

[Demigod Mac]

Ah ok thank you.
I though one Titan X wouldn't need external power supply, oops my bad it seems I misunderstood it, yike!

Time very soon to change my GTX 770 4gb, but for what then? ;o)

Cheers
Regis

It doesn't. Haven't tried Furmark yet but it made it through Firestrike Extreme without an issue.

 

[registudio]

It doesn't. Haven't tried Furmark yet but it made it through Firestrike Extreme without an issue.

hmm well very interested, my main purpose for it would be for using Octane, Turbulence FD, and Nuke / After Effects.
If someone can confirm there is no issue with the internal power supply, I would be very grateful.


Cheers
Regis

 

[Demigod Mac]

I can run Furmark tonight when I get home from work and let you know how it goes.

MVC did the same and the Titan X didn't cut off, unlike the original Titan.

 

[registudio]

It would be great, thank you very much

Cheers
Regis

 

[MacVidCards]

Think Different - 2012 rMBP with GTX970

Off to the gym for the daily torture session. When I return I will do Titan-X and some others.

Spoiler: the 980 beat original Titan, at least in this test. I found 980 to give me a 97. So almost EXACTLY what it should do, impressive considering that others running this on PC are likely on PCIE 3.

And if people sitting at work or wherever want to add their cards, great. Best if you also include Hardware Monitor or other app that can track power usage.

I try to do the power screenshot while this benchmark is running. Furmark will take it higher but that doesn't prove much. The power usage is pretty steady during bench, it drops during loading but then goes back to normal level.

EDIT: Just to shake things up a bit, here is a GTX970 running via TB on eGPU with my 2012 rMBP.Makes you think, doesn't it?

(Note the Valley score, pretty sure no 3,1 on the planet can beat that, and this via TB1)

 

[Demigod Mac]

FurMark results attached.

No problems, finished the benchmark and the highest TDP it got on the stress test (according to FurMark) was 86%. Stayed in the 82-84% range most of the time. Titan X is rated 250W so that would be 215W, still below 225W. Temperature peaked at 84˚C.

Worked fine in OS X FurMark too.

 

[registudio]

FurMark results attached.

No problems, finished the benchmark and the highest TDP it got on the stress test (according to FurMark) was 86%. Stayed in the 82-84% range most of the time. Titan X is rated 250W so that would be 215W, still below 225W. Temperature peaked at 84˚C.

Worked fine in OS X FurMark too.

Thank you very much for this test.

Cheers
Regis

 

[MacVidCards]

Titan-X #2 Gets 120 Points

Thanks for those results.

I just ran this benchmark and Furmark.

Hardware Monitor saw a Max of 224 or so Watts in Furmark. The old Titans would have shut down the Mac.

Drew 180 or so during OctaneBenchmark.

So we are clear, I am using a Dual 5680 machine with 88GB of RAM. So, plenty of other power use going on.

And the Titan-X was beaten by Dual 970s, and yes they cost less.

The question then becomes, can you give up 3 slots?

I am going to source some quality 6 pin splitters. I made my own. Once we get the parts together we will offer a 970 kit. Currently have quite a few already done, just need the splitters.

 

[lewnworxx]

EVGA GTX 970SC + EVGA GTX 970FTW combo - 164.87

Here's what I got on my EVGA 970 SC + FTW combo.

At any rate, this is probably the best bang for the buck for render horsepower out there right now.

It's 37.5% faster than a TitanX at 60% of the cost. That practically makes me giddy.

Since then I have picked up another pair of SC's new off eBay from 2 different sellers for $620 shipped total, which I'll drop into another 5.1 as a render slave.

My other two I got local from Fryes, so had sales tax making the total a bit more.

The 6 pin splitters I got off eBay from Jacobsparts @ $6.47 a pop.

I couldn't build em for that, let alone the time to source the connectors and solder them up, so that was a no brainer. I ordered 4 coming out of the gate to combine shipping having a hunch I'd be building a second 5.1 at some point.

It's a rare thing in techno land where one can take a several year old "obsolete" product, and for a very small investment capital wise get extremely good performance, even beating the brand new shiny and uber expensive (nMP) model.

Kudo's to the nVidia guys for not only making the technology in the first place, but for having drivers available for it on a "unsupported" platform. That's a rare thing these days.

 

[benjaprud]

Titan X does work with internal power, at least it works in my MP using 6 to 8 pin adapter. I've tried it with overclocking under Windows and GPU-Z reported a nearly constant 275W power draw.

Dual 970s is a better choice for GPU computing while Titan X is great for non-SLI gaming.

 

[registudio]

Well thank you everyone for sharing the cool tests you've made.

However, even if I do agree with you for rendering best price vs horsepower, it's great if you only use this rig for Octane let's say.

As I'm using as well TFD, and compositing package, 4GB memory will be too just for me and for instance TFD doesn't support multiple GPUs yet.

So my guess is that I will be obliged, sort of speak, to go with a Titan X to get in one card a lot of memory and a good fair amount of Cuda cores.

My existing 4 Titans will be hosted in a PC (for octane renderer engine purpose), and in the Cubix (for Octane renderer engine purpose) connected to my Mac pro, I will definitely take into consideration those results to complete / rebuild a rig.

Anyway thank you again for your time and your efforts to put all these test results available.


Cheers
Regis

 

[QuantumLo0p]

Ah ok thank you.
I though one Titan X wouldn't need external power supply, oops my bad it seems I misunderstood it, yike!

Time very soon to change my GTX 770 4gb, but for what then? ;o)

Cheers
Regis

Can you get one or two more 770's and do SLi? Skipping a generation and doing SLi can get you big numbers on the cheap and being overclocked GK104 680's they are relatively inexpensive and will still post some good numbers even though the TDP is much higher than the 680.

When I did 2-way 680 SLi they were performing as well as a 980 and now I have 3-way SLi which is my motherboard's max. It makes me wish I had purchased the Asus WS version which can do 4-way SLi. Currently doing ~275K points per day for Stanford's folding@home project. I plan to build another system sometime down the road and will use the next gen card unless it is merely an overclocked 980 in which case I'll wait for the next "real" generation. Obviously my build is more for FP calc performance than graphics although Siemens NX really runs well even though they are GeForce cards and not K.

 

[registudio]

I see, but Octane and other softs I'm using don't support SLi if I'm not mistaken.

Thanks
Regis

 

[philliplakis]

Guessing that many of the PC cards are running at PCIE 3.

Octane doesn't score on bandwidth. You should achieve the same score in a 1.1x

The scene is loaded into the cards vram then rendered...

It may be a mac os x issue overall, but the average 980 score atm is 99.

----------

Just to shake things up a bit, here is a GTX970 running via TB on eGPU with my 2012 rMBP.Makes you think, doesn't it?

Hence this score is 3 points above average on the average 970 of 81

 

[MacVidCards]

[/COLOR]

Hence this score is 3 points above average on the average 970 of 81

That is probably the small help that a 650M provides.

 

[MacVidCards]

GTX770 4GB uses 150 Watts to score 57

basically a hot rodded GTX680, when I do a bog standard 680 I will add to this post

So, using same power as GTX980, but only getting 57% of the work done.

Stock GTX680 2GB, gets 51%, uses a tiny bit less power.

So 980 is basically twice as fast on same power. Pretty incredible leap.

 

[philliplakis]

2x 780 ti & 1x K4200 = 232.72

Though i might add my benchmarks...

A little unfair maybe cause i run a 450w drive bay psu but any how here they are
The 780ti in slot 1 is 1071mhz and slot 3 is 928mhz
Boosta A is connected to the K4200 and Boosta B to the 780ti in Slot 3.

The K4200 pulled 90w at peak.

Slot 1- 780 ti = 100.22
Slot 2- K4200 = 39.31
Slot 3- 780 ti = 95.07

 

[philliplakis]

Photo

Pic of the inside