OTOY Forums

@Roubal - how did you make it??
Are you sure you used pathtracing?

Somebody in previous post had 8 GPUs and got 38ms/sec...

I'm having 12,5Ms/sec with 780Ti and 2 X 660 ...

@Roubal - how did you make it??
Are you sure you used pathtracing?

Somebody in previous post had 8 GPUs and got 38ms/sec...

I'm having 12,5Ms/sec with 780Ti and 2 X 660 ...

Yes, I used pathtracing, sure. But I noticed that there is almost no difference with Direct lighting for this scene.

GTX 580 are good. I have read somewhere on the forum that one Fermi core is equivalent to 3 Kepler cores, so the increase of power with recent cards is not actually huge.

BorisGoreta wrote: I wonder if you could stick two Netstor boxes in the main comp and would it work ?

I like the way you think Boris...

Seeker

@Roubal - no argument on my side, but Kavorka with single 580 had almost 4,00Ms/s.

No comparison to your time.
You messed some settings bro'

@slepy8 : Weird, I used the file as it is without moving the camera and I just adjusted to pathtracing... I did the test several times with different combinations of cards, with consistent results... I will try again when I will be on my workstation if I have some time.

EDIT : OK, I have found the trick : The value I got was with the "Octane Benchmark Trench (CLICK ME)" node enabled as it is in the original file. As it is, I get 48.99 Ms/sec with two GTX 580 enabled for rendering.

BUT, if I connect the node to the render target, I get 8.19 MS/sec with same settings. Direct lighting gives 31.58 Ms/sec.

I was using my 660 ti in my system since acquiring Octane last month until it seems Octane pushed it into a progressive failer, where the cuda cores would fail intermittently, then a week later I started to get ogl and screen corruption in modo.
Then last friday I started getting 2d screen corruption as well.
So time for a new video card
My wife and I were away on the weekend and I stopped in at a local retailer and picked up a EVGA GTX780 ti classified ( was the only 780ti they had in stock)
WOW this card is a beast!

Here is the score with my old GTX660 ti
Benchmark scene:

PT - 2.91 with alpha shadows off

Here is the new evga 780 ti classified out of the box
Default clock rates GPU boasted to 1162 mhz with octane going.
Benchmark scene:

PT - 9.60 with alpha shadows off
PT - 8.44 with alpha shadows on

I then overclocked it to
7200mhz on the memory from the default of 7000mhz
and in octane the gpu boast is at 1262mhz an increase of 100mhz
benchmark scene:

PT - 10.14 with shadow alpha off
PT - 8.89 with shadow alpha on

Temperature never got above 58°C while overclocked.
So far I'm extremely happy with the card.
Keith

58C? On air? What speed of the fan have you set? If 100% that must be loud...

kavorka wrote:you have the same exact card that I linked to?
I think the EVGA one is slower.

I got the exact same card you linked to. When I render to clock speed goes to 1160 mhz.
But I still got 7.6 Ms/s

I looked up the old Benchmark to compare.
18 June 2012:

rappet wrote:Hi,
I am very excited letting my new build doing it's thing with OR. This is the result:
16.72 Ms/sec (4x580, OR 2.57, driver 285.62)
4.18 Ms/sec with just one GPU.
To my surprise with SLI enabled it's just a little difference: 16.53 Ms/sec
greetz,

I am very excited letting my new build doing it's thing with OR (release 1.20).
This is the result:

4x GTX classified: 17.69 Ms/sec
2x GTX classified: 8.85 Ms/sec (turned off 2 to compare)

2x Titans Black: 15.03 Ms/sec
2x Titans + gtx275: 15.79 Ms/sec (turned monitorcard on as well)

Greetz,

octane_benchmark.ocs ... the one with DL/PT/PMC present as klickable nodes .. so I didn't change anything. Just load and klick on one of the three selections. This is thus WITH alpha shadows.

DL 56.5 Ms/sec
PT 16.3 Ms/sec
PMC 12.0 Ms/sec

Without Alpha shadows:

DL 62.0 Ms/sec
PT 18.6 Ms/sec
PMC 13.4 Ms/sec

2x 780 Ti (MSI Gaming OC cards)

I don't know if there are several different benchmark ocs out there. The one I used says this in the comment node:
Scene made by Francesco Legrenzi (http://www.francescolegrenzi.com), based on Greeble Richard Rosenman idea (http://www.richardrosenman.com).
Fine-tuned by OTOY.