OTOY Forums

Seekerfinder wrote:Sounds good. However, as suggested above, a 'single card' benchmark would help to compare apples with apples. I think there is a need to have that in addition to the 'mine's bigger than yours' club.

Best,
Seeker

In OctaneBench you can select the GPUs you want to run the benchmark on.

Seekerfinder wrote: However, as suggested above, a 'single card' benchmark would help to compare apples with apples. I think there is a need to have that in addition to the 'mine's bigger than yours' club.

do agree with this - to calculate things out ehen building a rig, from having single GPU charts would be very easy =) now these big rigs sometimes run watercooled, sometime non-stock voltages, sometimes mixed with cards & that makes a bit of pain to calculate out for new members (& some come pissed 'because info is not clear enough for them from first sight).

Performance table with some prices & value is something that a lot of new comers will thank for! Some of that could be done by hand (like updating price/value etc.) when we are going to have performance data =)

I am very interested to see how this plays out.

There's a very big project coming up for me soon and I need to know if a 980 is worth the investment... (compared to a 780 ti especially)

I just picked up a 980 today from NewEgg, should be here sometime next week. I will definitely test it when it gets in.

Seekerfinder wrote: Sounds good. However, as suggested above, a 'single card' benchmark would help to compare apples with apples. I think there is a need to have that in addition to the 'mine's bigger than yours' club.

Best,
Seeker

It totally aggree! I neeed the single card, non-overclocked Ms/sec performance for a defined benchmark and alpha shadows ON (it is just the use-case).

There is such a club? Great that's just what I needed to hear. I run on a GTX760m at home.

Where I Drove My Stake - Into The 1st Titan

Because Nvidia's release of the original ("o") reference design ("RD") GTX Titan ushered in a tremendous high water mark for low cost fast GPU computing, I coined a new measure. I call it "TE." "TE" stands for "Titan Equivalency." Since I use my Titans (and GTX 480s, 580, 590s, 680s, 690 and 780 Tis) mostly for 3d rendering, I defined TE in relation to OctaneRender's then current Benchmark Scene (version 1.20) in seconds, using Barefeats test results found at [ http://www.barefeats.com/gputitan.html ] as a base.

How I Drove My Stake - With Mathematical Ratios
What's going on, at base, is that I'm evaluating/comparing render time ratios, using division where the oRD Titan's render time of 95 secs. is divided by the render time of the GPU card in question. Here's my TE mathematical equation:

oRD Titan's render time in secs = 95; 95 divided by render time (in secs) of focus GPU card = TE of focus GPU;
where GPUs faster than oRD Titan will get values greater than 1; and slower cards will get values less than 1.

Why I Drove My Stake - To Make More Informed Purchase and Resource Allocation Decisions, As Well As More Accurate Job Completion Estimates
Since the oRD Titan that Barefeats used in his testing took 95 sec. to render that Octane version 1.2 benchmark scene, (1) one GTX 680 that takes 190 sec. to render that scene gets a TE of .5, or (2) 2 GTX 680s that together render that scene in 95 seconds get a total TE of 1, or (3) 8 Titans that each render that scene in 95 sec., each get a single TE of 1, but combined they have a TE of 8, or (4) as in the case of my GTX 690 (tweaked) that renders that scene in 79 sec., it gets a TE of 1.20 [ 95 / 79 = 1.20253164556962]. Remember to keep the following in mind as number of GPUs rendering together exceeds two. In OctaneRender, your performance will scale linearly when using the same model GPUs with the same setting. How this works with only two GPUs is easy - if one renders the scene in 200 sec., then the both of them will render the scene in 100 sec. But here's were the little tricky part starts: If you want to render that scene in 50 sec., it'll take twice the number of GPUs that it took to render the scene in 100 sec. That means it'll take 4. If you want to render the scene in 25 sec., it'll take twice the number of GPUs that it took to render the scene in 50 sec. That means it'll take 8. So if you keep this math in mind and just mull it over a little, it'll begin making perfect sense.

With this information, I can precisely define, for example, how many GTX 480 SCs it'll take in one of my systems to equal the performance of one oRD Titan or one GTX 690, etc.

Here are more Titan Equivalencies:

My GPU Performance Review

My CUDA GPUs’ Titan Equivalencies*/ (TE) from highest to lowest (fastest OctaneRender/ Benchmark V1.20 score to lowest):

1) EVGA GTX 780 Ti Superclock (SC) ACX / 3gig (G) = TE of 1.319
2) EVGA GTX 690 / 4G = TE of 1.202
3) EVGA GTX Titan SC / 6G = TE of 1.185
4) EVGA GTX 590 Classified (C) = TE of 1.13

Titan that Barefeats tested = TE of 1.0

5) EVGA GTX 480 SC / 1.5G = TE of .613
6) EVGA GTX 580C / 3G = TE of .594
7) Galaxy 680 / 4G = TE of .593

*/ For example, my EVGA GTX 780 Ti Superclock (SC) ACX / 3gig (G) with a TE of 1.319 is 1.319 times faster than the Titan that Barefeats tested, but my EVGA GTX 480 SC / 1.5G with a TE of .613 is only about 60% as fast as the Titan that Barefeats tested. Because of OctaneRender's perfect linearity two EVGA GTX 480 SC / 1.5G with a TE of .613 would be 1.226 (or 2x .613) times faster than the Titan that Barefeats tested and a tad faster than my EVGA GTX 690 / 4G which has a TE of 1.202.

It would be great to know the performance of these maxwell based cards with octane scenes that have displacement materials.
It seems to affect maxwell based cards by a big margin when there's objects with displacements.

Since I also recently got a GTX 980 (in replacement for my 680) I would like to know why is not as fast as we expect it is.
Also, I remember I tried the OctaneBench during the Beta testing, and it was cool. I look forward for the final version.

HI,
If i'm not wrong gtx 980 use cuda kernel 5.2. Cycles in blender uses 5.0 and gtx 980 is not as fast as expected, cuda kernel 5.2 will implemented in blender 2.73 and results should be better. May be it's the same issue in octane.

abstrax wrote:

Derya wrote:Can we please have an official statement regarding the new 980/970 cards from NVidia and why they aren't working as they should? After reading some threads it seems Octane doesn't recognize them properly. As Cuda fueled cards are essential for working with Octane this should have a rather high priority, shouldn't it?

We just got a 980 on Friday afternoon and we didn't have any time to do some proper testing, because we are currently trying to release 2.11. Until we have finished the testing / tweaking we can't say anything specific.

Hi Abstrax, did you have the chance to test the GTX 980 yet? Is there anything you can share regarding its performance in octane?

Thanks