SPECapc for 3ds Max™ 2015 - GPU Scores

[Dimitris Tolios]

After procrastinating for too long, I've published results with the SPECapc for 3ds Max™ 2015 benchmark on my lil blog.

 

SPECapc for 3ds Max™ 2015 - GPU Scores @ PCFoo.com.

 

Feel free to comment and draw your own conclusions, and also publish your own results when you get a chance.

A word of warning: the test process is lengthy, so be armed with patience and don't start this before or close to the end of your work day.

 

Cheers

[zlatkopetrovic]

As always, quite informative and useful. Would you say that 970 is better than 780 (non ti) regarding max/vrayrt?

[numerobis]

Thanks for the results! Very useful.

Did you run some tests with CPU at stock? Would be interesting to see how it scales with overclocking.

 

"and separately the “large model” scene score for each CPU."

I think this should be GPU?

[Dimitris Tolios]

numerobis, thanks for identifying the typo. Will run a test with the CPU @ Stock, probably with the 970.

 

Zlatco, I don't get any free samples, so I am forced to buy and sell the cards after some point (already way too many $ reserved)

I did not keep my Titan long enough for the Specapc 3DS 2015 to come out, so I did not have even a single run with it. The only GTXs in my arsenal now are the 970 and a 750Ti, and unless i find a 960 in a good price, I don't think I will be adding more results. I plan on expanding with some Radeon R9, starting with the R9 270 2GB and following with a R9 290 in the close future.

 

I did tho have a few runs of Specviewperf 12 with my Titan, and I can confidently say that the 970 does beat it in most OpenGL tests as far as viewport acceleration goes, and matches it in D3D.

As far as compute goes, the 970 is performing strong, but the results vary between applications. The only real test I know outside Chaosgroup's VRay RT test scene that you can find in their forums, is the one from BOXX, that compares VRay RT Cuda on a custom scene one of their clients provided: http://blog.boxxtech.com/2014/11/17/geforce-gtx-rendering-benchmarks-and-comparisons/

 

I've already commented on their post, that as far as "absolute speed" and a single GPU and Vray RT GPU, the Titan Black (and probably to the same extend the 780Ti) is still the fastest kid on the block. But the value of a 970 is undeniable, as you can get 3 970s and spare some change for the cost of a Titan Black. I would expect 970 to be very comparable in compute performance with the vanilla 780.

 

If I had a 780, I would probably not replace it with a 970. If I wanted more compute performance, adding a 970 to your 780 system (using either as a primary display card) would be more cost effective and faster than going for a single high-end Titan, unless you have to have the 6GB buffer.

Edited February 24, 2015 by dtolios

[NBraz]

Thanks for the benchmark results Dimitris! What kind of a framerate jump can one expect from a 750ti to a 970? I mean, how many poly does each handle? I had a GTX 580 and bought the 750ti when it came out and there was absolutely no difference in max viewport performance for 1/4 of the power savings and so i ditch the 580. But nevertheless i'm curious to know if you notice that much of a disparity between the 750ti and the 970 in max viewport performance.

[numerobis]

The 970 would be great! Thanks!

 

Btw. how do you calculate the results? I only get these XML-files and when i open the file called Results Workbook in Excel i get lots of errors and no values. But maybe my Excel 2003 is simply too old to handle the files...

[Dimitris Tolios]

You have to open the "virgin" specAPC2015.xlsm provided, enable editing and click the "Select Test Result Folder" button in order to browse and guide the macro within that file to calculate the composite scores for you. Don't save over the "virgin" file, as it appears not to be working after you do so. I just repeat the process and make notes of the results on a separate xlsm file where I average them out.

 

Admittedly I've only tried it with Office 2010 / 2013, so I don't know if Office 2003 will have issues with the macros in these files.

 

Each benchmark run is repeating the same tests 3 times, and I usually do 3 runs averaging them out - so we have 9-12 runs averaged out for the numbers posted.

 

Also not that the benchmark is not bug free, and it often hangs up on the completion of the 1/3 round of tests, never to complete the other two. Was a major pain to have slow cards complete it once, moreover 3 times.

 

@Nuno Braz,

The 970 appears to be much faster than the 750Ti in the Large Model sub-score than in the composite total score. It is hard to quantify the performance difference you will experience using it and tying it to millions of polygons and whatnot, as this is dependent to the specifics of each scene, how it is organized with groups / instances / proxies etc, as the adaptive degradation engine "works in mysterious ways".

 

Thus I feel more confident making predictions through reading results from a test that claims to be standardized like the Specapc, instead of the various qualitative methods which are either not objective, or tied to the specific organization of a scene that feels natural to me, but might be completely different to the one your workflow would produce.

 

E.g. a card might it might do great with a 100M poly scene that has a few dozen instances of a high-poly component, spreaded over a small area, vs a 100M poly scene with thousands of different objects, and massively better than a 20M poly scene that has just raw, unorganized geometry - say something you could get after importing a badly prepared model from an external source. Or a slow card with a big buffer might do better in a texture heavy, medium poly scene than a faster, older card with smaller buffer. But what makes a scene high or low complexity, or a texture size above average, is something arbitrary.

 

Does that mean that a 4x better score in Specapc will translate in 3-4x the performance in real life?

My guess is probably no, it won't. But I would thing that something that much faster, should at least give you tangible benefits under some scenarios.

Edited February 25, 2015 by dtolios

[numerobis]

Thanks for the info.

I have tried it again directly with the specAPC2015.xlsm file of the submissions folder, but it is the same file as the Results Workbook file from the start menu and I always get a runtime error when i click at 'Select Test result Folder'. I will try it with a newer Excel version. Thanks!