octane benchmark

[dave62]

since the gtx580 3GB arrived i started some benchmarking .
these are the first results.

everyone should feel free to use/ improve/ edit the ods (open office file) or maybe lets discus how make a nice benchmark environment:)
i also attachend the blend file which should no be altered to make the benchmarks compareable (at least please mention what you changed if so. like i did with the the sub-d levels in my case).

maxmem2.zip

(145.51 KiB) Downloaded 141 times

first conclusion:
maximum of 15-16mio tris will fit into 3gb vram atm (look into the chart to see the exact vram usage, but dont forget that you can only use 70-80% of your vram at all. idont know exactly, but the sys seems to reserve some vram) hopefully in future releases the preview of how much vram is left in octane will be more accurate.
also dont forget that a higher render res will need more vram too. so always keep in mind to keep some free overhead in your vram.
one more thing is clear: you need a lot ram too! using 8mio quads will take about 16gb ram while voxelizing...

cheers dave

[tehfailsafe]

Nice,
Can you post the obj and the ocs fle ? I don't use blender.


Also, make them all glossy and switch to pathtracing. I'd like to see the side by side of the 580 to 470 considering I have 470 and am about to switch to 580, I'd like to see the difference.

[dave62]

obj file with 8mio polys would be to big. easiest thing would be to download blender from http://www.graphicall.org/builds/ and export the scene as obj for yourself.
or i can upload an obj with only one suzanne and you can make an array for your own if you like?

[tehfailsafe]

Hmm, good point.
I have never opened blender, so even if it's free it would take a bit to figure out how to export correctly and all.
But redoing it as a max scene would definitely not work to benchmark, as it'd be different for sure.

What about the glossy pathtracing on the 580 vs 470? That's the main thing I'm interested in currently since I'm about to buy.

[dave62]

open blender->open file->export as obj ..i think it would be easy to do it. tschakka!! just give it a try:)

-470 vs 580 means sub-d level of 2 (1.8mio quads)
i will render it glossy with pt
..then i need to do it also with glossy and dl

ok wait:)

[dave62]

@tehfailsafe
i updated the chart.
only changing the mat to full glossy doesnt seem to slow down anything. therefore i only add comparisons between dl and pt..

cheers dave

[tehfailsafe]

Thanks!!

I asked for glossy since it seems that the longer I work on an octane scene, the slower and slower it gets. I tend to add more and more glossy so I thought it would slow it down during PT.

But strange, the higher poly version gets 9 ms/s while the lower poly gets 8 ms/s on the 580?
That seems strange, rendering faster with more objects...


Although overall it makes me happier, since it's closer to what I was hoping, a 2x improvement. I was a little worried when I saw the 470 get 11ms/s and 580 only got 17... The 580 is basically twice the cost of the 470, so I was hoping for close to twice the performance. Otherwise I'm selling my 2 470s to buy 1 580 and get worse performance...

But now I can upgrade to 3 580s since I could never get 6 470s to fit in my case!

[dave62]

[quote="tehfailsafe"]Thanks!!
...
But strange, the higher poly version gets 9 ms/s while the lower poly gets 8 ms/s on the 580?
That seems strange, rendering faster with more objects...


.../quote]
yes there must be something wrong in the chart. sorry.
will check it tomorrow, enough benchmarking for today:)
gtx580 has 512 cores 384bit 3GBvram and the gtx 470 has 448 cores 256bit 1,2GBvram
i think the render perfomance wouldnt double but loading is faster too.

cheers dave

[adrencg]

Good to know it can handle all of that. I don't think I've ever used more than 3 million polys in a scene, and that was a big deal to me.

[yoyoz]

Cannot import scene -> out of memory on GPU