GPU benchmarks

[Alan Iglesias]

mmm... Autodesk is the antinomy of improvement.

??
I was talking about May's Vray v3x SP...

-Alan

[fraggle]

mmm... Autodesk is the antinomy of improvement.

And timing-wise, it might be a good idea to implement the new file right after May's SP is released. Vlado is saying that it includes GPU speed improvements so we'd need to start a new graph anyway to keep the statistics consistent.

-Alan

[Alan Iglesias]

I might just create one. After all we don't need anything special, just something that takes long to render.

And timing-wise, it might be a good idea to implement the new file right after May's SP is released. Vlado is saying that it includes GPU speed improvements so we'd need to start a new graph anyway to keep the statistics consistent.

-Alan

[3LP]

I think we should get another benchmark scene to avoid confusion.
We could more than likely mix up results and it would make it more confusing than anything.

I might just create one. After all we don't need anything special, just something that takes long to render.

Would we go for something that's like 5 times slower, 10 times? More?

We should be prepared for big configs like quad titan x etc, so it last longer than 6 months.

Stan

[Alan Iglesias]

I think you'll see a flattening off with the scores on this test scene soon, as the proportion of the time spent preparing the raycaster, transferring the data to the cards etc becomes the restricting factor. A more complex test scene probably needs to be constructed, or a higher resolution setting of the current one.

We could stipulate cranking up Max paths/pixel to 1000, 1500, or something like that. It also might be a good idea to have folks note when they are overclocking their cards.

-Alan

[3LP]

I was thinking exactly the same thing after posting the previous message.

[MattClark]

I think you'll see a flattening off with the scores on this test scene soon, as the proportion of the time spent preparing the raycaster, transferring the data to the cards etc becomes the restricting factor. A more complex test scene probably needs to be constructed, or a higher resolution setting of the current one.

[3LP]

That would be a great idea.

I also would love to see a quad titan X setup,
This would finally reach the bar of being below 30sec (in theory).
GPU rendering has bright day coming...

[matanov]

I will see what can be done to see the chart somewhere like a sticky post or something like that ..

[3LP]

Well done!
Let's just post the image again so we don't need to scroll up too much

[matanov]

https://docs.google.com/spreadsheets...A0k/edit#gid=0

I've also added one exotic combination - 1 TITAN X + 2xQuadro M6000.

[Dariusz Makowski (Dadal)]

Any titan X tests yet?

Can any1 update 1/2nd post with charts?

[3LP]

This seems to be awesome news!

Stan

[savage309]

V-Ray RT GPU CUDA might render a bit slower than 2.4, but it produces different result too (a result that is closer to the production than before), so they can't be directly compared.
We do try to keep the GPU speed constant as we add more and more features (for the next SP for example, we plan to add more stuff like displacement, subdivision, tex baking, more procedurals (composite, remap, multiply/divide, texbeizer(labeled as output modifier in max) and more), GGX/GTR BRDF, round edges, anisotropy, UDIM textures, probably paging and more and at the same time to improve the performance).

[3LP]

Hi mate,
This is my post you're referring to, and the times are legit.
That card was literally incredible.
The main reason was that that card is one of the highest OC card straight in factory and on top of that I was able to manually OC it even more 15%.

With that card I rendered this animations :
http://forums.chaosgroup.com/showthr...-Archviz-BF-LC

I found this card so incredible that I wanted to purchase more of them but it has been discontinued since as the GTX9XX came out.
So I went for 3 GTX 970, but they are equal to 2 of those GTX780Ti.

Stan