.ext Vs .tx

[stephane.bonnot]

End of post and questions

So, to come to a conclusion ( sorry for the very long post), I would like to know if there is a reason for this difference between exr and tx behaviour (both are 16bit, tiled,mipmaped, compressed with standard recommended options).
Are there any drawbacks for the tx files vs exr ones that we might hit for real render scenes ( these were only tests scenes without gi or glossy stuff).
Is there any IT stuff that I can optimize to decrease render time with exr files ( higher bandwidth for server ? higher possible io count for server ? add a second gigEthernet to render machines ? etc).


Best Regards

Stephane.

[BenDel]

just some details to complete the post from Stephane Bonnot:

The Maya scene test:
- 152 Maps.
- the resolution is varying between 2K and 6K
- around 2.5 Go in total on disk: .exr with zip default compression, .tx with default compression (no access to the option with maketx.exe from oiio)
- 48 maps are 16b
- 104 are 8b
- only selfIllumination and opacity in the scene... no lights... no env... no GI....



so.. the two tests, comparing .tx and .exr are done with:

- one with: 104 .tx 8b, and 48 .tx 16b
- the other with: 104 .tx 8b, and 48 .exr 16b

both are mipmap filtered

[Dariusz Makowski (Dadal)]

Ok now that is very interesting !

I gotta read in to that tx. format. Is there any tool to convert my textures to it or something?

[vlado]

So, to come to a conclusion ( sorry for the very long post), I would like to know if there is a reason for this difference between exr and tx behaviour (both are 16bit, tiled,mipmaped, compressed with standard recommended options).

I'm not sure why exactly this happens. We handle the two formats identically once they get to V-Ray (though they take up different amounts of RAM depending on whether they are 8-bit or 16-bit). Obviously the libraries for loading the two formats are completely different (we use the opensource libtiff for .tx files, and ILM's OpenEXR library for .exr files); the physical data layout is probably also different.

Are there any drawbacks for the tx files vs exr ones that we might hit for real render scenes ( these were only tests scenes without gi or glossy stuff).

I don't think so.

Is there any IT stuff that I can optimize to decrease render time with exr files ( higher bandwidth for server ? higher possible io count for server ? add a second gigEthernet to render machines ? etc).

I don't have an answer to that either; we could try to do a similar set up and try to profile the whole thing, but it will take a while.

Best regards,
Vlado

[stephane.bonnot]

Ok, different results make sense since libtiff and openexr lib might not handle disk io the same way(readahead, cache, etc), and the physical layout sounds like another sensible reason. By the way, I would have bet on openexr for the best performance before these tests ...

Thanks a lot for the detailed tech info, we'll stick to tx for this project (very good performance and no issues with this format yet) and we will post here if we find some valuable tech info.

For other members reading this post, note that this specific issue is bound to file IOs, we hit issues with an extreme scene (more textures than RAM) and when our file server serving textures is under heavy load (I don't know yet if network bandwidth or disk IO count limitations or both are causing the issue), when run locally ( all texture files on the workstation with fast raid) the tests don't behave the same way.

Best regards
Stephane