Matt Pharr

With the latest version of nsight systems (catching up to my driver version, which it was complaining about being too recent for it), I am no longer seeing the GPU...

It looks like your Linux 3080 numbers are generally 10-15% slower than my Linux 3090 numbers, so that's good as far as that being roughly the difference I'd expect. So...

Sure, that'd be interesting to take a look at.

> Have you eventually described the mapping from v3 Disney to v4 new material models? Not yet. Patches welcome. :-)

The "good" news is that the bug is in the display-server code; the final EXR that's saved to disk doesn't have this issue..

Turns out that this boils down to the GPU path not using ray differentials for texture filtering. The parameterization on that model is funky, leading to a very wide filter....

That is very puzzling. Agreed with @pierremoreau. I'm also surprised that it happens in just those two files; I would imagine that default constructor is used in many more files,...

...this might be a g++ bug. [Here](https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68070) is an old bug report that fits the symptoms. It's not marked fixed yet, but it's pretty old... Could you try changing the...

So as far as a fix (or "workaround", I suppose), pbrt's CMakeLists.txt file should add `-fkeep-inline-functions` for gcc versions before v8? (Admittedly we don't know when this was fixed, but...

I've just (finally) pushed a fix that should in theory add that flag with versions of gcc before 8.0. If you have a chance to try it out and see...