We're trying to render some oceans and are getting clearly visible buckets which are even more pronounced when denoising.
We're coming over from normally mostly rendering with Arnold so we're lacking a bit of experience on where to start debugging this problem.

Does anyone know what might be causing this? And perhaps a way around it without just throwing more samples at it as the rendertimes are already starting to creep up.

Currently we're rendering with 64 primary samples and 1/9 min/max secondary at 4k.
Scene is being lit by one HDRI, one distance light and three area lights.

Any help would be appreciated.

Cheers,
Alex

Oh dear, this looks like a bug - Do you see this with the recent daily (or production) build? There were a couple bucketing bugs in 20.5 which were fixed since release. If so, can you please submit a bug with reproducible scene so that we may investigate this ASAP? Thank you.

Ah, sorry my bad.
I forgot to write that we're still on H20.0.688 so maybe this has already been fixed in an update.

I tried updating to H20.5.410 but the bucket error is still there so I've now submitted a scene that hopefully helps with debugging this problem.

I got a response from support and though I'd post it here in case anyone else stumbles upon it in the future.

The issue turns out to be some bad math and numerical instabilities with blinn (and phong) BSDF.
The fix does introduce some slight different energy output and thus a bit of look change, so we're not likely to backport the fix.
To avoid the the artifacting there are a couple things you can do:
A) switch to GGX model for the glossy components (recommended). or...
B) keep the roughness of Blinn above 0.03 or so.