Sorry for the late reply, but I didn’t know the answer to this at the time!
It’s bad because writing high res G-buffers at 90Hz uses a ton of memory bandwidth, and apparently AA is a must for VR, so you also have to run a post processing AA algorithm on top of that which eats even more bandwidth.
So memory bandwidth ends up being the bottleneck, and switching to a single pass (or barely deferred like a depth pre-pass) forward renderer with MSAA makes you go faster.
Why is deferred rendering bad for VR?
Sorry for the late reply, but I didn’t know the answer to this at the time!
It’s bad because writing high res G-buffers at 90Hz uses a ton of memory bandwidth, and apparently AA is a must for VR, so you also have to run a post processing AA algorithm on top of that which eats even more bandwidth.
So memory bandwidth ends up being the bottleneck, and switching to a single pass (or barely deferred like a depth pre-pass) forward renderer with MSAA makes you go faster.