I just tested my project in HDRP 10.2 and seems to be exponentially slower than the 7.x range
Is this something to be addressed ?
In the exact same project i get 15ms for my scene in 7.4.3 and 25ms for 10.2.
Also 3.8ms in render VS 11.6ms in 10.2. This is also a true difference as i see clear shuttering in 10.2, so is not just some counting error in the ms in the console.
All at same resolution and windows machine.
Is this some bug or is still in early beta and all will be fixed ? Otherwise HDRP is obviously not a game engine anymore, as is beyond extremely slow. It was already much much slower than standard with about same effects in 7.4.3, but 10.2 seems unusable.
Is 10.2 an official release or still in early beta and performance will be taken to 7.4.3 levels later ?
I did few quick tests with Ryzen 3900x + RTX 2070s (I’m mentioning the HW as you could get wildly different results on weaker gear). Following measurements are from release builds:
Old HD Template:
2019.4 + HDRP 7.5.1: 5.0 ms
2020.2 + HDRP 10.2.2: 5.4 ms
Viking Village ported to HDRP:
2019.4 + HDRP 7.5.1: 5.8 ms
2020.2 + HDRP 10.2.2: 5.2 ms
Here old HD Template became 0.4ms slower in newer HDRP but Viking Village with way more geometry was 0.6ms faster. I’ve also seen even bigger perf gains on 10.x on other project where large terrain was rendered so it’s I don’t think it’s fair to straight up claim 10.x is all around slower (as it’s not).
To get better idea where the differences came from, I also did development builds on the HD Template scene and attached profiler but it’s for some reason most of the time (~3.5 ms) is actually spent on waiting for gpu:
On 2019.4 it’s stalled mainly by Gfx.WaitForPresentOnFfxThread → Semaphore.WaitForSignal and on 2020.2 it’s stalled by GfxDeviceD2D11.WaitForLastPresent → DXGI.WaitOnSwapChain. I did all tests with VSync disabled so no idea what is happening here.