Thanks for the new RTX build.
Now the The new Cycles CPU 2.79 Benchmark has good performance with 256x256 tiles.
Latest build 2.80 (tiles 256x256): 51.32
your RTX build: 39.60
Without denoiser:
Latest build 2.80 (tiles 256x256): 49.63
your RTX build: 38.15
Looks like the denoiser only takes 1.5 seconds with 256x256 tiles and 2 gpus.
The denoiser is still a problem with smaller tiles with the auto tiles option even with one GPU.
Here is my test on the BMW GPU benchmark with only one 1x rtx2070.
The default without denoiser
your RTX build (auto tiles): 45.07.
your RTX build (256x256): 45.77.
With denoiser:
your RTX build (auto tiles): 54.95.
your RTX build (256x256): 48.36.
As you can see the denoiser with small tiles is slow even with only one GPU. The 256x256 tiles to denoise only takes 2.5 seconds the auto tiles take almost 10 seconds with one gpu. I remember Lukas saying that small tiles have performance issues with the denoiser.