How to accelerate the rendering process?

When I trained the model with 8G-1070Ti and 24G-3090Ti, the speed increase was not significant. Compared the original-version, the taichi-version is 2-3 times slower, is it possible to speed up the taichi-version further?