Are there any tips and tricks to reduce memory usage? I am interested in simulating structures that are on the order of ~10λ. However, the number of orders required to reach convergence is quite large, and the GPU memory required quickly exceeds that of any commercially available GPU.
Below are plots for simulating a ~0.5λ x 6λ device with 1 layer.
-
GPU memory as a function of "order factor" where my order vector is [10, (order factor)*10].
-
Corresponding transmission as a function of order factor. The transmission doesn't converge within the range of tested order factor.
-
GPU memory as a function of resolution. I'm surprised to see that this barely changes.
Are there any tips and tricks to reduce memory usage? I am interested in simulating structures that are on the order of ~10λ. However, the number of orders required to reach convergence is quite large, and the GPU memory required quickly exceeds that of any commercially available GPU.
Below are plots for simulating a ~0.5λ x 6λ device with 1 layer.
GPU memory as a function of "order factor" where my order vector is [10, (order factor)*10].
Corresponding transmission as a function of order factor. The transmission doesn't converge within the range of tested order factor.
GPU memory as a function of resolution. I'm surprised to see that this barely changes.
