Pooling GPU memories with ITensorMPS

ITensor Julia Questions
1

Hi all,

Does anyone have any experience with using multiple GPUs that share their memory for MPS computations?
Unfortunately, my TDVP2 simulation throws “out of GPU memory” due to too large bondim..

Would be thankful for any leads.

P.S.
I saw this one Memory bottleneck with GPU backend for DMRG. @mtfishman- Does “write-to-disk” still operate (also for tdvp?), has it been tested?

2

I can’t remember if we’ve tested it, in principle it should work since TDVP shares the same data structures as DMRG internally.