Denis Demidov

Hello, > Is there a more effective approach to expedite the convergence of the linear system? I am uncertain why the convergence for such a matrix is slow and how...

It looks like the problem is just too difficult to solve with a simple iterative approach. Intel Pardiso is a direct solver, so it does not have this problem. Direct...

I am not familiar with Kratos internals well enough to understand what is the problem here, sorry.

That is a known issue/normal behavior. The first run is ["warm up"](https://www.google.com/search?q=nvidia+cuda+warmup), when the driver does things like kernel compilation for your specific device, caching etc. The numbers in the...

`profile.toc()` [returns](https://github.com/ddemidov/amgcl/blob/57a8140c8e9cb71d5f01824aa0e934bf71f34e9d/amgcl/profiler.hpp#L101) time since the initiating `tic`, so you could use that. There is no method to get the total time across all iterations, so you could either accumulate it...

Can you save the matrix and the RHS vector on the step that results in nans? You can use `` for that: ```cpp amgcl::io::mm_write("A.mtx", A); amgcl::io::mm_write("b.mtx", h_F.data(), h_F.size()); ```

Hello, here is an example: https://gist.github.com/ddemidov/3292c6fea940c785dc8b43461846ae45

Hi, 42ee9da2057153b1aef734e8e462c846a33aa805 should fix this.

Hi, can you provide a minimal example that reproduces the issue?

The difference between using `shared_ptr` and not is just a matter of convenience. Look closely into what's happening in the example on your first link: the solver is constructed inside...