pardiso stops with error message -2 in phase 33 when increasing the number of rhs from 600 to 700. 600 rhs runs fine with a memory use of 146.2GB ("RES") reported by "top". The system has 256GB of RAM, so there is still +40% left. What is the issue when increasing #rhs by 100??
This may happen because of Pardiso consumes too many memory in the case of rhs. How many threads do you use? what is the problem size? Please check what iparm returns in the case of 500 and 600 threads for example.
first it is 700 to 800 rhs, not 500 to 600 ............. my fault.
iparm with 700 rhs is:
iparm with 800 rhs is:
so iparm for 700 is 477,759,815, where iparm for 800 is 1,727,119
That all doesn't make sense to me especially since the manual states that iparm is calulated during phase 1.
Further if that is the memory usage in kbyte then 700 rhs was at 477,759,815*1000/10^9 GByte which is 477.75 GB. But the computer has only 256GB and I am 100% certain that it never went into swap.
The factor contains 126,792,401 non-zero elements and has a row/column dimension of 2,421,403. I need to solve for ~50,000 rhs, which I supply in chunks of up to 700 rhs. Because from what we discussed here
I want to keep the number of RHSs as many as possible.
CPU is Intel(R) Xeon(R) CPU E5-2697 v4 @ 2.30GHz, so effectively 36 cores and threads.
Note that I don't store a rhs matrix of 2.4Miox50,000. The rhs are generated by a pre-mult step with a sparse rectangular matrix before they go to the solver.
Can you specify the Intel MKL version which you are using? I believe we have improved the memory consumption for the multi-threaded execution of the Intel MKL Pardiso in MKL 2019u2.
some small correction - actually the fix of the problem will be available into the next ( nearest) version MKL 2019 update 3. the current ( latest) available update 2 doesn't contains such kind of fixes...