HPF and cray T3E

July 31, 1998, 18:37

I am currently working on a DNS simulation. I have recently begun using HPF on a cray T3E. Has anyone had any experience using high performance fortran on a cray T3E?

August 4, 1998, 05:41

We are now porting some CRAFT codes, developed on the T3D (http://sawww.epfl.ch/SIC/SA/publicat.../7-95-37a.html) to a T3E, using HPF-CRAFT. The codes are essentially devoted to time-dependent and DNS of buoyant flows. However, up to now, our results in terms of speed-up, scalability etc., are below our *reasonable expectations*. This might be due to the PGHPF compiler we have on our machine, which is still under development. For this reason, we are also working on some *hybrid* solutions, i.e. HPF+shmem (not portable) and/or HPF+MPI (portable) for the Poisson solver, which constitutes the most time-consuming part of the code. Is someone out there with similar experiences ?

July 31, 1998, 18:37	HPF and cray T3E	#1
Frank Muldoon Guest Posts: n/a	I am currently working on a DNS simulation. I have recently begun using HPF on a cray T3E. Has anyone had any experience using high performance fortran on a cray T3E?

August 4, 1998, 05:41	Re: HPF and cray T3E	#2
Enrico Nobile Guest Posts: n/a	We are now porting some CRAFT codes, developed on the T3D (http://sawww.epfl.ch/SIC/SA/publicat.../7-95-37a.html) to a T3E, using HPF-CRAFT. The codes are essentially devoted to time-dependent and DNS of buoyant flows. However, up to now, our results in terms of speed-up, scalability etc., are below our reasonable expectations. This might be due to the PGHPF compiler we have on our machine, which is still under development. For this reason, we are also working on some hybrid solutions, i.e. HPF+shmem (not portable) and/or HPF+MPI (portable) for the Poisson solver, which constitutes the most time-consuming part of the code. Is someone out there with similar experiences ?