bdobbins@gmail_com
New Member
Hi guys,
This is several questions all in one:
1) We're running CCSM3 (beta14) here - I know of the internal CCSM3.5, though. Is this something we should consider? I know from a long-ago look at things, the build system was changed considerably and might be much easier to use. I'm mostly interested in knowing what the major differences in performance are as well as how different CCSM3.0 and 3.5 are scientifically. I know that's a big question, so even if the answer is, "If you don't know, you don't want it!", that's fine. Just wanted to ask. ;-)
2) In another post on this board, Karen showed instructions for using the 10.1 Intel compilers - I'm planning on going to 11.0 and a preliminary test showed some issues, most of which I'm sure are pretty simple to work out. (And some system ones, too, such as CLK_TCK -> CLOCKS_PER_SEC). If I'm going to put in the effort to patch these, is there someone at NCAR to whom I can send the fixes? Is that even of interest to people, or is 4.0 close enough to release that additional patches to 3.0 are not terribly interesting? (Assuming Intel 11.0 runs correctly, of course.)
3) Finally, I know Meng-Pai Hung, a PhD student at OSU, achieved about 13.17 years/day on an Opteron cluster as OSC for the T42_gx1v3 runs he was doing. Does anyone have any data on Xeon (Harpertown) systems? The memory subsystem is poor, so load-balancing on-node is tough, but it's got a decent FPU. I'd also love any Nehalem timings people have. And ORNL's "Jaguar" achieved 40 years/day, but I can't find what resolution that was achieved at, anyone know?
Thanks very, very much,
- Brian
This is several questions all in one:
1) We're running CCSM3 (beta14) here - I know of the internal CCSM3.5, though. Is this something we should consider? I know from a long-ago look at things, the build system was changed considerably and might be much easier to use. I'm mostly interested in knowing what the major differences in performance are as well as how different CCSM3.0 and 3.5 are scientifically. I know that's a big question, so even if the answer is, "If you don't know, you don't want it!", that's fine. Just wanted to ask. ;-)
2) In another post on this board, Karen showed instructions for using the 10.1 Intel compilers - I'm planning on going to 11.0 and a preliminary test showed some issues, most of which I'm sure are pretty simple to work out. (And some system ones, too, such as CLK_TCK -> CLOCKS_PER_SEC). If I'm going to put in the effort to patch these, is there someone at NCAR to whom I can send the fixes? Is that even of interest to people, or is 4.0 close enough to release that additional patches to 3.0 are not terribly interesting? (Assuming Intel 11.0 runs correctly, of course.)
3) Finally, I know Meng-Pai Hung, a PhD student at OSU, achieved about 13.17 years/day on an Opteron cluster as OSC for the T42_gx1v3 runs he was doing. Does anyone have any data on Xeon (Harpertown) systems? The memory subsystem is poor, so load-balancing on-node is tough, but it's got a decent FPU. I'd also love any Nehalem timings people have. And ORNL's "Jaguar" achieved 40 years/day, but I can't find what resolution that was achieved at, anyone know?
Thanks very, very much,
- Brian