General Discussion Undecided where to post - do it here. |
Reply to Thread New Thread |
![]() |
#5 |
|
Asher, this prof's jobs will likely be run by HIS grad students. Very few jobs are actually run by profs. 90+% are grad students. Also, I'm not sure how "nice" factors into it (if it's the Unix type of "nice" which assigns system resources). The cluster assigns the next job in line for which there are adequate resources. So if there's 300 processors free, one guy submits job A for 400 processors and then the next guy submits job B for 200 processors job B gets run immediately while job A goes on the queue. It's not usually an issue. Even the big jobs get run within a day or two.
|
![]() |
![]() |
#6 |
|
|
![]() |
![]() |
#8 |
|
|
![]() |
![]() |
#10 |
|
|
![]() |
![]() |
#14 |
|
|
![]() |
![]() |
#16 |
|
|
![]() |
![]() |
#17 |
|
gcc is a 1984 Toyota Corolla to Intel's 2009 McLaren F1.
Intel's compilers are far, far faster. They'll utilize SIMD hardware far, far more often in far, far better ways. You can write some benchmarks yourself, or google for them. http://www.luxrender.net/forum/viewtopic.php?f=21&t=603 On average, Intel's compiler is about 10-20% faster than gcc on the exact same code, sometimes much higher (depending on the use of vectors). 10-20% is massive for HPC. If your compute nodes are x86, use Intel's compiler. If they're PPC, use IBM's compiler. They're both in another league from GCC in terms of performance. GCC is a utilitarian/toy compiler. The kind of **** you compile crappy desktop apps with. |
![]() |
![]() |
#20 |
|
|
![]() |
Reply to Thread New Thread |
Currently Active Users Viewing This Thread: 7 (0 members and 7 guests) | |
|