|
|
|
|
Joined: Oct 2003
Posts: 15
Forum Member
|
OP
Forum Member
Joined: Oct 2003
Posts: 15 |
Hi All,
I've got quite unusual benchmark results on relative performance
of Intel Xeon and AMD Athlon machines and it would be nice
if others would share their experience regarding performance of
new Intel Xeon machines on CHARMM jobs.
Here are details of my tests.
CHARMM version: c30a1
Test case: cbenchtest from the CHARMM standard tests
Operating system: Linux Red Hat
Athlon machine Linux kernel 2.4.20-20.8.smp
Xeon machine Linux kernel 2.4.20-20.9.smp
Single processor calculations took:
AMD Athlom MP-2000: ..... 36 min
Intel Xeon 2.8 GHz: ..... 33 min
This is a bit strange, is not it? One would assume that
performance should be improved with higher CPU frequency
but this does not happen in this case.
It would be nice to learn, are these numbers consistent with
experience collected in other labs?
With kind regards,
Victor
Victor Anisimov, PhD
Alex MacKerell group
School of Pharmacy
University of Maryland
|
|
|
|
|
Joined: Sep 2003
Posts: 8,660 Likes: 26
Forum Member
|
Forum Member
Joined: Sep 2003
Posts: 8,660 Likes: 26 |
The Biowulf cluster at NIH has some of the Intel 2.8 GHz nodes, and I've tested those vs. AMD p1400 nodes (both node types dual proc), with 2 different size systems (no. of atoms) using MPICH ethernet for parallel. Some results (brief MD with PM Ewald, elapsed time in seconds) are:
ca. 49K atoms
N Int. AMD
1 1103 1740
2 706 1228
4 728 1251
8 771 1217
ca. 20K atoms
N Int. AMD
1 480 681
2 263 380
4 231 369
8 216 488
The Intel nodes are faster, but not 2x the p1400. However, the AMD based machines run significantly hotter, and can cause major cooling problems when used in a cluster.
Rick Venable
computational chemist
|
|
|
|
|
Joined: Sep 2003
Posts: 175
Forum Member
|
|
Forum Member
Joined: Sep 2003
Posts: 175 |
there is a web site which has more on Intel vs AMD AMD vs Intel
|
|
|
|
|
Joined: Sep 2003
Posts: 4,883 Likes: 12
Forum Member
|
|
Forum Member
Joined: Sep 2003
Posts: 4,883 Likes: 12 |
I get 19 minutes for cbenchtest/mbcodyn.inp on a 2.8 GHz Intel Xeon machine, GNU/Linux 2.4.20-9smp, using c31a1 compiled with pgf90 (v 5.0), but I do not have an AMD system to compare with.
Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden
|
|
|
|
|
Joined: Sep 2003
Posts: 175
Forum Member
|
|
Forum Member
Joined: Sep 2003
Posts: 175 |
On the http://r.cmm.ki.si/vrana/amdvsintel.html the best result is obtained by compiling it with the gcc for P4. The results in the columns B are for the same benchmark (cbenchtest/mbcodyn.inp) and is also 19.1 minutes, but it is for 2.4GHz P4 (not Xeon).
|
|
|
|
|
|
|
