Re: amber-developers: UNC JAC benchmark?

From: David A. Case <>
Date: Tue, 27 Apr 2004 15:34:50 -0700

On Tue, Apr 27, 2004, Carlos Simmerling wrote:

> is that JAC timing of 202ps/day for single 3.2ghz Xeon at UNC/duke
> correct? (on the benchmarks page)
> if yes, then why is the "fix" number for UNC/duke much closer to the
> data?

Just to make sure you are looking at the correct ratios:

For clock speeds, unc/bohr = 3.2/2.4 = 1.33

for JAC (2 cpu) unc/bohr = 330/165 = 2.0

for fix (2 cpu) unc/bohr = 120/78 = 1.54

My guess (assuming no typos....) is that the new 3.2 GHz PIV have a much
bigger L2 cache, which is why both ratios are better than the ratio of
clock speeds per se. For the smaller JAC problem, the "extra" cache on
newer machine is a real win, but for the bigger system, both processors
blow cache more often, and the "win" for the unc system is not so great.

Note also that the relative speeds of altix vs. unc are reasonably similar
for both the jac and fix benchmarks.

And, according to Bob, going to ifc8 will make the new 3.2 Xeon chips look
even better than the ifc7 numbers shown on the benchmark page.

[Now, you can wait for the correct answers from others on this list....]


David A. Case                     |  e-mail:
Dept. of Molecular Biology, TPC15 |  fax:          +1-858-784-8896
The Scripps Research Institute    |  phone:        +1-858-784-9768
10550 N. Torrey Pines Rd.         |  home page:                   
La Jolla CA 92037  USA            |
Received on Wed Apr 05 2006 - 23:50:03 PDT
Custom Search