jump to navigation

64 Bit Machines December 13, 2006

Posted by gordonwatts in computers, physics.

Almost everyone is looking towards 64 bit machines for future computing. Inner detector and calorimeter simulation can take up to 18 minutes to simulate and digitize. We need to make billions of these events to understand the data we will collect. So anything that makes the simulation go faster (cheaper!) is something that we and the LHC experiments are going to be interested in.

But, being scientists, we need to test everything. I saw a talk this morning measuring the CPU and memory requirements for running on 64 bit machines versus 32 bit machines. In absolute time it looks like we get a 30% speedup, which is great! You can build the programs that you run on a 64 bit machine in 32 or 64 bit mode — the speed up is minor: you get most of the gain by moving to the 64 bit machine and a few more percent by rebuilding in 64 bit. Unfortunately, the memory usage almost doubles. Ouch.

But even more interesting, if instead of using time as a benchmark, one normalizes to SpecInt 2000 bench marks the 64-bit machines are slower than then 32 bit machines! What this means is that if you judge what processor you are going to buy based on the benchmark, ATLAS simulation won’t speed up as much as you might like. This might also just be saying that SpecInt is the wrong benchmark to be running for HEP.



1. nt moore - December 13, 2006

While putting together a small cluster I wondered about similar 64/32 bit speed claims. One informative page seemed to be the one that the GMP library authors keep at (http://www.swox.com/gmp/32vs64.html). I ended up buying a AMD Athlon64 x2 machines, which were considerably cheaper than the intel counterparts. The main disappointment I’ve found thus far is the remarkable immaturity of 64 bit linux – I tried Suse and Scientific Linux. I’m not a guru, so its a big deal for me to compile scilab, latex, gnuplot, gfortran, etc all from source. Have you LHC guys made it past this obstacle?

2. gordonwatts - December 14, 2006

We have people who do this for a living. We are using Scientific Linux 3x and 4x (migrating to 4x now). We have gotten to the point where stuff runs, but not everything, and not everything correctly on 64 bit. But the machines people are putting together are production machines, so I doubt they care about things like latex. As the machines become more popular, I would guess that will be fixed up.

The general 64 bit problem seems to be industry wide — and is associated with this new architecture showing up (I would guess). The Windows world shows similar signs of imaturity, for example. Other than wait, I don’t think there is much that can be done.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: