ADF on Mac Pro Fires on All Four Cylinders

Last week Scientific Computing and Modelling (SCM) made available the first release of the ADF Quantum Chemistry package for Intel Macs. When I am not moonlighting for MacResearch, I am the developer responsible for the Mac ports of ADF, and last week's release represents for me the culmination of several months of development activity, both on the number crunching back-end, and the ADFLaunch application (which is exclusive to the Mac).

The Intel Mac release of ADF represents a new milestone in ease of use. Although ADF has been capable of running in parallel for as long as anyone can remember, until now this has involved installing extra software (eg. PVM) and careful configuration of one's run time environment. The Intel Mac release of ADF2006.01 has been linked with OpenMPI, and configures itself automatically at run time, so when you run it on a 4 processor Mac Pro, it will run on all 4 cores in parallel. No need to install additional libraries, or configure your environment — like all good Mac software, it just works.

To celebrate the release of ADF for Intel Macs, and the introduction of a new Benchmarks page on MacResearch, I did some benchmarking on a 3.0Ghz Mac Pro. "So how fast is that new Mac Pro?", I hear you ask. In short, it is the fastest machine we have ever benchmarked ADF on, core-for-core. For full results, check out the benchmark table.

Comments

Slight mistake in report

I think you mean memory in GB, not MB.

I have a hard time believing you managed to get a Mac Pro with 4MB. :-)

Great news on the automatic parallel configuration -- as someone who uses a range of quantum chemistry packages, that's definitely not the norm.

And yes, it's nice to see some benchmarks of the Mac Pros against some Opterons.

Opteron

The detailed report states that the Linux machine is a dual-core Xeon 3Ghz.

Good point. It is GB

Good find. Thanks. I'll fix this.

Drew

---------------------------
Drew McCormack
http://www.macanics.net
http://www.macresearch.org

Indeed

Thanks. I'll fix this too.
It should be Opteron.
Note also that I forgot the optimization (-O3) flag in the compiler options for Opteron. In case anyone was wondering, the results are with the -O3 flag included.

Drew

---------------------------
Drew McCormack
http://www.macanics.net
http://www.macresearch.org