64-bit and Dual-CPU questions

[grassman]

I apologize if these questions are a little rudimentary, but I basically have no experience with anything other than 32-bit scalar systems (although those I know pretty well). Any information will be greatly apprecated!

I am currently running VASP on 32-bit Intel Pentium 4 computers with 2GB RAM (and in one case with 4GB). We are considering starting some calculations that will require more than 4GB of memory. Unfortunately, the very best we can do right now (with Red Hat Enterprise WS 3) are calculations of ~3GB... Anything larger is not allowed due to the 32-bit architecture. We know that we can solve this problem by parallelization, however the cost of moving in that direction is beyond what we can currently afford (at least as far as clusters or large SMP machines). Another option seems to be to move to 64-bit processors (AMD Opteron or Intel Xeon with EM64T, for example). Our questions are as follows (and hopefully you can answer them):

1) Is there anything special that needs to be done in the compilation of VASP that would allow it to be run in a 64-bit operating system, such as Red Hat Enterprise for AMD64, or is it pretty straightforward (as in the 32-bit compilation, but just with a 64-bit compiler)?

2) Are there still limits to the amount of memory that can be accessed at one time for a single calculation? That is, with a single Opteron computer with 6GB of RAM and 64-bit Linux, should we be able to run a calculation that requires more than 4GB of RAM? Or, instead, will we still need to parallalize VASP on, for instance, a dual-processor machine?

3) Do anyone recommend any particular Linux versions for performing the above-mentioned calculations? We're leaning toward Red Hat Enterprise (WS 3 or 4), since we've been using Red Hat for some time now (but change can sometimes be a good thing).

4) And also, if we were to go dual-CPU, I'm guessing that VASP does indeed need to be compiled parallel in order to be able to utilize both processors, correct? (Sorry if this seems like a silly question, but since I've never used anything parallel, I just don't really know for sure.)

Thanks again...

[peterklaver]

Hi grassman,

I just joined this forum, so you may have all answers already. Just in case it is not so:
- you can run the parallel version on a single-cpu box. I sometimes run a 3-way split calculation on my dual-G5 PowerMac if I need the full memory that that machine has (and MacOS 10.3 is still partly 32-bit). So you don't need multiple machines to acces a lot of memory, a single machine should just have enough memory slots.

For building a G5-cluster for VASP I put easy instructions on
http://titus.phy.qub.ac.uk/group/Peter/

kind greets,
Peter

[grassman]

Thanks for the response. I ended up buying a single-cpu 64bit Xeon system w/6GB of RAM (I already have experience with the Intel systems and compiler software, didn't want to have to purchase Portland Group compiler to use the Opterons efficiently). With Red Hat Enterprise 4.0 (EM64T version) and 64bit GOTO libraries, VASP runs quite well (and quickly) and I can access all of that RAM for a single calc, which is what I was wanting (and I can upgrade to dual cpu later on if/when I want). Thanks again for your response, though.