Ab initio calculations of molecules are one of the most resource-consuming calculations that one can imagine. When considering the size of typical molecules that modern organic and bioorganic chemists are trying to synthesize or characterize, and the lessons we learned about sub-par accuracy of semiempirical methods, it is easy to dream up calculations that cannot be performed with given hardware and time limitations. A good computational chemists is the one who knows what kind of method and basis set give the desired accuracy for a given problem using available computational resources. In many cases, calculations on the system that experimental chemists are interested in are impractical, and then a reasonable molecular model that captures the essence of the process should be studied instead. For example, if one is interested in a reaction with of 4-octylpyridine with dodecyl chloride using toluene as a solvent, maybe modeling this as a reaction between 4-methylpyridine and ethyl chloride in the homogeneous dielectric can reveal the essential energetic features of this reaction?
The speed of calculation is a very strong function of the number of basis sets used. More atoms means more basis sets; a better basis means more basis functions. Electron correlation methods are more expensive than Hartree-Fock calculations. MP2 calculation takes about twice as long as single point SCF, MP3 may be several times longer, MP4 could easily be orders of magnitude more expensive. Any calculation requires some amout of memory (RAM) which is requested in megaword units. 1 Megaword is about 8 MB, so requesting 200 MW means that 2 GB of RAM should be available. On 32 bit computers (ccl workstations), this is pretty much the limit. On 64 bit computers, one can ask more memory (kalev has 6 GB installed). If you can fit two-electron integrals into memory, the calculation will be very fast. If you cannot, it will recompute integrals, or store them on disk, both of which are slow options. However, if your calculation is small, and you ask too much memory (e.g. Ne in cc-pVDZ with %Mem=400MW), the calculation takes more time than it needs because Gaussian spends time freeing up a 4 GB segment even though it only needs 4 MB!
The speed of cortrelated calculations depends stongly on the amount of available hard disk space and disk speed. The shared disk on ccl workstations that you have been using is the very slow, and halts to grind if multiple users are running ab initio calculations on it. Gaussian is configured to write its large files to a scratch disk; PC GAMESS is configured to write them to the current directory. Thus, use the fast local disk /local/chem226/PERM when running PC GAMESS jobs, or log into the workstation. The workstation has a pair of dedicated disks in RAID-0 configuration offering 1 TB of temporary storage, so specifying a large MaxDisk (e.g. MaxDisk=1000MW) is OK and recommended for correlated calculations.
Choice of program is another factor that determines the performance. In our environment, for typical ab initio calculations that you are doing later, PC GAMESS / Firefly offers better performance than many competing programs. Please see its tutorial for details.