Tuesday, July 24, 2012

Theoretical Solutions

Assuming that the maximum price we are willing to pay is fixed, the price of the CPU as a percentage of the entire system would be a good candidate for a criterium, especially since that would give us a nice translation to compare against other optimization parameters such as amount of memory and CPU to memory bandwidth. However, this kind of information doesn't seem to be readily available on the Web, and it may be particularly hard to pry it from a PC vendor.


FLOPS characteristics on the other hand are indeed available, but although they may be good for comparing machines used for actual computing, they aren't very useful in benchmarking desktop machines, because they favour CPUs with good floating point performance, which is of lesser importance in office work, especially with the advent of powerful GPUs to handle eye candy. Benchmarking is better, but to know which benchmarks to pick, one must study the characteristics of the benchmarks, and have a good idea of what software is to be run on the machines.


Based on a short investigation, I come up with two candidates to obtain benchmarking data from: the infamous Tom's Hardware, with a choice of benchmarks to pick from, and PassMark software for a single benchmark that incorporates all use a desktop PC is likely to be put to. My advice would be to specify minimum requirements for any number of these benchmarks for a CPU to be considered.

Remains the question of what these minimum requirements are to be. In order to answer this question, we need to know how fast the graphs of CPU computation prowess slide by in time. If we had a list of release dates of the processors shown in the charts, we could just slide a 6 month window across the charts and we would have a good approximation of what would be reasonable to ask. However, I was unable to find such a list, and release dates of individual processors seem harder to come by on the Web than expected.

Source

0 comments:

Post a Comment