Originally Posted by
aster
How good an indicator is this single benchmark that this site uses? If say I have a processor with 4638 as the benchmark and a new one has 8089 then we're talking about am almost 75% increase in performance here?
Passmark (the benchmark used at cpubenchmark.com) is a VERY blunt instrument; it's good for brief overall comparisons, but it should be taken with a grain of salt.
Keep in mind that they are performance overall, not per core, and don't reflect differences in the Intel turbo boost feature; a 4-core model with a 4000 passmark and a 2-core model with a 4000 passmark are going to have very very different peak single-threaded performance.
Originally Posted by
aster
Now I'm trying to use it to compare servers (thinking about upgrading vs staying with an old spec server), where I take it such a benchmark would probably be more useful as it doesn't involve graphics/games/etc.?
The quality of Passmark gets worse with servers, where the data set they use is smaller, and where bigger differences in the workload are likely to be present.
If you can find a benchmark specific to your workload (e.g. SPECjbb2013 for java application servers running a transaction-processing workload) that's best. If not, SPECint_rate is usually better than passmark for a general server workload, and they're measuing similar things.
Given that a big part of my job is evaluating servers, what's your workload, and what's old server vs. the upgrade you're considering? If the server is older than 3-4 years, the other really big factor is electrical cost; machines older than the Nehalem generation (Xeon 5500) in particular have pretty bad power scaling features and are going to use a lot more electricity even at idle than a newer one with modern design.