There are lots of Geekbench threads with some people boasting about how high there scores are, what do they really mean?
....
I'm just trying to understand that the real world performance is relative to these scores. Obviously upgrading CPU's in a machine will increase the performance, but how much?
Thoughts?
It all depends on what you mainly use your system to do and whether you can use the bench results to perfect your system.
Except when tuning a system - something less important the more a system is closed, Geekbench scores mean less to me than Cinebench 11.5 scores, Photoshop test scores and Blender scores. Why? There is no Geekbench application, other than the benching software itself [but see below]. Because these other benches are measures of performance of applications that I use and I use Cinema 4d the most, Cinebench 11.5 is the most important bench to me on a day to day basis. If I used my system(s) mainly for web surfing, managing e-mail, creating word documents, etc. none of these benches would mean much. SSDs and adequate ram would give a greater overall feel of high performance, given how fast entry level systems are today.
However, there is a ratio between Geekbench scores and Cinebench 11.5 scores from which one can be used to roughly predict the other. Moreover, Geekbench gives the user more task data to judge relative system performance at certain tasks, such as integer, floating point, memory and stream performance (any of which, however, can skew a test in a way that may be completely meaningless for most users). The integer, floating point, memory and stream performance tested by Geekbench 2 can be further broken down into elements that may or may not be useful to a particular user, e.g., the integer test consists, in part, of text and image compression routines, and the floating point test consists, in part, of blur and sharpening routines, either of which may be important information to one who is comparing a system because she/he compresses a lot of video or uses the blur and sharpening routines in Photoshop a lot. But there is another test geared to some of the most hoggish tasks in Photoshop. A Cinebench 11.5 score, for example, just reveals a single measure of performance of certain Ogl video routines and the speed at which a particular frame is rendered using one core or multiple cores, but I render thousands and thousands of frames so this gives me useful information about whether my system is tuned just right. There is also a relationship between how fast one core can render a single frame and all of your cores can render it, showing a number revealing how linear your core performance is additively and the speed up from hyperthreading. Thus, these benches do give some useful information to those who can put it to use. Geekbench 2 data gives me more particularized information to finely tune my systems than does Cinebench 11.5, but in the end when I successfully tune a system to the max using Geekbench, my Cinebench renders are faster also. So in sum, what is most important is whether any of the information yielded by the bench is relevant to what you use your system for and whether you can and do use this information and the features that they measure often enough to make them important to you.
My most revealing bench was one that I created when I had used Geekbench 2 and Cinebench 11.5 to get my WolfPack1 (aka FrankMacWinTosh and FrankHacWinTosh) as finely tuned as I though was possible. I ran (all at the same time - low scores to prove it) (1) CPUTest to test for stability, (2) Maya [rendering a 90 frame HD animation that it completed at a rate of 3 seconds per frame], (3) Cinebench 11.5, (4) Geekbench 2 and (5) mCoreTest64. Note that WoffPack1 was underclocked, but turbo biased [13,13,13,13,14,14 bins per each w5680] to run at 2.2 GHz base at that time. This more greatly simulates my real workflows under crunch time. See pic below for results. So if you're curious about how your system stands up to your crunch time whatever that may be, I advocate coming up with those things you do and placing as much of a load on your system as you would under that crunch and see how it performs, then tune it some more if you can.
