Don't compare what you get per result + or - versus claim, but what is granted per hour over a longer series. Then, they're not the same CPU and everything in/on the box with that (duh), thus compare the benchmarks. The integers part (Dhrystone) is the one mostly doing the faul play in the computation for claims. Whilst a client is dumb and just claims credit for run time per benchmark, the servers also look at actual throughput for the zero redundancy results.

Anyway, the horse is in coma, 500 or more threads existing on these forums discussing merit, exactness, validity, methodology and more ... fresh minds see it differently than I do.

There's an invisible multipurpose benchmark at the front of each task for HFCC, FAAH and DDDT that is compared with the server reference mean, so it's known how well each individual client does up front, irrespective of the client benchmark itself.

The claim works as (Dhrystone+Whetstone) / 480 * CPU time.

The 480 divisor varies slightly. Don't know why.

Have fun, with crunching.