I have released a new free, multi-core benchmarking software.
The software measures your system's actual efficiency increase when a cpu- and memory-intensive task is taken from old single-threaded design and made to work with thread counts from 2 to 8. If two threads takes half as much time as one thread, that would be an efficiency of 2.00x.
In addition to displaying information on the screen, the software can write out a tiny file containing your CPU's ID String (ie "Intel Celeron"), your clock speed, your average core score, and your speed score. This information can then be uploaded to the MCB website to build a publicly viewable database containing CPU strings, maximum clocks, average clocks, average core score and average speed score.
In the preceding example, I used an Athlon 64 X2 -- dual core, so I expected to get a final average core score of 2.00x and indeed it was 2.17x.
How can it be higher than 2.00x? Well, as a single-threaded task, L2 cache misses hold up processing, everything is fairly serialized. Break that task into two pieces (or more) and now you have new possibilities. One thread's processing will hide or mask the L2 cache miss latency of other threads in the same group.
I have seen Celeron M, a single core processor, achieve consistent high 2.xx to low 3.xx scores on some tests.
Please do try the software, and upload the MultCore.BIN file it produces.
For those of you who might be paranoid, the contents of the file is:
48 characters CPU ID String
4 bytes mcb version (4)
4 bytes clockspeed (240 = 2.40 GHz)
4 bytes core score (217 = 2.17x)
4 bytes speed score (0 to 36000)
4 bytes validation value (used to verify that the information is not fake)
60 bytes reserved future expansion (zero)