unlikely cpu benchmarks ?

[W Gretsky]

Looking at the mega page, I am puzzled by a couple of seeming inconsistencies.

1. It appears that the Core i3-6100 at 3.7 GHz is faster than the Core i3-6300 at 3.8 GHz on single thread but slower on overall CPU mark when they are obviously the same design on a different clock speed? If it were due to non-cpu system differences then at least it would be consistent?

2. Equally puzzling, looking at the single thread mark, the Core i3-6320 at 3.9 GHz with 4 MB L3 cache runs 7% faster than a Core i5-6600 at 3.9 GHz with 6 MB L3 cache?

[David (PassMark)]

The i3-6100 and i3-6300 are very very close in performance. The 2% clock speed difference is nearly meaningless as other factors like the RAM, motherboard & cooling solution used can have a greater and 2% influence. In other words, the margin for error in the results is greater than the performance differences in these CPUs.

The Core i5-6600 only has a base frequency of 3.3Ghz. While the i3-6320 has a base frequency of 3.9Ghz. While the i5-6600 can turbo to 3.9Ghz, it seems this still doesn't match the i3-6320. I don't know the exact reason why. Maybe turbo lag? I don't think there is anything wrong with the benchmarks results however (in the sense that they reflect the real world situation). If you look at cache size per core then the i3 has more cache.

Of course the i5 is way better than the i3 when doing multi-threaded tasks.

[W Gretsky]

Thank you for your response.

Ad 1., my main point was that the results appear inconsistent not only because the slower processor is faster on single thread (which you addressed), but because one processor wins on single thread and the other one wins overall using both cores (which you did not address) - on the same sets of hardware - while the processor design would be identical.

[W Gretsky]

Ad. 2, with respect to the L3 cache per core, during a single thread test does the benchmarking software keep the other 3 cores busy while testing a single thread, or are 3 cores idling while one is running the single thread mark?

(and in that case wouldn't the 6MB all be available for the 1 tested core?)