Originally Posted by Electroneng
I have always found the passmark benchmarks to be inconsistent. To be consistent, one needs to run the benchmark multiple times and take the average of the reading. 3 scores on one processor vs 1000 of the other is not consistent.
The I7 2600K scores 9512 on a a couple of runs while the I7 2600 scores 9072. These are the same processors with the only listed difference being the unlocked multiplier. Explain the difference?
Using your calculations on these inconsistent benchmarks:
2600k- 9512/3.4 = ~2797
2600 - 9072/3.4 = ~2668
950- 6335/3.07 = ~2063
980x - 10460/3.33 = 3141
The 2600k (3 tests) has a 35.5% advantage over the 950 (100's of tests)
The 2600 (2 tests) has a 29.3% advantage over the 950 (100's of tests)
The 980x (100's of tests) has a 12.2% advantage over the 2600k
The 980x has a 17.7% advantage over the 2600
Once the averages of 100's of tests are listed for the 2600 series, this data may become usable.
Cinebench provides a much better picture of CPU performance.
These are the posted runs:
I7-930/950 at 4.2ghz = 7.21
I7-2600K at 4.2ghz = 8.15
I7-980X at 4.2Ghz = 10.83
I score the same at 4.2ghz on my 980x
I7-2600K provides 13% more CPU power then the 950
I7-980X provides 32.8% more CPU power then the 2600K
No where in any of these calculations have I seen a 6% differential between any of these CPU's
Yes, there is more statistical significance in larger sample sizes with normal data sets... but this isn't a normal data set: the facts are that CPUs are so homogeneous as to render large sample sizes redundant so I disagree with a blanket statement that the Passmark is invalid / inconsistent due to small samples.
I do agree with you however that there is a 2600 vs K discrepancy and suggest that there may be several reasons such as ES sample inclusion (probable) or possibly a difference in SKUs which we do not know (unlikely but possible). It will be interesting to see the difference in these scores a couple of weeks from now.
Additionally, the Cinebench CPU performance test is by definition graphics with CPU rendering based and not base CPU computational performance, so I strongly disagree that its a better overall benchmark for CPUs as a discrete component. In fact, Maxon themselves represent the test as a net system
benchmark for industries / applications such as animation, film, television, CAD, etc. and not as a CPU comparison product. Various Cinebench 11.5 results sites show CPU scores with an inconsistent correlation to clock and stronger correlation to GPU thus proving this point. So in this context it's a poor tool.
The 6% difference was my guess by looking at the chart between base scores for the 2600K and the most affordable gulftown, the 970... the difference is actually closer at 5.1%... I was simply making the point that to the user, the out of box stock clock SB product has a nearly equivalent performance level at a large savings on two less cores with higher OC potential....
In reality nobody is going to clock them alike so the 4.2 argument is moot.. the real issue will be as it always is... the best bang for the buck.
A- The 2600K has a significantly larger OC overhead which will likely get it near or at parity with a overclocked Gulftown for a fraction of the price.
B- It kills the 4 core I7's
btw... this is all in good fun... healthy debate / devils advocate stuff.. so no offense intended...Edited by anuoldman - 12/29/10 at 12:34pm