Benchmarking results per core.

[20nm]

In fact the i7 has 8 thread execution units. 4 cores + the new hyperthreading on each core. So do you divided by 8 or 4? Further it has turbo mode that overclocks a single thread. So after dividing by 4 (or then you need to add 10% to 20% back on for turbo mode. But only for i7s. In addition it is naive to just divide by 4 (or as going 1 core to 4 cores doesn't generally quadruple performance.

For machines that have "several quads" or several CPUs of any sort we label this. The name will be [Dual CPU] or [Quad CPU] and they all appear in the same multiple CPU chart.

I divide by 4. The 8 threads give efficiency. The HT gives you efficiency at handling multiple programs.
http://en.wikipedia.org/wiki/Multithreading_%28computer_hardware%29#Hardware_co st

Before when I was talking about dividing by 8 is because the machine has 2 cpus, with 4 cores.

Does CPU index reported by your Performance Test 7 use all 4 cores or not?

If you want to be picky you can provide 3 indexes,
1. CPU index
2. CPU index divided by the number of cores
3. Index of a single core, no multi-tasking (no ht benefits).

[David (PassMark)]

Say a i7 3.2Ghz has 7932 index and 4 cores, in theory the performance per core should be 1983

In fact the i7 has 8 thread execution units. 4 cores + the new hyperthreading on each core. So do you divided by 8 or 4? Further it has turbo mode that overclocks a single thread. So after dividing by 4 (or then you need to add 10% to 20% back on for turbo mode. But only for i7s. In addition it is naive to just divide by 4 (or as going 1 core to 4 cores doesn't generally quadruple performance.

For machines that have "several quads" or several CPUs of any sort we label this. The name will be [Dual CPU] or [Quad CPU] and they all appear in the same multiple CPU chart.

[20nm]

Guads= Quads

We are interested in the performance per CORE in a cpu. Not the aggregate performance of all the cores maxed out in a single CPU.

If you don't do this, we have sit around and calculate how many cores, and cpus is the cpu performance index comprised off.

Say a i7 3.2Ghz has 7932 index and 4 cores, in theory the performance per core should be 1983.

It would be good to have reports on the site as to how much performance a machine has on an individual core for those cases when you need to chose a machine that is the fastest to run applications which are single threaded.

We can do it by hand searching the Intel site. To get a valid result we need to have this inforrmation for all the cpus on the site. Doing the calculation by hand for me is not feasable.

[David (PassMark)]

There are very very few machines that are under clocked. You are probably confusing a low measured CPU speed (due to the CPU being partially / fully idle) with underclocking.

When running the actual tests the CPUs should detect the load and run at full speed.

and often you don't know that the machine has several Guads

Err, yes. The Guads are an eternal mystery.

You can set the tests to be single threaded from the Edit / Preferences window in PerformanceTest. But you are correct we don't report information in the charts about how well a CPU will perform if just 50% loaded, or 25% loaded, or just 12.5% loaded (in the case of 8 cores).

[20nm]

The results on CPUbenchmark.net are often misleading.

You can have a machine that is underclocked and gives you the idea that the machine is faster.

Sometimes you don't know how many cores a cpu has, and often you don't know that the machine has several Guads.

So we need the CPU interger reported/meassure on a per core basis as often applications are single threaded and we don't know what machine are going to run on best.