Discussion CPUz Benchmarking thread

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
With so many questionable Rocket Lake CPUz benchmarks out there I thought it might be clarifying to have an "official" benchmark thread of our own.

In order to specify the actual clock speed of the tested CPU let's use the "Average Effective Clock" that we are using in the "Handbrake" testing thread in this forum.

As always I'm open to all suggestions but here's what I'm thinking.

Download and and run "HWinfo." You can download the portable so you don't have to install anything. I run it "sensors only" so as not to clutter up the screen.

You need a version of CPUz that has the benchmark version 17.0164. This is of course important;)

Run HWinfo and note the "Average Effective Clock." You can highlight this line to it's easier to pick it out. You can also customize the "layout" in HWinfo's settings to get rid of a lot of the info you don't need.

1. Open CPUz and HWinfo so they are both visible on the screen.
2. Run "Bench" on CPUz and immediately reset the HWinfo timer to reset values. It's the stopwatch icon at the bottom of the sensors window that you should have open.
3. While CPUz is running the multithread bench note the "Average Effective Clock" value. You want to remember the last value before it starts to decline. This is your average effective clock for the multithread run of CPUz's bench. This frequency number will start to decrease as soon as CPUz switches to the single thread test because the other cores will become "unloaded."
4. Note which core loads up during the single core test.
5. Run CPUz again. This time as soon as the "Average Effective Clock" starts to decrease press the stopwatch reset and now watch the average clock of the CPU that is loaded to near 100% during the single thread bench run. Remember the highest value before it starts to decrease when the test ends. This is your single core clockspeed.

The column I have created in the table below titled "Actual MT/ST Ratio" is the multithread score extrapolated from the single core clock and then divided by the single core clock at the same speed. This is the actual ratio because what CPUz reports a value using the "boosted" single core speed divided by the lower "all core" clockspeed. This gives us an idea of how many cores the HT/SMT of our CPU is "worth" during this test. For example, my 4770k with 4 physical cores performs like 5.14 cores. So in essence the hyperthreading virtual cores provide 1.14 additional compute in terms of physical cores.

Let's see how these various cores perform when we nail down their actual average clockspeed during the benchmark testing.

I think. If I'm off-base I'm sure one of the many really well-versed people on this board will help me out.

Interesting to note my Haswell does the best with HT performance on this test. Then my Kaby Lake R, and finally Broadwell. My old Penryn has no HT and actually underperforms in multithread mode. I had to extrapolate the score for my PIII-M since it wouldn't run the later versions. I ran old/new versions on my 4770k to obtain a multiplier estimate a score for the new benchmark version.


UserCPUModel if laptop, memory specs if desktopCoresST ClockST ScoreCoreMHz/(CoresxMHz)MT ClockMT ScoreActual MT/ST RatioHT/SMT "Worth" this many cores
Hulk4770k16GB DDR2-80043890418.7Haswell0.10836912042.95.141.14
Hulk8250uMicrosoft Surface Laptop 243291395Kaby Lake R0.12028871733.95.001.00
Hulk5200uLenovo t450s22570279.8Broadwell0.1092494775.82.860.86
HulkT7200Dell Inspiron 640m2199880.5Penryn0.0401998157.31.95-0.05No HT!
HulkPentium III-MCompaq EvoN600c1200025.9Tualatin0.013No HT,No dual core,NoNuthin'!estimate score from v1.75

1609613975842.png
 
Last edited: