PassMark CPU Bench Single-Threaded Perf Chart just seems wrong

[themanuel]

The subject chart is often recommended as a resource when deciding what processor to buy for emulation.

However, the results don't make sense to me whatsoever:

Processor                                                        Score
Intel Core i7-8565U @ 1.80GHz                         2378
Intel Core i3-8350K @ 4.00GHz                         2336

If this truly represents single threaded operation, how can the i7 have the same performance as the i3, which is running at more than twice the clock speed?

It just doesn't add up and makes me question using this chart when considering my next upgrade.  You could argue that the i7 has a larger cache, supports more instruction sets, etc., but non of that can make up for the enormous difference in clock speed.

The only possible explanation would be if the chart represents operation on a single thread but on all available cores, or if the results are normalized for clock speed.  However, I don't see that stated anywhere on the page.

Can anybody shed some light on this?

Edit:
Well, after clicking on that i7 CPU I learned that it has turbo speed up to 4.6GHz, which probably explains the discrepancy. It also has only 4 cores, just like the i3-8350K.
I guess this answers my question but feel free to weigh in if you have additional insight.

[TKSilver]

Keep in mind that the scores are averages across multiple users and in the case of laptops\tablets multiple different devices. So your millage may (will) vary depending on cooling and power delivery choices the device maker makes. This is especially true with the U series CPUs since they rely heavily on turbo boost for single threaded performance. The last 5 total scores (I wish passmark would show the individual stp scores) out of the 205 samples it used vary in between 8575 and 10013 (the total score for the CPU is 9088). For the 8350K it is even more varried with the low last 5 being 8955 and the high being 11592 (the average is 9219), but that is to be expected on a CPU that can overclock.

[themanuel]

You bring up very good points and I should have used desktop processors for my comparison, since they offer more stable baseline conditions. Regarding variability, it it to be expected although many of the processors in the list have thousands of samples, which provides a decent comparison of nominal performance.

Now, I was assuming the scores there reflect no overclocking. Allowing OC results without showing the clock speed would really muddy things up.
Do you know if that is the case?

[JonnyH]

As you mentioned, you're comparing base, not boost clocks. The base clocks are calculated to assume all cores loaded and be within the TDP (with a fair bit of fudge factor, but close enough) over a length of time (IE it can boost higher for shorter workloads). These benchmarks only loads a single core, so the power "saved" from the other cores being idle can be used to boost the one loaded core higher, and the benchmarks aren't that long, allowing it to boost higher. So the results are pretty much "Best case" for the low powered chip SKUs (like the ones ending in "U")

i7-8565U boosts to 4.6ghz
i3-8350K clocks at 4.0ghz.

So a lot closer, and if the i7 in this device is somewhat power/thermally limited, it could explain why it might "average" 4.0ghz and come up with exactly the same result.

But that doesn't mean it'll be the same at running dolphin - as the workload is over a much longer period of time, and loads more than 1 core, so the results could diverge significantly - so be wary in using this to make purchasing decisions with dolphin in mind!

[themanuel]

Yeah, I always prefer unlocked units so that I can unleash the full potential, knowing that the turbo functionality is limited as you described, but you have helped me understand that those limitations may be of a greater magnitude than I thought.

How about overclockable CPU's on that list; are the reported scores from tests at the base frequency, or at whatever clock the user had it set when running the test?

[TKSilver]

The clock speed that the CPU was at when running the test. Some of the K series and overclockable AMD CPUS have as much variability as the laptop CPUs because of differences in overclocking.

Also to add on to JonnyH's point. More core in some cases can actually be detrimental to Single Threaded heavy workloads. More cores = more transistors, more transistors in the same size package/with similar power and thermal limits can impact heat generation and available power. One reason 16+ core server\workstation CPUs are not generally recomended for gaming or emulation, even when they have higher TDP and physical size. In laptops using the iGPU also can dump extra heat In the CPU package and decrease max frequencies, though if your discrete GPU solution is overloading the heat dissipation capabilities of laptop's cooling solution then that can be just as bad (*cough macbook *cough)..

[themanuel]

(05-18-2019, 06:57 AM)TKSilver Wrote: The clock speed that the CPU was at when running the test.

I was afraid that would be the answer. If they can't segregate the results by clock speed, or at least show the average clock speed for all the samples, it really limits the usefulness of the database.