SP ratio KS vs. K/KF and raw data
But back to the sensible SP values of the cores. Here I have now plotted again the values for the P cores from the previous articles of K and KF CPUs against those of the KS and normalized them relative to the number. So here you can see only the relative distribution in percent, not the actual number per SP value per SKU. At this point, it should be emphasized once again that the 13900KS has a different scale for the SP values and these cannot be directly compared with K and KF.
While the 13900KF shows an almost evenly bell-shaped distribution, the K and KS SKUs are shifted to the left and right, respectively. This is actually obvious because, as already mentioned, the chips with a working iGPU are divided into K and KS, but the chips without a working iGPU are not. And if you pick out the best from the amount of potential K CPUs, this automatically results in a larger share of inferior CPUs among the rest. Here it also becomes clear once again that you certainly don’t get a potato CPU with a 13900KS, although no main prize in the Silicon Lottery is guaranteed either.
But what would have to be done now so that the SP values of KS, K and KF can be compared? Attention, speculoos alert! In the following, I have artificially corrected the SP values of the KS CPUs with a factor so far downwards, until the distribution over the SP values of from K and KS CPUs together coincides as well as possible with the KF CPUs. This is based on the assumption that all 3 SKUs are theoretically produced in the same way, with the same production tolerances, and thus the same proportional accumulation per SP value should result.
And the result is not 42, but -6. Now, of course, such a data alignment is not an exact science and the curves do not really coincide, but it should not be too far from reality either. Because it has also been shown in practice that the 13900KS does not clock higher in the same ratio as their P SP values are larger. In concrete terms, this would mean that a 13900KS CPU with P SP 126, for example, is about as good as a 13900KF with P SP 120. Conversely, the best 13900K CPU with P SP 123 actually tested by us would then still be of slightly higher quality than the best 13900KS CPU with P SP 127.
Though, in practice SP values are no exact science either. It’s actually not uncommon that a CPU with theoretically slightly worse SP values can achieve minimally higher clock speeds and vice versa. Again different is the behavior under extreme cooling methods, where for example at -190 °C an SP 120 might outclock an SP 125 that was faster at ambient temps with normal cooling methods. In the best case scenario SP values, cookies, CPU Force results etc. are only reference points, and in the worst case they are completely meaningless, as we saw today. At the end of the day, you can only do one thing to be sure: Real testing in the actual usecase.
Whoever has a better idea to evaluate the masses of data and somehow bring the K and KS together to the same denominator is welcome to try it for themselves. Because as of today, all of our raw binning data for Raptor Lake CPUs is publicly available at the following Google Sheets link: https://docs.google.com/spreadsheets/d/1vZfWYrC3HGF4mVzirq7v9PFGJWYGy5_DvR32ssIPFdU/
As David Kriesel once put it so beautifully: “Raw data is hot!” (roughly translated) In this sense, have fun with it! If you discover any more interesting correlations in the data, or maybe just find an error in the formulas, feel free to let us know in the forum thread! 😊
13 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Mitglied
Mitglied
Mitglied
Mitglied
Veteran
Urgestein
Neuling
Urgestein
Urgestein
Veteran
Urgestein
Veteran
Alle Kommentare lesen unter igor´sLAB Community →