What do you do with two Core i9-12900K trays? That’s right, binning! Today we have a very special treat for you from the south of Germany. The system integrator MIFCOM has kindly provided us with a large amount of Alder Lake i9 CPUs so that we can put them through their paces and examine them for differences in clock speed. Originally, the premise was to find a CPU with the fastest memory controller possible for our DDR5 tests. However, we were able to collect a lot of other interesting data while testing the CPUs, which we naturally don’t want to withhold from you. The result was a partly very different picture, because somehow there was no such thing as THE best CPU. But I don’t want to spoil anything yet!
A huge thank you goes to the colleagues at MIFCOM, who entrusted us with this terrific loan!
But we are not the first to receive such an honor. The German overclockers Bullshooter, Sergmann, DaGunzi and quiekMew have already tested several hundred CPUs shortly after Alder Lake launch and have kindly made the data publicly available. I will use this as a reference in the course of my piece today, thanks to collegial permission.
Binning, VID and Silicon Prediction
“Silicon lottery” is a common term in enthusiast communities to describe the variation in the quality of computer chips. Put simply, there are inevitably slight quality variations in the manufacturing of semiconductors, which is why each individual chip requires different amounts of voltage to achieve a certain clock speed.
This is precisely where so-called “binning” comes into play, whereby chips are individually tested for their characteristics and then assigned to a grade (bin). The top grade or top bin is usually referred to as “golden” because these chips are particularly rare, powerful and correspondingly valuable. Likewise, parts of a chip, such as individual cores, may not work at all. However, instead of disposing such a chip completely, the defective parts are simply switched off and the CPU is then recycled as a lower SKU. For example, this way i7-12700K CPUs are created from the same silicon blueprint as i9-12900K CPUs.
And even if all functional units of the CPU are working, the different voltage requirements have to be considered so that each CPU also reaches the specified clock and of course does not become unstable. Instead of finding a common denominator, so to speak, and simply specifying a voltage value with which all CPUs would run stably, this is instead tailored to each individual chip in modern CPUs.
For this purpose, each CPU is precisely tested at the factory for its requirements and then the necessary voltage is permanently stored as a VID (Voltage Identification Definition) table. With Intel CPUs since Rocket Lake, each core even has its own VID table to enable the most efficient operation even in partial load scenarios or with sleeping cores. Conversely, too much voltage would mean unnecessary power consumption and waste heat.
Asus has used this VID table for several generations to calculate a “Silicon Prediction” (SP) on their Maximus and Strix motherboards. The principle is relatively simple: If Intel has already programmed a lower voltage for a CPU from the factory, there is more potential left for raising the voltage and thus overclocking. Since Alder Lake CPUs have Performance and Efficiency cores with their own VID tables, the SP value is broken down again for each. With a 2:1 weighting, the final result is the total SP value. Using the example above: (101 * 2 + 79) / 3 ~= 93. A higher SP value is better.
Now it should be mentioned that the VID or the SP value does not tell the whole overclocking story by far. Equally important is of course the cooling of the CPU and also the “leakage”, another characteristic of each individual CPU influencing how much resistance, thus power consumption and waste heat it has. While the latter has almost no influence on the achievable clock in normal operation with conventional cooling methods, this however becomes especially important in extreme overclocking. This is because CPUs with high leakage can often be clocked significantly higher at temperatures below zero than CPUs with low leakage.
21 Antworten
Kommentar
Lade neue Kommentare
Veteran
Urgestein
Veteran
Mitglied
1
Mitglied
1
Urgestein
Mitglied
Neuling
Mitglied
Urgestein
Urgestein
Mitglied
Urgestein
Veteran
Urgestein
Veteran
Veteran
Alle Kommentare lesen unter igor´sLAB Community →