MC SP and its low significance
Recently, Asus introduced an additional SP function with BIOS updates, which can now also evaluate the quality of the CPU’s integrated memory controller (IMC/MC). However, this value is not nearly as reliable as the SP of P and E cores for several reasons.
According to this post by Asus R&D engineer shamino, here only the System Agent component with its voltage is tested. The other parts of the CPU, which are also responsible for the working memory, with their voltages CPU VDDQ (VDDQ TX) and VDD2 (MC Voltage), are not tested. Furthermore, with these voltages there is no V/F curve ex works, but only a fixed stock value, which is used for all frequencies. Thus, Asus cannot rely on Intel’s preliminary work here, but has to test the scaling of voltage and frequency at runtime themselves.
This in turn brings other difficulties. On the one hand, this test depends on the CPU’s temperature, so we could measure fluctuations in the MC SP of up to 5 points from test run to test run with the same CPU. On the other hand, a test run to determine the MC SP takes about 1 minute, which generates a considerable amount of additional time for data acquisition. Accordingly, we only determined the MC SP data at all for those CPUs that already have the highest SP values for P and E cores. Thus, no general conclusions about the MC SP distribution can be drawn from this, but at least we can see whether a high MC SP is related to a high SP for the cores, for example, or not.
By the way, the MC SP is not part of the overall SP and is also not dependent on the RAM kit used. In addition, I was already able to test that a high MC SP is often, but not always, also indicative of the maximum RAM clock rate that a CPU can support. However, this maximum bootable RAM frequency is also not in linear relation to the maximum stable RAM frequency. As an example, I have additionally tested 3 CPUs on a Maximus Z790 Apex motherboard and also for real stability in Y-Cruncher, Karhu and TM5 “Extreme1@Anta777”.
A higher MC SP is therefore far from being a guarantee for a higher stable or even bootable RAM clock, but at most one of several prerequisites.
Summary
The findings are sobering insofar as no conclusions can be drawn from one SP value to the others and the question about RAM-OC potential cannot be answered easily with the MC SP. This was already the case with the last generation, where we were also allowed to do some “bucketing”, albeit with a much smaller number of CPUs. I’ll link you to the article again here:
At least we were again able to collect data about how the SP values are distributed over a larger amount of chips, and this time with a statistically non-insignificant amount of CPUs. For example, if you have a Raptor Lake i9 CPU with a total SP of over 110, you can consider yourself lucky and be among the top 3%. Most CPUs then exist at a value around 100 and everything that falls below 90 are the proverbial potatoes.
Mind you, all these statements only count for i9-13900K and 13900KF CPUs. i7 and i5 SKUs and also older CPU generations are again in other SP ranges and therefore cannot be compared directly via the SP value! It also remains to be seen whether the upcoming i9-13900KS CPU will fall into the same scale as the other i9 chips or whether Asus will again use other rating constants here. Hopefully, we will be able to go on a silicon treasure hunt for the launch next year and possibly also check the older SKUs for any changes in the silicon quality since their launch.
The significance of the RAM controller with the MC SP is significantly lower. A very low value here also means low potential, but conversely a high value does not necessarily mean high potential. Furthermore, the maximum bootable clock or the clock that can be used for benchmarks do not always behave in relation to the maximum stable clock.
We can only hope that the MC SP metric will be further developed and refined. Especially since we have only talked about DDR5 so far and DDR4 can behave completely different with the same CPU, as we could already see in the article linked above. But at least I can let you know that my MC SP 73 CPU manages to keep the Asus Maximus Z790 Hero running stably at the maximum RAM clock from the QVL at DDR5-7800. A supposedly “bad” MC SP value does not necessarily mean the end of the day.
At this point, I’d like to say thanks again to MIFCOM for their support – if you’re looking for a new gaming PC, you should definitely check out their website or their new showroom. All that remains is for me to wish all readers a Merry Christmas, whether you spend it tinkering with your PC, spending time with your loved ones, or both.
31 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Veteran
Mitglied
Urgestein
Mitglied
Veteran
1
Urgestein
Moderator
Mitglied
Urgestein
Urgestein
Urgestein
Mitglied
Urgestein
Urgestein
Mitglied
Urgestein
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →