{"id":66261,"date":"2021-07-19T20:30:32","date_gmt":"2021-07-19T18:30:32","guid":{"rendered":"https:\/\/www.hwcooling.net\/?p=66261\/"},"modified":"2021-08-03T21:35:24","modified_gmt":"2021-08-03T19:35:24","slug":"specs-of-intel-alder-lake-125w-k-processors-leak-clock-rate-pl2","status":"publish","type":"post","link":"https:\/\/www.hwcooling.net\/en\/specs-of-intel-alder-lake-125w-k-processors-leak-clock-rate-pl2\/","title":{"rendered":"Specs of Intel Alder Lake 125W ‘K’ processors leak: clock rate, PL2"},"content":{"rendered":"

Last week we’ve had reports of the big.LITTLE Intel Alder Lake processors reaching surprising performance<\/a>: with just 8 “big” cores out of the total 16, they are said to surpass the 32-threaded Zen 3 in Cinebench R20. Now we the specs of all three 125W unlocked models, with clocks and power draw in boost, aka PL2. Looks like 10nm process will finally lower the power consumption compared to the 14nm CPUs\u2014but not by much.<\/p>\n

The specifications of Alder Lake desktop processors (aka 12th Core generation) for the LGA 1700 socket have appeared on the Chinese forum Zhihu. As usual, you should still take them with a grain of salt, but the data do look quite plausible. Intel has reportedly been supplying so-called qualification samples to its partners for a few weeks now, which are pre-production CPUs that have the same specifications as the final models. The specifications are probably derived from the QS chips.<\/p>\n

Intel will reportedly label the “big” high-performance cores in Alder Lake processors (these have the Golden Cove architecture, which is said to have up to 20% better IPC than the current Willow Cove\/Cypress Cove on Tiger Lake and Rocket Lake processors) as “P-Core<\/b>“, where ‘P’ probably stands for Performance. In contrast, the “little” cores with Gracemont architecture (speculated to have IPC close to Skylake processors, however the cores lack HT andonly provide 1 thread) will be called “E-Core<\/b>” in Intel’s marketing language, which likely means Efficiency.<\/p>\n

By the way, although we still use the term “big.LITTLE”, that was originally promoted by ARM (who however kind of phased out the use of the label in the meantime, too), Intel does not use this term. The company refers to this concept of processors employing multiple types of cores as “Intel Hybrid Technology”.<\/p>\n

Specifications of Alder Lake for LGA 1700<\/h3>\n

Now let’s proceed to the specifications. The highest performance model should be Core i9-12900K<\/b> with eight P-Cores and eight E-Cores and 30MB L3 cache. Its clock speed should be consistent with the previous leak. The maximum boost is 5.3 GHz\u2014the same level as what was reached by 14nm Intel CPUs after years of tuning<\/a>. This can be considered a success, as increasing the IPC did not decrease the achieved clock rate.<\/p>\n

This frequency will be available for one or two P-Cores\u2014this may be dependent on temperature via Thermal Velocity Boost, and this clock is also likely only reached on favored cores. But that is nothing new. If you put all eight P-cores under load, their maximum clock speed is 5.0 GHz\u2014this is an increase over all-core boosts of existing 14nm CPUs (except for the atypical Core i9-9900KS<\/a>).<\/p>\n

\"\"<\/a>
Processor sample of Intel Alder Lake-S for LGA 1700 socket (Source: VideoCardz)<\/figcaption><\/figure>\n

The E-cores of this processor have a maximum boost of 3.9 GHz. This time it is available for four cores simultaneously loaded, but it is not entirely clear whether this is meant for all cores being active in one of the two 4-core clusters into which the E-cores will be grouped, or whether this clock will be provided by four cores divided across the two clusters. When all eight E-cores are under load, the maximum frequency will be 3.7 GHz.<\/p>\n

Boost power draw remains at high levels: PL2 at 228 W<\/h3>\n

Of course, if the entire CPU is under load at once, which occurs in multithreaded applications, then it is possible that the frequencies will be lower than the specified maximum values. Intel is again going to specify the maximum power consumption at boost, the so-called PL2. The value of TDP (which is also called PL1) is 125 W, which means that at base clock (which we do not yet know) the power consumption shall not exceed 125 W. However, it can be exceeded in the boost and the processor will be allowed to draw power up to the so-called PL2 value. This is set at 228 W for Alder Lake. Therefore, even with these CPUs being manufactured by a more efficient 10nm process, they will maintain the same boost behavior with power draw significantly exceeding the TDP as was present in previous generations.<\/p>\n

However, the boost power consumption should be lower than with 14nm processors, such as the i9-10900k or i9-11900k\u2014these have a PL2 of 250 W or 251 W<\/a>. That said, 228 W means not even a 10% reduction, when going down from those values.<\/p>\n

Intel currently specifies that this PL2 power consumption value should be drawn only for a limited time (so-called Tau) and then it is supposed to drop to PL1 (TDP) level. For the 125W Comet Lake and Rocket Lake processors, the Tau was 56 seconds (28 seconds for the 65W models), but we don’t know if this will change or stay with the Alder Lake. However, the general problem is that Intel allows motherboard manufacturers to ignore the Tau limit and set the duration of PL2 power draw to unlimited, and they can also increase PL2 and PL1 themselves. While office or business OEM PCs typically adhere to the limited duration of the high-power boost, desktop motherboards sold separately on the retail\/DIY component market, especially the more expensive Z-series chipset motherboards, tend to ignore the limits and the processor can permanently be at that higher limit in such boards. Thus, Alder Lake would draw up to 228 W on such motherboards\u2014depending on the task running, naturally. Power consumption should be lower in games, but\u00a0 you could shoot above 200 W when encoding using x264 or x265, for example.<\/p>\n

\"\"<\/a>
LGA 1700 socket components for Alder Lake processors (Source: Taobao)<\/figcaption><\/figure>\n

Core i7 and Core i5: less cores, constrained L3 cache<\/h3>\n

In addition to the flagship Core i9-12900K, a cheaper Core i7-12700K<\/b> will be offered. This is a 12-core that has 8 big P-Cores and four little E-Cores. The clock speed of the P-Cores has a maximum of 5.0 GHz, and this frequency is applicable to a maximum of two cores. All-core boost, i.e. maximum if all P-Cores are loaded, is 4.7 GHz. Little cores also have a variable maximum, although there are only four of them here. When the maximum of two is loaded, their clock speed can be 3.8 GHz, and 3.6 GHz when all four are loaded.<\/p>\n

Again, we do not know the base clock speed. TDP is 125W, but the maximum boost power draw of PL2 is allegedly the same as for i9\u2014up to 228 W. An interesting information is the L3 cache capacity, which is 25 MB. As we wrote recently (see link bellow), the L3 cache will probably work in such a way that one 4-core cluster of little cores will have one common block of L3 cache belonging to it (while big cores each have their own piece of L3 individually). For the Core i9 model, the L3 cache has 3 MB per core\/cluster, so we have 8 \u00d7 3 MB + 2 \u00d7 3 MB, a total of 30 MB.<\/p>\n

Based on the 25 MB total L3 on the Core i7-12700K, Intel seems to limit the L3 cache of the Core i7 from 3 MB to 2.5 MB for each P-Core core (or E-Core cluster). Interestingly, the four little cores on the i7 probably come from two clusters, because we still have 8 \u00d7 2.5 + 2 \u00d7 2.5 MB L3 cache. When disabling the little cores, Intel does not turn off the whole cluster at once, but only disables part of the cores in each of the two clusters\u2014while the rest then has the entire original L3 block retained.<\/p>\n

Read more: Intel Alder Lake: information on cache in Golden Cove and Gracemont and how the big.LITTLE works<\/a><\/b><\/p>\n

It seems there will be three 125W Alder Lake models with unlocked multipliers for overclocking\u2014the last one is the Core i5-12600K<\/b>. For this model, Intel will disable two more big cores, leaving 6 high performance P-Cores and 4 E-Cores. These are probably again in two clusters and the processor again seems to have 2.5 MB of L3 cache per core, as the total capacity of L3 cache is 20 MB (6 \u00d7 2.5 + 2 \u00d7 2.5 MB = 20 MB).<\/p>\n

With this processor, the maximum clock speed of big cores is 4.9 GHz, and again it will be available on a maximum of two cores. The maximum boost for all big active cores loaded is 4.5 GHz. For little cores, the clock rate was reduced by 200 MHz compared to the i7 model\u2014the clock for a maximum of two active E-Cores is 3.6 GHz, for all four E-cores active the maximum is 3.4 GHz. This CPU also has 125W TDP\/PL1, but again has the same 228W PL2 specified. It seems Intel will not set different boost power consumption limits for individual models.<\/p>\n

\"\"<\/a>
Reported specifications of 125W Intel Alder Lake desktop processors (Source: Zhihu)<\/figcaption><\/figure>\n

The leak does not indicate pricing yet (and it is probably too early for that). We do not have information about the specifications of the integrated GPU either. But the previously leaked information<\/a> states that the iGPU in mobile Alder Lake should have 96 EU (768 shaders) of the Xe architecture, seemingly similar to that of Tiger Lake and Rocket Lake. However, the desktop Alder Lake we were just discussing is supposed to have a slower iGPU with only 32 EU (256 shaders), which means it should be quite similar to Intel UHD Graphics 750 in Rocket Lake processors.<\/p>\n

Read more: Leaked info on Intel Alder Lake: PCIe 5.0, DDR5, platform and performance<\/a><\/b><\/p>\n

DDR5 in Z690, but cheap motherboards rather with DDR4?<\/h3>\n

It has also been leaked previously that the processors will have a PCI Express 5.0 \u00d716 controller for the graphics card, but probably only PCI Express 4.0 \u00d74 for NVMe SSDs. DDR5-4800 memory will be supported, which will be used on more expensive motherboards with the Z690 chipset. According to Twitter leakers, however, supposedly cheap motherboards will usually use DDR4 memory (DDR4-3200 will be\u00a0officially supported). Most or almost all motherboards with H610 and B650 chipsets could be outfitted for use with DDR4 memory, and even some cheaper Z690s are supposed to use DDR4 memory instead of DDR5.<\/p>\n

\n

All B\/H series boards are DDR4 and low end Z690 is DDR4 also.<\/p>\n

— Kepler (@Kepler_L2) June 30, 2021<\/a><\/p><\/blockquote>\n