Intel today released the Alder Lake 12 Core developer documentation detailing the architectural configuration of the next generation of processors, and the optimization of the size-core hybrid architecture.
Intel confirms thatAlder Lake-S Desktop Edition has two different cores, one is full of 8 cores, 8 small cores, 32 units nuclear display, the other is 6 large cores, 0 small cores, 32 units nuclear display, that is, some models will have no small core.
The mobile version code is unified as Alder Lake-P, replacing the previous U-series, H-series division, there are two different cores, one is 6 large cores, 8 small cores, 96 units nuclear display, the other is 2 small cores, 8 large cores, 96 units nuclear display.
In other words, the mobile version will have up to six large cores, but all models will have small cores, and the core is three times larger than the desktop version.
It is worth mentioning that the instruction set supported by the size of the nuclear is exactly the same,The only exception is the AVX-512, which is not supported by existing small cores, so if the nuclei are turned on, the nuclei will automatically disable the AVX-512.
On the other handOEMs can decide for themselves whether to block small cores, not shown in the BIOS, this time you are free to turn on the AVX-512.
Intel also confirmed that the 12th generation of Core will be a full coverage of desktop, notebook, ultrabook, flip, two-in-one and other different product form areas, the corresponding models from the fourth quarter of this year began to release.
Intel gives three levels of optimization of size and nuclear scheduling efficiency that are of most concern to everyone.
One is that there is no optimization.
The software itself, or the developer, does not consider the size of the nuclear allocation at all.Based entirely on Intel ITD feedback and algorithms, the Windows System Scheduler intelligently allocates threads and dynamically distributes load.
In most cases, they can do well, but in some cases they may assign critical tasks to small cores or non-critical tasks to large cores, especially those that use multiple middleware components and schedule threads on their own.
The second is Good Optimization.
The software has certain settings that can take advantage of hybrid architectures, but there is no targeted full rewrite.
At this point, the priority load is assigned to the large core and the task system is scheduled according to the number of large cores and the maximum number of threads required for the load to ensure load balance.
There is also thread priority, QoS API, to ensure that the load is given to the right core, such as games, rendering assigned to large cores, background threads to small cores, and so on.
The third is Best.
The software can take full advantage of the hybrid architecture task system and create two thread pools.
One is a pool of priority threads, facing the large core, performing tasks that require a large core or are assigned to a large core.
The other is a pool of secondary threads, facing small cores, performing non-critical tasks such as shader compilation, audio mixing, push streaming, and decompression.
To further optimize the system and increase load balancing,Developers also need to deploy task assignment algorithmsWhen the large core load is too high, turn the priority thread into a secondary thread and hand it over to the small core.
This is especially important for notebooks because they have relatively few large cores.