![]() The E-Core disabled performance of both operating systems is the same. Corona renderer seems to have a big problem with Windows 10, where the scheduler is sending traffic to the E-cores exclusively and the P-cores sit idle. The benchmark simply uses all cores available and scales nicely. We also tested with the E-cores disabled completely, and Windows 10 is slightly faster than Windows 11 even here, although the newer OS gains massive ground.Ĭinebench R23 sees absolutely no difference between the two operating systems regardless of whether we toggle the E-cores on or off. Considering Windows 10 doesn't interface with Thread Director at all and Windows 11 does, this would suggest that the issue lies with Intel's Thread Director, not Windows 11 itself. On Windows 10, wPrime traffic somehow does get sent to the P-cores, where we see a massive performance gain. Synthetics like SuperPi and wPrime love Windows 10 because they're not optimized for modern Windows in any way, and for whatever reason, Thread Director misdirects wPrime workloads to the E-cores. ![]() In our testing, aggregated across all tests, we find that Windows 11 is on average 6.5% faster than Windows 10, but there's a lot more to this. ![]() Much of this functionality has existed in Windows for years, so it's not like Windows 10 is totally unable to make such scheduling decisions. Windows 11 also comes with some improvements to its API for software developers to use, so they can tell the operating system whether something is a background thread, important, or even time-critical or not. To the best of my knowledge, nothing is set in stone yet regarding updates, so we'll have to be patient. I think it should also be possible to distribute those changes through Windows Update for those motherboards that no longer receive BIOS updates (in a couple of years). Going forward, Intel does have the ability to update that model, possibly through microcode updates wrapped into motherboard BIOS updates. It does not "learn"-the model and its decisions are fixed and deterministic. In their press briefings, Intel clarified that Thread Director uses a pre-trained AI model internally. CPPC2 is only a static recommendation that can't change while the system is running. The big difference compared to mechanisms like CPPC2 "Preferred Cores" is that Thread Director runs continuously on the CPU (it's independent and doesn't use up CPU time) and monitors the situation with millisecond accuracy. Intel has worked closely with Microsoft to make sure the new operating system queries all the right capabilities and is set up to benefit from "Thread Director"-a dedicated piece of silicon in the processor that recommends how to distribute threads to Windows. The main reason Intel is recommending Windows 11 over Windows 10 has to do with changes to the thread scheduler, the piece of Windows that decides which core(s) to run any of the hundreds of concurrently running threads on. This, however, comes with the requirement of software optimization (as with Arm big.LITTLE in its initial days back in 2013). ![]() Under the right circumstances, the new cores can be made to contain lightweight background tasks, and contribute to reduced system power consumption. The "Gracemont" E-cores operate in an optimized performance/watt band, and turned out to be a lot faster than expected, enabling Intel to sneak up on AMD's high-core count "Zen 3" processors. While there's people claiming that Intel only introduced E-cores to dial up core counts, we think there's actual tangible benefits to this new approach, and we're just at the beginning of this journey. ConclusionIntel Core "Alder Lake" is a very interesting processor architecture, and given Arm SoCs have had Hybrid architectures for over eight years now, the latest Intel chips can only be considered a step in the right direction. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |