当前位置:网站首页>Solve the problem of intel12 generation core CPU [small core full, large core onlookers] (win11)
Solve the problem of intel12 generation core CPU [small core full, large core onlookers] (win11)
2022-07-06 16:37:00 【Ruo Miaoshen】
List of articles
This article seems to have nothing to do with software , Actually, it has a great influence .
You can skip the boring recording process , Look directly at the solution :)
( One ) Causes and problems
1.1 upgrade 12 Generation core
Upgrade the desktop computer at home a few days ago ,CPU The picture is cheap and uses the one without nucleus Intel12 Generation of core QS edition .
Reference link : Parameters .
as well as :《 The first 12 On behalf of Intel core Hybrid Technology How to work 》
Its own operating system is Win11 了 , So I didn't reinstall ( The test conditions are more fair ).
This generation and the past Intel Processors are different , For the first time, the design of large and small cores is used ( Of course, nuclear size is a folk saying ).
- Big nucleus ( Performance core ):Performance-core, abbreviation p-core.
- Micronucleus ( Efficiency core ):Efficient-core, abbreviation e-core.
This one is specific CPU Yes 8 individual p-core,8 individual e-core. adopt BIOS switch e-core You can get different combinations of core numbers .
So here we go 8P8E For short, the default full open 8 individual p-core,8 individual e-core.
and 8P0E Is to close all e-core.
1.2 The running speed of the program is worrying
I'm very happy after the upgrade , I immediately tested it with my own program :
The whole core 8P8E I used it 35 second ,
Than before upgrading i9-9900(ES) Of 40 second Only a little faster .
Than the company's i9-9900KS Server's 28 second More than a little slower , Is slow , Slow down , Slow down ……
The observed phenomenon is probably what the following task manager looks like :
The small core is full ( Last 8 individual ),
The big core is idle ( front 16 individual : Multithreading x2).
Try a few of your own programs again , The result is almost .
adopt BIOS Medium close small core , The test is much faster again .
About the following :
Program | 8P + 8E | 8P + 0E | 9900(ES) | 9900KS | Company |
---|---|---|---|---|---|
Go Multithreading | 35 | 26 | 40 | 28 | second |
Pascal Single thread | 68 | 48 | 79 | 70 | second |
Spark on VM | 120 | 60 | 80 | second |
Repeatedly switch on and off the machine to test various conditions .
It is similar to the conclusion on the Internet , A very strange summary comes :
- The program indicates 16 Threads or more , It will only be full 8 A small core .
- Step by step BIOS Reduce the small core :8->6->4->2->1. Even if only 1 A small nucleus will only be full of small nuclei , The big core continues to watch ( The program is extremely slow ).
- Only completely close the small core , Only then can the big nucleus normally participate in the work ,CPU You can also use 100%.
( Two ) Theoretical achievements are displayed
Although the actual program testing is hard ……
But under various evaluation software , There is no such problem , The performance is strong enough to break the ceiling .
2.1 CPU-Z
2.1.1 【8P0E】
2.1.2 【8P8E】
2.2 Aunt Lu
Also called entertainment master , But it seems that everyone likes her ……
2.2.1 【8P0E】
2.2.2 【8P8E】
2.3 Other test software
Self testing CB edition R15,R20 The results are basically the same , It looks very good .
Chess doesn't seem to recognize the size of the core , The result is a little strange .
For more tests, please see professional evaluation .
In short, the evaluation results of the software are good !!! Poor operation of actual program !!!
( 3、 ... and ) solve : Set the power mode
️ Notice that it's not Power scheme in the control panel , Keep there equilibrium Don't move .
3.1 Set up
Get into : Set up -> System -> Power Supply -> Power mode -> Best performance .
Here's the picture , You don't even need to restart :
Add : On the other one is not 12 Generation core CPU On .
Same location , No, Power mode This setting . Here's the picture :
3.2 The test again
The test again , It is observed from the task manager that the big core is also normally involved in work :)
The supplementary figure is as follows , The task is much faster , The screenshot is complete …… Khan, .
Tested that the small core just now is full , The procedure of big nuclear onlookers , This time it's just 17 second .
Than 8P0E Of 26 second Significantly improve , Than before setting the power mode 8P8E Under the 35 second It's a big difference .
give the result as follows :
Program | 8P + 8E | 8P + 0E | 8P8E performance | 9900(ES) | 9900KS | Company |
---|---|---|---|---|---|---|
Go Multithreading | 35 | 26 | 17 | 40 | 28 | second |
Pascal Single thread | 68 | 48 | 59 | 79 | 70 | second |
Spark on VM | 120 | 60 | 54 | 80 | second |
PS: Best performance In mode , Observe that single threaded programs still switch between cores , Most of the time it runs on small cores ……
It seems that the scheduling and Win10 Dissimilarity , And not satisfied CPU It may fall on the small core ... This In addition, continue to study .
Multithreaded program running log :
( Four ) Discuss
4.1 Operating system and size core scheduling
The news from the Internet is , must Win11 To properly schedule the size of the core .
It is also said that Win10 In fact, there is already ( because ARM qualcomm CPU Well ) But the scheduling strategy is not open .
I don't see anything about Win11 It must be opened 【 Best performance 】 Pattern …… Or I don't look carefully .
In short, a lot of time was wasted in the process ……
Unwilling to turn off the small core ...
If you turn off the small core , That is, an enhanced version 9900KS 了 ...
4.2 The core of efficiency is not weak
From the test results , The small core is not weak .
I found This detailed evaluation I mentioned ,
After rough calculation ,12 Generation nuclei are equivalent to 9 Generation normal core runs in 3.7GHz Frequency .
Except for no hyper threading , I won't support it AVX512, Just buy one and get one free :)
Last , I still hope to exchange small cores for large ones .
If you have to add a limit , that 8 Change a small core 4 A big core
Add : Let's meet each other halfway Intel,8 Change a small core 2 A big core is ok
边栏推荐
- Chapter 5 detailed explanation of consumer groups
- js时间函数大全 详细的讲解 -----阿浩博客
- AcWing:第56场周赛
- SF smart logistics Campus Technology Challenge (no T4)
- 解决Intel12代酷睿CPU单线程调度问题(二)
- pytorch提取骨架(可微)
- Market trend report, technological innovation and market forecast of China's double sided flexible printed circuit board (FPC)
- useEffect,函數組件掛載和卸載時觸發
- Codeforces round 797 (Div. 3) no f
- Advancedinstaller installation package custom action open file
猜你喜欢
Li Kou - 298th weekly match
第五章 Yarn资源调度器
Pytorch extract skeleton (differentiable)
sublime text 代码格式化操作
QT style settings of qcobobox controls (rounded corners, drop-down boxes, up expansion, editable, internal layout, etc.)
Installation and use of VMware Tools and open VM tools: solve the problems of incomplete screen and unable to transfer files of virtual machines
业务系统从Oracle迁移到openGauss数据库的简单记录
第6章 Rebalance详解
Li Kou: the 81st biweekly match
Flask框架配置loguru日志庫
随机推荐
第6章 Rebalance详解
QT实现窗口置顶、置顶状态切换、多窗口置顶优先关系
Codeforces Round #802(Div. 2)A~D
Problem - 1646C. Factorials and Powers of Two - Codeforces
Codeforces Round #800 (Div. 2)AC
Market trend report, technical innovation and market forecast of China's desktop capacitance meter
第一章 MapReduce概述
Bidirectional linked list - all operations
简单尝试DeepFaceLab(DeepFake)的新AMP模型
(lightoj - 1354) IP checking (Analog)
树莓派4B64位系统安装miniconda(折腾了几天终于解决)
875. Leetcode, a banana lover
SF smart logistics Campus Technology Challenge (no T4)
第6章 DataNode
解决Intel12代酷睿CPU单线程只给小核运行的问题
Classic application of stack -- bracket matching problem
Generate random password / verification code
Some problems encountered in installing pytorch in windows11 CONDA
Base dice (dynamic programming + matrix fast power)
AcWing:第58场周赛