MTK Dimensity Tier List (Messy Naming)

Updated on 2024/07/21.

Disclaimer: This list is base on Wikipedia data. The grouping solely depends on CPU architectuname and process node, CPU frequency and GPU are ignored. 

TSMC N4P, 4* Cortex X4 + 4* Cortex A720 (fastest SoC for phones now):

• Dimensity 9300

TSMC N4P, 1* Cortex X3 + 3* Cortex A715 + 4* Cortex A510 (Comparable to Snapdragon 8 gen 2):

• Dimensity 9200(+)
  

TSMC N4, 1* Cortex X2 + 3* Cortex A710 + 4* Cortex A510 (Comparable to Snapdragon 8+ gen 1):

• Dimensity 9000(+)

TSMC N4P, 4* Cortex A715 + 4* Cortex A510 (Comparable to Snapdragon 8+ gen 1, slightly weaker ST performace):

• Dimensity 8300

TSMC N4, 4* Cortex A78 + 4* Cortex A55 (Comparable to Snapdragon 7+ gen 2(which is downclocked 8+ gen 1, weaker ST performace):

• Dimensity 8200

TSMC N5P, 4* Cortex A78 + 4* Cortex A55 (Comparable to Snapdragon 7+ gen 2(which is downclocked 8+ gen 1, weaker ST performace):

• Dimensity 8100
• Dimensity 8000

TSMC N4P, 2* Cortex A715 + 6* Cortex A510 (Low big core count but should be very efficient thanks to N4P):

• Dimensity 7200
• Dimensity 7350

TSMC N4, 4* Cortex A78 + 4* Cortex A55 (Performance comparable to Snapdragon 778G, but should be very efficient thanks to N4):

• Dimensity 7300(X)

TSMC N6, 4* Cortex A78 + 4* Cortex A55 (Comparable to Snapdragon 778G):

• Dimensity 8050 (1300 renamed)
  
• Dimensity 8020 (1100 renamed)
• Dimensity 1300
• Dimensity 1200
• Dimensity 1100
  

TSMC N6, 2* Cortex A78 + 6* Cortex A55 (Comparable to Snapdragon 695):

• Dimensity 7050 (1080 renamed)
• Dimensity 7030 (1050 renamed)
• Dimensity 7020 (930 renamed)
  
• Dimensity 1080
• Dimensity 1050
• Dimensity 930
• Dimensity 920
• Dimensity 900
  

TSMC N7, 4* Cortex A77 + 4* Cortex A55 (Comparable to Snapdragon 865):

• Dimensity 1000(+)

TSMC N7, 4* Cortex A76 + 4* Cortex A55 (Comparable to Snapdragon 855):

• Dimensity 820
• Dimensity 800

TSMC N6, 2* Cortex A76 + 6* Cortex A55 (Comparable to Snapdragon 765):

• Dimensity 6080 (810 renamed)
• Dimensity 6100+ (slower than 6080)
  
• Dimensity 810

TSMC N7, 2* Cortex A76 + 6* Cortex A55 (Comparable to Snapdragon 765):

• Dimensity 6020 (700 renamed)
• Dimensity 800U
• Dimensity 720
  
• Dimensity 700

 

 

 

 

 

QMK, 1000Hz Polling, NKRO, eagar debounce

近排又砌咗隻鍵盤，順便研究埋點樣將佢嘅latency降到同其他電競鍵盤一樣。

雖然呢個group玩自組鍵盤已經係小眾，會自己flash firmware嘅只會更少，但希望呢個方法可以令人少個理由買手感相對差嘅電競鍵盤，試下自組鍵盤。

鍵盤嘅latency主要有3部分: matrix scanning, debouncing, USB polling.

Matrix scanning: 即係控制器掃描所有軸嘅速度，通常都少於1ms.

Debouncing: 機械軸係用2塊銅片做觸發，但銅片喺接觸嘅瞬間會喺0同1之間彈嚟彈去，需要一個delay去無視呢啲輸入。Cherry MX及同類通常都係5ms, 新嘅1億次壽命Cherry MX Hyperglide係1ms, 光軸係0.

USB polling: 用過Gaming mouse都應該識，1000Hz = 1ms, 125Hz = 8ms.

USB polling可以用呢個program試，重可以順便試埋key order嘅問題。

https://github.com/ad8e/input-polling-test

你可以試下同時撳2粒制(例如A, S)，係1000Hz嘅話應該會見到delay係1-7之間嘅數，而且有時係AS, 有時係SA.

遺憾嘅係好多自組鍵盤出廠都係125Hz, 所以delay好多都係10ms左右，好在有好多有線自組鍵盤都用一個強大嘅firmware叫QMK，呢個firmware最大賣點係可以設定鍵盤唔同鍵嘅功能。

https://config.qmk.fm/#/

如果你肯自己compile firmware嘅話，就可以改USB polling rate做1000Hz, 用其他debouncing algorithm去隱藏嗰5ms嘅delay, 甚至自己寫C code都得。

首先喺電腦裝QMK

https://beta.docs.qmk.fm/tutorial/newbs_getting_started

然後喺qmk_firmware/keyboards入面搵返自己隻keyboard

喺config.h度加：

#define USB_POLLING_INTERVAL_MS 1

#define FORCE_NKRO

#define DEBOUNCE 5

再喺rules.mk度加：

DEBOUNCE_TYPE = sym_eager_pk

Compile完之後用QMK toolbox flash入隻鍵盤，咁你嘅delay應該就會係2ms左右。

有錯請指正，以上資料都係用一晚Google得嚟