In the earliest days, people made a computing integrated circuit ( Circuit
, or IC
), wrapped it with resin and made a small piece ( Chip
), put it on the circuit board ( PCB
), make it into a computer ( Computer
), they distinguish this small piece from other small pieces and call it The central processing unit ( CPU
), the person who writes the software simply calls it the processor ( Processor
).
At this stage, different words refer to basically the same thing.
Later, with the development of technology, in order to provide stronger capabilities, they first installed multiple Chip
into the same PCB
, so that a computer appeared with multiple Chip
, they call this Multi-Processor,MP
. If these Processor
can be treated the same, without each special processing, it is called isomorphic multiprocessing ( SMP
), otherwise it is called asynchronous multiprocessing ( AMP
).
MP
is necessary to stretch the circuit far away, it is not efficient to encapsulate it in a computer, so there is another person in a Chip
Inside, the calculation circuit has been made multiple copies, and each copy is called a "" Kernel" or you translated it as kernel here ( Core
), but still encapsulate the whole Chip
called a CPU
.
But the software has called an executed hardware unit CPU
(that is, the hardware thinks Core
), from here on, the names used by software and hardware are not the same thing . In Chinese, the software refers to the protected software component used by the operating system to manage all resources, also called the kernel ( Kernel
), and each managed software component is called the processor ( Process
, Chinese translation is "process"), which is more likely to cause people's misunderstanding. In fact, no matter Kernel
or Process
, they are all software, and they are things that run on CPU/Core
, which is only a distinction within the software. Just like the relationship between our human body and thinking, CPU
is the body, software is the soul, Kernel
is the soul, Process
is the soul (of course we don't study it carefully). the exact semantics of both )
Core
More and more, a circuit is already difficult to manufacture, they start to make it separately, and then package it in a Chip
Inside, that independently manufactured circuit, some people call it a DIE
, some people call it a Cluster
. The software doesn't actually care, the software only cares about how many " CPU
" (the chip says " Core
"), but CPU
are in different locations. The memory speeds are not the same, so the software will think they are on different "zones" ( zone
).
In order to further optimize Core
, the designers found that many circuits of Core
are idle most of the time, and they let multiple cores reuse these circuits. This looks like a hardware Core
, which provides multiple software to see CPU
, they put this CPU
, called " hard thread " ( hyper-thread
).
Well, I wonder if you fainted as a software engineer? The software engineer asked: Stop so much nonsense, just say how many "fake CPUs" you have, and I will deal with them uniformly.
但硬件说,我有30
个A cpu
, 10
个B CPU
, 4
个C CPU
, 15
pcs MPU
(microprocessor, understood as simpler CPU
), of which A CPU
contains 2 Even a hard thread, three connection methods, B has a hard thread, four configurations...
The software said, stop, stop... Forget it, let's say it separately, you CPU
gave me the same thing, we call it SMP
(note that the meaning of this name has been changed from the previous one different), this CPU
I use it uniformly, and the memory is also accessed uniformly. If you are not like this, I call it "asymmetric multiprocessing" ( AMP
), then you can tell me how to use it one by one.
Did the reader faint? Have you come and let me wash my brains and use the definitions on our server chips :)
A separate hardware processing unit, seen by software as something " CPU
", we call it Core
. But there are exceptions: Core
can have multiple hyperthreads ( hyper-thread,HT
), if there are HT
, each HT
is a software to CPU
, but it doesn't have all the computing resources.
Multiple shared one-level cache
Core
, we call one Cluster
.
Multiple cluster
combined circuits that can be freely combined and packaged, we call a super CPU CLUSTER(SCCL)
.我们还可以把网卡, sas
东西封装成超级IO cluster(SICL)
,从而PCIE
速度的更接近CPU
的IO
device (such as the network port of 100G
).
Combining multiple sccl and sicl
together in resin, we call one chip
.
Multiple chip
are put into one board, inside core
used as SMP
Processor
each chip
Processor
, or from the perspective of PCB
aka a slot ( Socket
).
The circuit board and various external network circuits are packaged into a server. We call one Machine
, and multiple Machine
are combined into a cabinet, which we call one Rack
. For external advertising, we call one of us Rack
can provide more than 10,000 pieces, and the working frequency is 2.6G+
Core
:
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用
。你还可以使用@
来通知其他用户。