Hi all. This is my first article. Excuse me if I make any mistake.
Today discussion is about general PC architecture. The idea is to try to keep it least technical so that non-tech guys can also understand. This should help reader selecting proper system and understanding what does the system specification means. Remember, it is not description parts of PC, its architecture.
There are two major architectures. Intel architecture and AMD architecture. It is not the case that the architectures are strictly followed, there can be some changes from generation to generation.
Difference between architectures is not huge and if one is understood, the other is not difficult. For easy understanding, only desktop motherboard is taken in perspective and not notebook. There will be some minor changes in notebook design to go well with form factor and power.
First is widely used architecture...
Intel Architecture
-----------------------
If you have a little bit knowledge about PC, you may not need explanation. Just referring diagram would be enough for you.
Let me start with the part that we all may know, processor or CPU. As you can see in diagram below, CPU(Central Processing Unit) is not actually in center.
It (processor) talks to ‘North Bridge’, which in turns talks to ‘South Bridge’. Why so much complexity involved? Why do we have 3 different chips to control and why can’t processor itself connect to all devices? There are some sensible reasons.
First, work distribution. If High priority tasks and low priority tasks, both are given to a single chip, it is going to be difficult to keep everything in sync. So better is to let once device handle high speed task and let another handle slow speed. Even technology used for high speed chip should be better than that of low speed chip. And they both can talk to each other using an interface which is good for both.
Second, upgrading the device. If everything is controlled by one chip and later any interface/part (processor, PCI, Memory controller, iGPU, etc...) needs to be upgraded, rest of all interfaces need to go under testing before it is out in market. Each interface takes quite a good time to qualify.
Third, end user choice. Everyone has different choice. Some need higher processor, some need iGPU and so on. Keeping different chips will solve this issue to great extend. But finally, it’s all up to manufacturer which combination they want to roll out in market.
And Finally, if everything is given in a single chip, how will they make profit which they can make by selling 2/3 different chips!!!????
Ok, that was too much of reasoning. Let’s move back to our discussion.
North Bridge is also called MCH (Memory Controller Hub) or GMCH (Graphics and Memory Controller Hub). It communicates to CPU using protocol called Front Side Bus (FSB). It is a proprietary protocol by Intel. (For techies, FSB is a quad pumped interface. If it is written FSB is 800MHz, actual clock is just 200Mhz.) Higher the is FSB, better the speed of system.
Next is Memory or RAM. Currently, more popular are DDR2 or DDR3. As its name says (DDR – Double Data Rate) the data flows at double the rate than frequency of interface. This is highest bandwidth interface on the motherboard.
Then comes PCIe. As you can see, PCIe is there with North Bridge as well as with South Bridge. PCIe devices which may use memory frequently, should be nearer to memory for less latency. These kind of devices should be connected to PCIe connector of North Bridge. iGPU uses system memory, which is generally connected to North Bridge. It is necessary for iGPU to access memory faster so placing iGPU in North Bridge makes more sense than in South Bridge.
At last, North Bridge is connected to South Bridge. We’ll discuss about this after we finish with South Bridge.
South Bridge is also known as ICH (I/O Controller Hub) or PCH (Platform Controller Hub). It takes care of low speed interface. Low here is with respect to Memory or FSB. These includes PCIe (again), USB, SATA and IDE (for Hard Disk and CD/DVD drives), PCI, FireWire, Audio, LAN and Super I/O. Let’s take them one by one.
PCIe here can be some card which doesn’t need more memory access. Though the cards which require frequent memory access can be connected, but in that case, performance will suffer. PCIe is a serial interface and signalling is differential.
USB are driven directly from South Bridge. Current standard is USB 2.0 which supports 480Mbps (Mega BITS per second). They are backward compatible and support USB 1.1 as well.
SATA and IDE are used to connect Hard Drive and Optical drives like CD/DVD/BD drive. IDE are not seen on latest motherboards. SATA has higher speed then IDE and it is serial interface which uses fewer pins for communication.
PCI is the most common interface and it is parallel interface. There are many add-on cards which sit on PCI like LAN, USB, FireWire, and Soundcards.
FireWire sits on PCI bus. This is different then what is mentioned in PCI. Here an IC for FireWire is soldered on motherboard so add-on card is not needed. This is good as it leave a PCI slot unused.
Audio is available as interface in the South Bridge and it doesn’t go to connector directly. This being more analogue and very low speed interface, a codec IC needs to be connected in between. SB talks to it in digital manner only.
LAN/MAC also connects to external IC and from that it goes to connector. It (the external IC) provides physical layer functioning (what kind of electrical signal need to be sent) and more importantly, protect South Bridge from voltage spikes which comes from cable if something is wrong on other end or in between.
And finally it’s Super I/O. It handles very low speed interface like Keyboard, Mouse, Serial port, Floppy drive etc. Speed of these interface are not even 1% speed of SATA or PCI. So it doesn’t make sense to connect it directly to South Bridge. Even connecting all devices needs more no of pins on South Bridge. All these devices communicate to Super I/O which in turn communicates to South Bride using very low pin interface called LPC (Low Pin Count). Flash ROM/BIOS also sit on same LPC bus. Sometimes Flash ROMs works on SPI interface and they are connected to SIO or directly to South Bridge.
A simple hierarchy can be seen here. Super I/O takes data from all very low speed interfaces and communicates to South Bridge using very few pins and at low speed. South Bridge communicates to all low speed interface and then forward data to North Bridge. North Bridge has logic to route the incoming data to CPU or Memory. It also takes care of data from CPU to memory and vice versa and from PCI/PCIe card or GPU/iGPU to memory and vice versa. It also defines priority to data and piping mechanism for data flow.
There are crystals oscillators on the motherboards which are connected to the bridges. Bridges have internal PLLs to convert that low speed clock of oscillators to high speed clocks. Clocks given to almost all interfaces are driven by North Bridge/South Bridge. FSB clock is driven by North Bridge or sometimes an external PLL IC (which is driven by North Bridge).
Let’s have some very basic about FSB and CPU Speed. As it was discussed earlier, if FSB is 800Mhz, then clock speed would be only 200Mhz. So, what does CPU speed shows? And from where that clock comes? One thing to remember is that CPU doesn’t produce any clocks. CPU takes FSB clock as base and multiplies it to get CPU Speed. This multiplier ranges from 4 to 10.
So, that was all about basic Intel Motherboard Architecture.
Now we’ll move to...
AMD Architecture
-----------------------
I hope you will not need much explanation on it and just by looking at diagram, you’ll understand most of it.
The first and major difference is ‘memory’, which is controlled by processor itself. Next is protocol between Processor and North Bridge. It is HT (Hyper Transport). Apart from this, there are no major differences.
Advantage/Disadvantage of integrated memory controller
Processor is a device that makes most use of memory. So it does make sense to keep it nearer to CPU.
If Intel architecture is observed then it’s realized that processor need to go through North Bridge to access memory. This takes more clock cycles and more time which result in slightly lower performance. But keeping it (memory controller) in North Bridge has advantage for PCIe devices that makes direct use of memory.
In AMD architecture, processor can access memory directly without much latency. That is good for performance but when other devices needs to access memory, latency goes up for them. Moreover, CPU needs to come into the picture, hence CPU performance goes somewhat down.
Coming back to discussion, if you see above diagram you can observe that, if iGPU is not needed, then North Bridge is just redundant because PCIe is/are available in south bridge as well. Even iGPU can be integrated in South Bridge.
Same as FSB, HT’s speed is also 4 times the HT frequency and AMD CPUs also have multiplier ranges from 4 to 10.
That’s all about it.
Single Chip Solutions
--------------------------
Just now I stated some reasons why separate North Bridge and South Bridge makes sense. But that was more for Intel then AMD.
What are the arguments to move to single chip solution?
1. Keeping North Bridge and South Bridge in same chip with iGPU also in it makes a lot sense. You just need processor and a chipset and you are good to go.
2. The major benefit is in Notebook space where space is constraint.
3. Take North Bridge and South Bridge and fuse them into one. Only one part in the production. This saves money.
4. If you are competitor of a company with 2 chip solution, you have the chance. The reason is, board manufacturers look for parts that take less space so they can save money on boards.
5. Upgradation cycle for toys attached to Bridges is slow. So upgrading chip won’t be issue for long time.
In today’s modern PC serial/parallel port, floppy drive, PS/2 keyboard/mouse is not used. So in that case SIO IC can also be avoided and all this peripherals can use USBs.
For AMD, Single Chip Solution came earlier then in Intel for a simple reason of memory controller. Intel didn’t get good success with single chip solution until 2008 when 9400M was launched.
Intel’s Nehalem Architecture
------------------------------------
With Core i7, the first processor with Nehalem architecture, Intel follows almost AMD way of integrating memory controller in processor. Still Nehalem is 2 chip solution. It’s out with new DDR3 only. Nothing much to say about Nehalem at this time.
So, I declare end of discussion from my side.
Your comments and questions are welcomed.
The data provided here was as of 25th Dec 2008. Author doesn’t take any responsibility for information provided here.
