Форм-факторы материнских плат (Basic PC architecture)

Basic PC architecture

 

 

The basic components that make your PC work

 

Okay, a little while ago I asked for input on what guides to write next and thank all of you for a lot of reactions! Most asked for were graphics cards, motherboards, SSDs and a basic overview. I will address all of these (and more) in separate writeups, but to be able to understand why one component may be better than another we'll need to do the overview first.

 

Basic components

 

I think most of us know what the basic components inside the computer are but I'll list them first and then go into detail on each of them:

 

The processor or CPU

The mainboard with all it's components / chips.

The memory

The PCI Express slots (discussed in a separate article here)

The harddisk and/or SSD (to be discussed in 2 upcoming writeups)

The graphics card (discussed in a separate article here)

The power supply or PSU (discussed in a separate article here)

 

The CPU, mainboard and memory are closely tied together so we'll be discussing these three in this article. Your choice of CPU architecture determines what type of mainboard you'll need and mainboard architectures between AMD and Intel are different. The CPU choice also determines the type of memory you'll need.

 

The mainboard (AMD)

 

For AMD mainboards we'll mostly be talking about the latest generation of socket AM3+ mainboards. Most of the design however is identical to older AMD mainboards. Please bear in mind that the only real AM3+ boards are the ones with an AMD 9xx based chipset. When we talk about the various features of AMD boards here we are referring to 9xx based boards. If you have an older chipset you may not have all of these features.

 

AMD based mainboards have 2 chips that comprise the mainboard chipset: the northbridge chip and the southbridge chip.

 

The Northbridge

 

The northbridge on AMD mainboards is mainly home to the PCI express controller. This chip therefore determines how many total PCI express lanes can be used on a particular mainboard. This is why 990FX boards can do x16/x16 dual graphics and 990X and 970 boards can only do x8/x8 or single x16 graphics. The northbridge also houses the IOMMU unit which is a type of memory controller that allows other chips direct access to the computer's main memory in a flexible manner.

 

The Southbridge

 

The southbridge on AMD mainboards houses the AMD SATA controller, AMD USB2 controller, the controller for the legacy PCI bus and hosts and several single lane PCI express connections. It also connects to the onboard audio chip on the mainboard and the onboard network controller(s) on the mainboard as well as the legacy I/O controller chip which houses things like the old parellel printer port, serial port and PS/2 mouse/keyboard connectors. The way these items are wired up to the southbridge is either via an internal PCI bus connection or an internal PCI express connection. To the operating systems these devices appear the same way as when they would be on an expansion card in a PCI or PCI express slot.

 

In AMD architecture the CPU houses the memory controller and so the CPU connects directly to the DIMM slots from it's socket. The CPU also connects directly to the northbridge chip but communicates with the southbridge chip and any devices wired to it via the northbridge chip. For reference, here is a schematic overview of how it all ties together:

 

 

In AMD architecture the connection between the CPU and the northbridge is what is called a HyperTransport link. This name is not unique to this connection, in multiprocessor (multiprocessor, not multicore!) AMD systems the links between the processors are also HyperTransport links.

 

The communication link between the northbridge and southbridge chips is called Alink Express (version 3 in the case of the 9 series chipsets). This name is unique to that specific connection.

 

The mainboard (Intel Sandy Bridge socket LGA 1155 with P67/Z68 chipset)

 

One of the things you'll immediately notice when looking at the Intel mainboard diagram is the absence of a northbridge chip:

 

 

 

Sandy Bridge socket 1155 systems don't have a northbridge chip and instead the PCI express connections that are used for the graphics card(s) run directly from the CPU to the slots. The southbridge chip connects directly to the CPU and houses the same basic functionality as the southbridge chip in AMD based systems.

 

On the 1155 platform the link between the CPU and the northbridge (Z68/P67) chip is called a DMI link. This same link is used on the older X58 enthusiast platform for communication between the northbridge and southbridge chips.

 

The mainboard (Intel socket 1366 with X58 chipset)

 

The older (but still very much up to date for it's age) socket 1366 with the X58 chipset strongly resembles the current AMD architecture. There is a Northbridge and a Southbridge chip, both of which house the same basic functionality as they do in the AMD Northbridge and Southbridge chips.

 

 

The mainboard (Intel Sandy Bridge Extreme, socket LGA 2011 with X79 chipset)

 

Sandy Bridge E with the new and massive socket LGA 2011 as well as the new X79 chipset is a lot beefier than the mainstream LGA 1155 platform in almost every respect. It has a high number of PCI Express lanes running directly from the processor, double the memory bandwidth and the processors come in quad and hexacore variants. The CPUs all have 10MB cache, an almost 67% increase over the mainstream Core i5! The PCI Express controller on the CPU die is wired up slightly differently so that it has lower latency than it's older Sandy Bridge counterpart and does a speedy 1GB/s per PCI Express lane of which it has a total of 40! The diagram shows a clear difference in all the stuff connecting directly to the CPU although the X79 Southbridge has the same facilities as the current P67 and Z68 chipsets. The exception here is probably the SSD caching feature although that appears to be an all-software solution that is programmed to work only on Z68.

 

 

 

 

In general

 

For all architectures discussed here (AM3+, LGA 1155, X58 and LGA 2011) processors the memory controller resides on the CPU. The CPU therefore determines whether dual channel, triple channel or quad channel memory is supported. Obviously the communication lines from the CPU to the memory need to be on the mainboard as well. This is one of the reasons that socket 1366 with triple channel memory has more pins (1,366 pins) than Sandy Bridge (1,155 pins) which only supports dual channel. Sandy Bridge E (socket 2011) has a massive pin count of 2,011. Although officially, socket 2011 supports PCI 2.x tests have recently confirmed that it runs successfully on 3.0 speeds with the new HD7970 graphics card from AMD.

 

 

 

 

 

 

 

 

 

 

 

The CPU

 

Okay now that we have the mainboard basics down, let's do a basic overview of the CPU. There are a number of things of importance about the CPU when building a PC:

 

The package

 

This is the other side of the socket if you will, a CPU in an LGA 1155 package will fit in a socket LGA 1155, AM3 package will fit in AM3 as well as AM3+, an LGA1366 package will fit in socket LGA1366 and so on. Socket LGA1155 has 1,155 pins running from the CPU to the mainboard, LGA1366 has 1,366 pins and socket LGA2011 has a mindblowing 2,011 pins. AMD's socket AM3 has 941 pins and AM3+ 942 pins. The lower number of pins on AM3 and AM3+ is mainly due to the PCI express lanes not running from the CPU but from the Northbridge chip.

 

More or less pins does not mean faster or slower, better or worse. There are advantages to integrating a lot of stuff onto the CPU die but also some disadvantages. Lower latency between various parts can be one advantage. The more you pack onto a die however, the more heat it will obviously produce and at some point the extra space stuff uses up on the processor die might also be better spent on packing more cores or a larger CPU cache (more on this later).

 

 

 

The package really is just a package as you can see in this picture here :-) The piece of silicon in the middle of the package is the actual processor (or: what you pay the real money for). The metal casing that usually encloses this is really the first stage of the heatspreader :-)

 

Processor cores

 

Moving on to what is actually in (or on) that piece of silicon, obviously there will be 1-8 processor cores on there. Below are two closeup pictures of Intel's Sandy Bridge and Sandy Bridge E processor dies that highlight the various things that are on it.

 

 

 

To see which part of the die is occupied by which components, click on the links:

 

See all

Cores

Graphics

Cache

I/O

 

Cache

 

It used to be that cache was not as overlooked as it seems to be lately. The cache memory of a processor is actually one of the most important factors that determine performance! The general rule here is that the larger the cache the better the performance. There is a point at which larger cache is going to slow the processor down but the cache sizes on modern processors are obviously not so large that this occurs (who buys a more expensive CPU that is slower?).

 

This leads us to another factor of how cache size is determined. Cache memory on the CPU is a special and ultra fast kind of memory that takes up a lot of space on the processor die. Making it larger costs money but it also takes up space that is sometimes better used for putting an extra processing core on that same die. So usually the cache size and the number of cores are a compromise that (should) make for the optimal performance. It's worth noting that besides the cache size, there are different ways a cache can be organized which can also lead to performance differences. On top of all that, the way a computer program goes about doing it's thing can impact the performance of a CPU cache in different ways as well :-)

 

Since we're on the subject of cache, one of the reasons that the Core i7 is faster than the Core i5 is it's larger cache size of 8MB vs 6MB. This is over 30% extra cache on the i7. Hyperthreading can actually slow down programs that are not specifically designed to take advantage of it. Turning it off on a gaming PC leads to basically an i5 with 30% more cache (and thus better performance in most if not all applications) and most likely a better overclockable processor than the i5 2500K. Better overclockable is due to the i5 and i7 basically being the same processor with some features disabled at the factory. The dies that come out top-notch become i7s and the "lesser" dies get some stuff disabled and become i5s :-) Selection in the factory is however also done with supply and demand in mind so it can happen that most of the dies come out very good and still get features disabled because of market demand for the cheaper editions. How many dies come out good vs not so good is however kept confidential by manufacturers. It's one of the reasons that a certain batch of cheap processors may be much more 'overclockable' than another production run of the same processor type.

 

Now back for a minute to why cache matters. The fact is that most of the operations that a processor performs takes place in the cache. A processor in the old days read an instruction from memory, performed an operation and then went on to read the next instruction from memory. Modern processors read those instructions from their cache and this cache is intelligent enough to predict what is needed next and fetches this information from the main memory into the cache before it's actually needed. 6MB of cache for instructions is actually huge. It fits the entire text of the Bible at least 1.5 times! A lot of what a computer does is loops in which the same series of instructions is carried out a certain number of times. As a result the CPU does not need to access the main memory very often. This is the primary reason you don't see any benefits from faster memory!! Only in tasks like video encoding does main memory usage spike but for the vast majority of stuff that a CPU does the memory speed really doesn't matter all that much! This is also the reason that overclocking your CPU without overclocking the memory gets very good results: with overclocking the CPU you are also overclocking the CPU cache.

 

The memory controller

 

As was mentioned in the mainboard overview, the memory controller determines how many memory channels a processor can handle. For both AM3+ and 1155 packaged processors this is dual channel. For 1366 processors this is triple channel and for the new LGA 2011 processors (Sandy Bridge E) this is quad channel.

 

This memory controller also determines what type of memory can be hooked up to the processor (for AM3, AM3+, LGA 1155 and LGA 2011 this is DDR3) and determines the maximum supported datarate. For AM3 (Phenom II) this speed is up to DDR3-1333 but most AM3 mainboards allow you to overclock the memory controller and thus have speeds of DDR3-1600 and higher. For Sandy Bridge, Sandy Bridge E and Bulldozer, higher speeds are supported out of the box.

 

Integrated graphics

 

As you can see, with Sandy Bridge LGA1155 processors the integrated graphics is also on the processor die. For AMD chipsets with integrated graphics this GPU part is on the northbridge die. The 9xx series of AMD chipsets however does not have any integrated graphics and neither do the socket LGA2011 Sandy Bridge E processors.

 

Inside the cores

 

One of the main reasons why current Intel processors are faster than current AMD processors has to do with the way things are organized inside the cores themselves. As I explained earlier, basically all a CPU does all day long is fetch an instruction from memory (cache), carry out that instruction and then move on to the next. The reason Intel processors are currently faster is - simply but accurately put - the fact that inside the cores themselves the Intel processors are organized more efficiently. This allows the Intel core to process more instructions in a given amount of time than it's AMD counterpart while clocked at the same frequency.

 

Manufacturing process

 

Year after year the transistors (building blocks that chips consist of) get smaller. Current Sandy Bridge processors are manufactured on what's called a 32nm (nanometer) manufacturing process while AMD's Phenom II processors are still manufactured on a 45nm process (Bulldozer is also 32nm). The smaller the transistors on the die are, the less excess heat they generate and the cooler a processor will run. Besides running cooler the processor will also use less power. This is the primary reason that older processors run hotter than the current Sandy Bridge processors. I cringed just a few weeks ago when a reviewer was surprised that the "Hyper 212+ cools an i5 better than a Phenom II". This is a load of bullcrap. The Phenom II simply produces more heat!

 

With smaller transistors achievable clockspeeds also get higher. Smaller transistors is one of the reasons you can clock your i5 a little higher than your Phenom II. I'm purposely saying "a little" here because contrary to popular belief not nearly all i5 processors can reach 4.5GHz. In fact, quite a few of them hit a brick wall at around 4.2 - 4.3 GHz while the Phenom II hits this wall at 3.9 - 4.0. Only 100-300MHz difference in most cases. Remember though, that the i5 can process more instructions per second than the Phenom II can so despite the difference in maximum clockspeeds being small, the i5 still wins.

 

While we're on the subject of manufacturing: CPUs (and all other chips) are made using a process called "lithography". Basically this means that transistors are etched onto a clean, empty silicon surface in much the same way as photo's were developed before the age of digital cameras. This process is not an exact process, meaning that some processor dies come out better than others in the factory. This makes each and every processor die unique and is the reason why one i5 will produce more heat than another i5 with the exact same model number. It is also the reason that "stock voltage" is differerent for each and every single i5 (or any other processor). The stock voltage for a processor is entered into it at the factory. If your processor runs a few degrees hotter than your friends identical processor with identical coolers there is nothing wrong with yours, this is just luck of the draw. It's also the reason that some i5 processors can overclock higher than other "identical" i5 processors. There is nothing wrong with yours, just don't forget that even the unlocked i5 2500K is only guaranteed to operate successfully on stock clocks!

 

Overclocking possibilities

 

It was suggested that I add a little on locked and unlocked multipliers. Just be aware of this: if you want to overclock a Sandy Bridge socket 1155 processor you will need a version with what is called an "unlocked multiplier". The i5 2500K is one such unlocked processor. AMD processors can also be overclocked fairly well if they have a locked multiplier, although this is somewhat more involved. Most high-end AMD processors come only in the unlocked variety however so there's really no reason to buy a locked AMD processor to save money. More information on multipliers can be found in the overclocking guide in the sidebar.

 

Sandy Bridge E (LGA 2011) systems have some headroom in the base clock department that their little brothers on socket 1155 don't have. This is likely a result of the Sandy Bridge E systems not having onboard graphics as well as having an improved PCI express controller on the processor die.

 

It's also worth mentioning that even the "locked" Intel processors do have limited overclocking potential. One such processor is the Core i5 2500 which can be overclocked to 4.1GHz using the CPU multiplier. Another example would be the up and coming Core i7 3820 (Sandy Bridge E) that allows for a multiplier overclock of 4.3GHz and can be clocked even higher than this if you also adjust the base clock frequency. DISCLAIMER: always check the specifications of components you want to buy with the data you find on the manufacturer's website (in this case Intel.com) as mistakes can be made both on this site and a lot of other websites, including those of resellers!

 

Memory

Now that we've touched on the subject of the CPU, let's move on to memory. It's important to have read the part on CPU cache because that's a major influence that can conceal the performance of your system memory and render memory speed inconsequential for a lot of tasks!

 

Memory Voltage

 

Before continuing on the more involved characteristics of computer memory, let's address the memory operating voltage first. The current mainstream DDR3 memory can come at different standard operating voltages. Always consult your mainboard manual and the website of your CPU manufacturer to find out what memory voltage your CPU and mainboard are designed to work with! For all Intel Sandy Bridge processors the maximum voltage recommended by Intel is 1.5V. To be clear: 1.65V is not 1.5V. The answer to the question "will 1.65V work too?" is a resounding "maybe". If you are looking for certainty, look up the recommended specs on the CPU manufacturer's website and in your mainboard manual. Some people play fast and loose with memory voltages but this website is not the place to be looking for approval to go ahead with this.

 

DDR Memory

 

The current generation of memory that is used in PCs is DDR3 memory. DDR is an abbreviation of "Double Data Rate". The 3 in the name means that we're currently on the third generation of DDR memory.

 

Double Data Rate means that the speed at which DDR memory can transfer data is twice the clockspeed. In effect this means that DDR3-1600 runs on a clockspeed of 800MHz, DDR3-1333 runs on a clockspeed of 667MHz. Notice the absence of the MHz unit after 1600 and 1333. This is the correct notation and avoids confusing the datarate with the clockspeed.

 

Memory Timings

 

Aside from the clockspeed the memory is running on there is another major factor that determines memory performance: timings. You might have read specifications like "DDR3 1600 CL9" or "DDR3 1600 9-9-9-25" or something similar. The latter part of the memory specification are the timings. The most important memory timing is what is called the "CAS Latency". CAS is shorthand for "Column Access Strobe". We'll get back later to what this means exactly. The CAS Latency is the first number in a row of timings, so given the example, "9-9-9-25" would mean a CAS Latency of 9. The number behind the CL notation also specifies the CAS Latency so for the example "CL9" means a CAS Latency of 9.

 

Timings and Clockspeed

 

For memory timings there is a simple rule but with a caveat: the lower the timing value the faster the memory reacts to a read or write request from the processor or other components in your system. However memory timings are not measured in milliseconds or nanoseconds, memory timings are measured in clockcycles!

 

As a consequence memory timings can only be directly compared if the clockspeed of both DIMMS is the same. A value of 9 for CAS Latency means that it takes 9 clockcycles before data can be read from a DIMM module from the time this data was requested until such time that the data is ready to be read. This is where clockspeed comes into play. In order to compare the memory reaction speed of memory that runs on different clockspeeds we need to know how long a clockcycle takes.

 

Let's consider 2 examples:

 

1. DDR3 1600 CL9

 

As stated earlier the clockspeed of DDR3 1600 is 800MHz. 800MHz means 800 Million clockcycles per second. 1 clockcycles thus takes 1 / 800,000,000 seconds = 0.00000000125 seconds. Not a very long wait ;-) To make this number somewhat easier to read we'll convert it to nanoseconds (1 nanosecond is 1 billionth of a second). 0.00000000125 seconds x 1 billion = 1.25 ns (nanoseconds).

 

Now that we know how long a clockcycle takes we can calculate the real CAS Latency of our RAM in nanoseconds: CL9 in our case is 9 x 1.25 = 11.25 nanoseconds!

 

2. DDR3 1333 CL7

 

Now that we have calculated the CAS Latency for our DDR3 1600 CL9 we get to why CAS Latency can be so deceptive. DDR3 1333 runs at a clockspeed of 1333 / 2 = 666.5MHz. One clockcycle for this memory thus takes 1 / 666,500,000 seconds = 1.5 nanoseconds. A CAS Latency of 7 clockcycles for this RAM thus means a real CAS Latency of 7 x 1.5 = 10.5 ns!

 

So which RAM is fastest? You might be tempted to think that DDR3 1600 is faster than DDR3 1333 but this is not always the case! Every time the memory receives a read request for a specific memory location it takes the amount of time specified by the CAS Latency before actual data begins streaming from the DIMM module. In our example the 1333 memory does this significantly faster! Once a specific memory location is opened and data starts streaming, the datarate is what determines how fast all subsequent data from that location is sent. This obviously goes faster with the 1600 memory as opposed to the 1333 memory!