This needs a lotta formatting, and a lot of updating. Perhaps it should be split into several pages? Any and all help welcomed, this was written 3
years ago. — Cynos 04:47pm Fri 17 Nov 2006
Ok I’m gonna try and update this to include new platforms, dual core cpus, PCIe vid cards, DDR2 ram etc etc - pablo 09:20pm Tue 21 Nov 2006
It turns out the CPU HARD DRIVE articles are from this guide. I updated the info in them slightly so you might want to check them out — budget 01:16pm Wed 22 Nov 2006
I was not happy with my “So your building a new PC?” thread. In my eyes, it was not detailed enough about hardware and brands, and was generally all to brief. So I have taken the liberty (and several hours :p) to create a guide which is much more detailed about what all those stats and bell and whistles mean, and other commonly misunderstood areas of computer technology (such as hard drive cache, and its effect on performance as an example). I am not going to be talking Apples, as they are generally too expensive and incompatible with PC games to include in a guide for PCs which is focused on gaming. My main aim of this guide: to put all that jargon and tech-talk into a format which just about everyone can understand, and to dispel some of that marketing crap talk :)
Get yourself nice and comfy, you’re about to learn a lot more about computers (unless you’re already up to speed :p ). For ease-of-use’s sake, I’m using the analogy of a human body to explain what each component of the system does.
The “brain” of your computer, the CPU is responsible for three main things: calculating, writing/reading to memory, and decision making based on instruction sets and thus moving from one instruction set to another based on these decisions. I’m not going into the technical part of CPUs, as they are extremely complex and would be useless information you don’t need to know. Just to say that the CPU is what brings all the other parts of the system together to create the system with tangible results :)
The FSB of a CPU is one part of a two-part step for calculating a CPUs clock frequency (speed). The other is the multiplier, and you get the clock speed simply by multiplying the FSB by the multiplier. So, on my XP1800+ which has a FSB of 133Mhz, and a multiplier of 11.5, the clock speed is 133 x 11.5 = 1533 MHz. Now, this is just a core value.
On AthlonXP processors, the FSB is double-pumped, so my XP1800+ has 133 dual-pumped to to give effectivly a 266FSB. The FSB found on the Athlons ideally suits DDR RAM. Athlon64 CPUs work in a different manner, see “HyperTransport” below.
AMD formerly produced/sold 133Mhz, 166Mhz and 200Mhz FSB AthlonXP chips (effective 266, 333 and 400 FSB respectively).
Intel’s latest Celeron, Celeron D, Pentium 4, Pentium D and Core 2 processors are quad-pumped. So, I also own a Pentium 4 1.8 GHz. This has a stock FSB of 100Mhz, but is quad-pumped to give an effective FSB of 400. This FSB is best matched with Rambus RAM or Dual Channel DDR RAM.
Intel currently produces 133Mhz, 200Mhz and 266Mhz FSB CPUs (effective 533, 800 and 1066 FSB respectively).
HyperTransport is a technology found on AMD Socket 939 and 940 platforms which replaces the traditional “Front Side Bus” but can be treated in a similar manner when determining clock speeds. Default clock speed for the HT Link is currently 2000Mhz using Athlon64 X2 processors but most older chips on the two platforms operate on an HT Link speed of 1600Mhz. To arrive at this speed, first we take what is now referred to as the “CPU Frequency” of Athlon64 CPUs, which by default is 200Mhz across the board. On the latest X2 processors this has a 5x HT multiplier applied to it to give a real speed of 1000Mhz, and then double-pumped to 2000Mhz, whereas older chips have a 4x multiplier applied to give 800Mhz/1600Mhz real/effective speeds. In terms of performance there are no real-world gains to be made from having a faster HT Link as this bus is nowhere near saturated by today’s software, but this could obviously change in the future. To obtain the overall clock speed of a Athlon64 chip, take the “CPU Frequency” and times it by the CPU’s multiplier, for example an Athlon64 X2 3600+ has a multiplier of ten, so 10x200Mhz results in a default clock speed of 2000Mhz.
Three major manufacturers: AMD, Intel and VIA (in alphabetical order thank you :p )
Athlon XP/XP1700+ - XP3200+ (Palomino, Thoroughbred and Barton cores, 32-bit, 266/333/400 FSB, 462 pins), Athlon MP/MP1800+ - MP2600+ (same as XP range, plus ability to run in multi-CPU motherboards, 266/333 FSB), Opteron/240 - 244, 140-144 and 840-844, (64-bit, server based, iSSE2, 940 pins, on-die dual channel memory controller),Athlon 64FX/ 51 (64-bit, iSSE2, 940 pins, on-die dual channel memory controller), Athlon 64 XP/3200+ (64-bit, iSSE2, 754 pins, on-die memory controller)
Upcoming CPUs: Hammer/clock speed TBA (939 pin replacing the current 940pin FX)
Pentium 4/1.7GHz – 3.2GHz (Northwood core, 32-bit, iSSE2 included,512kb L2 cache, 533 FSB (“B”) and 800 FSB (“C”), Hyperthreading (3.06Ghz and all “C”) 478 pin), Celeron/1.7GHz – 2.6GHz (400 FSB, 128kb L2 cache, iSSE2 included, 478 pins), Xeon/2.0GHz – 3.06GHz (400 FSB, 533 FSB, 256/512kb L2 cache, server based CPUs, Hyperthreading, iSSE2, 32-bit),
Upcoming CPUs:
Prescott core/clock speed 3.2Ghz (800 FSB, Prescott New Instructions, Enhanced Hyperthreading)
VIA: Used to be a player in the CPU market, but has not released a new CPU since the C3 series. Lagging behind AMD and Intel in terms of technology.
Current CPUs:
C3/700 MHz – 1.0GHz (super silent, requiring a heatsink but no fan to operate, 64kb L2 cache)
Upcoming CPUs:
unknown
RAM, if you didn’t know, stands for Random Access Memory, and to use the human analogy, its like your spine - the basis of communication between the brain (CPU) and the rest of the body. It is the memory that a CPU can read and write to at a speed many times higher than that of the hard drive. Generally, the more memory you have, the better you computer will run. Your OS will specify a thing called a swap-file. When your system runs out of memory, your OS will use this swap-file as a type of memory on your hard drive. The only problem with this is that your hard drive is many times slower than your RAM, and thus performance will suffer. With more RAM, your system will use the swap-file less, and you will see a performance increase. You can see this increase in loading a game of Battlefield 1942, comparing 256mb to 512mb of RAM. The system will use all your RAM up and continue onto the swap file, but as you increase your RAM size, it will use the swap-file less, and thus load a lot quicker.
Also, more memory allows more applications and processes to be running before your system runs out of memory, and stalls or crashes.
Speed/Type: Nowadays, there are virtually 2 exclusive types of RAM, with a very small 3rd type. The first two are: SD and DDR. SD is older and technically obsolete, running at a much lower speed. The key point of the RAM speed and type is to provide as much memory-bandwidth as possible to the CPU. Bandwidth is like a road. The higher the bandwidth, the more lanes on the road, and thus the more traffic the road can hold before becoming “saturated”. PC133 SD RAM provides 1.08GB/s of memory bandwidth. This was okay when CPU architecture could only handle that much (i.e. up to the end of the Pentium 3 range of CPUs, who just had 133 effective FSBs). But, with Athlon XP and Pentium 4 CPUs, this was hardly enough. Thats where DDR RAM comes into play. DDR stands for Double Data Rate, and runs and the speed is effectively doubled. First modules of DDR were suited to dual-pumped 100 FSB (200 FSB effective) Durons, and was called PC1600 because it provided 1.6GB/s of bandwidth. Now, with minimum FSBs of 133 (266 effective), 266 MHz DDR is the minimum you can buy, and its called PC2100 because, you guessed it, it provides 2.1GB/s bandwidth. DDR has continued on this path, right through to the latest PC4000 modules, at an effective speed of 500 MHz :O The third and rarely used type of RAM is Rambus. It was the pinnacle of RAM right through to when dual-channel DDR became available. It runs at insane speed and provides huge bandwidth, and is run in pairs. Its ratings refer to the speed - PC800 meaning 800 MHz. The highest commonly available RD RAM is PC1066 at 1066 MHz. RD RAM’s downfall is price - it is hugely more expensive than DDR to buy. The focus of the Pentium 4 architecture was RD RAM’s huge bandwidth, and a main reason why P4’s have such big FSB’s compared to Athlons.
Intel: E7205 Granite Bay, 865G, 865PE, and 875P
AMD : nVidia nForce2
support a new DDR feature called dual-channel. Dual-channel is where two sticks use two memory controllers together, to create a low-latency 128-bit memory interface out of two 64-bit controllers. Example: one controller can be gathering info, the other publishing it. They are independent of each other, but work towards the same end, and halve latency because of this. The memory bandwidth is also effectively doubled (think of the difference between a 64-bit and a 128-bit interface on a graphics card), provided the FSB can keep up with it. Dual-channel works best on Pentium 4 systems, as the FSB can allow this doubling of bandwidth, whereas Athlons only benefit between 5 and 10 percent. The other downfall of dual-channel is that for it to work best, you need matched pairs. Thus, same speed, same brand and almost same batch are required to get the most out of dual-DDR. Dual-channel can best be described by the road analogy : the road now has the same number of lanes coming the other way, effectively doubling the traffic through-put.
Overall, aggressive (lower) timings are better, obviously, but you must find a balance between timings and speed. The difference between CAS2 and CAS2.5 in real performance is often very small, a couple of percent at most. Nice timings are the icing on the cake as far as I’m concerned. As for changing them - they’re usually in your BIOS somewhere, just dont do something stupid with them, okay? Otherwise you’ll either have a very unstable system, or your system won’t boot at all, and you’ll have to clear your CMOS.
Major RAM manufacturers:
Top brands (this is subjective, and from my experience): Corsair > Mushkin > Crucial > GeIL > Winbond > OCZ > Kingston
Corsair, undoubtably the producer of the best RAM in the world, has outstanding RAM that will be guaranteed to run at extreme timings. Mushkin, relative new player to New Zealand’s market, is quite a bit cheaper than Corsair, and offers a range of top-quality RAM. GeIL has had some problems of late with quality control, but their latest “Golden Dragon” series of RAM seems to be top quality, and able to run in dual-channel modes at extreme speeds and timings. Winbond, a manufacturer and supplier of RAM chips to Corsair, Mushkin and GeIL also produce a budget type of RAM, thats not quite as fast as the other manufacturers, but is damn cheap to boot (we’re talking hundreds of dollars cheaper). OCZ have had a hard time lately, with a few bad modules tarnishing their reputation, but after restructuring are looking at the high-end market again with a range of dual-channel enhanced PC3200 and PC3500 RAM. Kingston have their famous HyperX RAM to their name, specializing in high-speed RAM to run in dual-channel, but they sometimes miss the mark.
The storage place for all types of data, from your Operating System to video and sound files to games. An often overlooked component, the quality and speed can have a big impact on the performance of the entire machine, especially in terms of load times and failure rates. Old systems had hard drives of a few hundred megabytes (mb), but nowadays 80gb (gigabyte = 1024mb) and larger hard drives are commonplace. In this FAQ I will deal only with ATA and sATA drives, as SCSI drives are rare in desktop machines, and are a completely different technology altogether.
Major factors of Hard Drives:
Speed: This is usually the rotation speed within the hard drive. Older drives were 3,600 rpm, then moved onto 5,400 rpm and now 7,200rpm are the standard hard drives for consumers. The faster the hard drives, the lower the seek time.
Seek time: The delay between when a CPU requests data from the hard drive, and when the first byte of data is sent to the CPU. Measured in milliseconds, it is a popular measure of the performance of a hard drive, as the seek time is pretty uniform for a hard drive and is measurable. The lower the seek time, the faster the hard drive is. Typical hard drives that are sold today typically have a seek time of 9 - 12 milliseconds.
Interface: There are two common interfaces around now. These are Parallel ATA (PATA) and Serial ATA (SATA), with SATA being the newer technology, which will improve in theoretical speed from 150mb/s now to 600mb/s in 2007 (according to Seagate). ATA stands for Advanced Technology Architecture, and PATA being used since the early 1980’s are the industry standard in desktop hard drives. This technology has now aged and is about the oldest piece of technology you’ll find in a computer system today. PATA has allowed expandibility like no other computer technology, but has reached the end of its usefulness as it reaches its limits. SATA is the next step on for hard drive technology, it has much smaller cables, allows “hot-swapping” (taking out/adding in drives while the computer is on), and is also much faster. 72mb/s is the current maximum for ATA100 drives, but this will be dramatically increased as IDE technology improves over the next couple of years.
Latest chipset motherboards usually have SATA build into them, while with older motherboards you can buy SATA controllers that usually plug into a PCI slot. Gone are the bulky PATA cables too!
Cache: Like a CPU, hard drives have caches. Most drives today have either 2mb or 8mb caches, where recent data is stored, ready for use again if needed. This makes drives quite a bit faster, but makes them a bit more expensive accordingly.
Size: Everyone should know about this. Measured in Gb today, these can range anywhere from 20gb to 300gb. Common values are 80gb and 120gb now, and can be bought cheaply to.
Cooling?: As hard drives get faster, they also get hotter (similar to CPUs). Having airflow over a hard drive is essential for improving a hard drive’s life expectancy. Many enthusiasts now include front panel fans as essential components of their systems, and with good reason. Hard drives can reach 40 degrees plus! That heat has to go somewhere, and its best if doesn’t go stay in the hard drive itself, but is convected out into the case, and then out of the case.
Manufacturers/Brands Major players: Western Digital, Seagate, Maxtor, IBM - there are other smaller manufacturers, but these are the major ones.
Characteristics of each manufacturer:
Western Digital: Arguably produces the fastest hard drives, but they are often hot and noisy as well. Western Digital also offers 3 year warranties on just about all of their drives, which make them very popular, as hard drives are the most common pieces of hardware that fails in modern systems. They aren’t the cheapest drives around, but they make up for this in improved performance and quality. They have also recently released the Raptor series of SATA drives, that spin at a wicked 10,000rpm and compete with many SCSI in terms of performance. However, this drives are currently only available in 36gb sizes, so are limited to enthusiasts only.
Seagate: One of the oldest manufacturers of hard drives, Seagate is known for its quiet, cool drives that are cheap. Their Barracuda series of drives are highly popular for their big sizes and small prices. Their performance is not up their with Western Digital, but they are good drives overall. Only come with 1 year warranties though, except their 8mb cache drives, which come with 3 year warranties. They were the first to release a desktop SATA solution as well. Seagate are also known for their high-speed Cheetah series of SCSI drives that spin at an insane 15,000rpm.
Maxtor: Maxtor have had problems in the past with their Quantom Fireball and Bigfoot drives, but they’re latest drives are pretty much top-quality. They also offer the only ATA133 drives, which are totally pointless in terms of actual performance. They are usually priced between Seagate and Western Digital. They also produce the “huge” hard drives, from 160gb to 300gb in size. 1 year warranties.
IBM: Tainted with a bad batch of Deskstar hard drives that have haunted them ever since, IBM also produce top-notch drives for desktops and laptops. Typically a bit more expensive than Western Digital, they are quiet drives, and quite good performers too. The quality control of IBM’s drives have improved drastically since the bad-batch of drives, so don’t hesitate to buy one if the price is right.
The thing that shows the output of the rest of the system, the monitor is a very important component of any system, and should not be overlooked when considering upgrading/buying new. They also are a subject of much jargon and subjectivity, so I will try to remain as objective as I can. But the monitor is often a situation of user-preference that wins out in the end.
Main factors of a Monitor:
CRT/LCD: Cathode Ray Tube (CRT) monitors are the cheaper of the two, and are a similar sort of technology found in CRT TVs. Liquid Crystal Displays (LCD) are much more expensive, but are coming down in price all the time. The difference? Cathode Ray Tubes are technically obsolete, are much bulkier than LCDs because of the large tubes that are required in them and the choice of many gamers because of the no ghosting present on older and cheaper LCDs, and the price factor. LCDs have no refresh rate as such like on a CRT, and are much easier on your eyes, especially over long periods of time. LCDs are very thin, and much more portable than CRTs, but both have their limitations.
Benefits of CRTs: They are a lot cheaper than LCDs. A good quality 19” CRT will set you back between $600 and $800, whereas expect to pay at least double that for a good quality 18” LCD (you’ll see why I have a different size for both shortly. CRTs also have no problems with ghosting like LCDs do, which is where the edges of an object are blurred and appear to have a “ghost” double behind them. This is a big problem on LCDs with fast-moving objects, typically games and some movies. CRTs also are more flexible in terms of resolution, running any resolution you choose (up to the limit obviously), compared to an LCD which usually has a “native” resolution that the LCD looks best at.
Benefits of LCDs: First off, they dont have a refresh rate as such. Refresh rates are how many times a second the screen refreshes, or redraws. The human eye picks up this refresh, but does not show it to us in our vision unless the refresh rate is really low. This refreshing of the screen can make our eyes sore and tired after a while on a CRT, whereas LCDs dont have this problem. LCDs are also highly portable because of their thin screens, and take up much less desk real estate.
Size: The most important statistic to just about everything, its the physical size of the display, measured diagonally from corner to corner. This ranges from old 14” up to 22” screens. There is also another difference between LCDs and CRTs here too. CRTs have a slightly lower visible area than what they say, because the plastic surround covers some of the tube. For instance, my 19” CRT has a viewable area of only 18”. LCD displays don’t have this problem, and the size that is on the box IS the viewable size. Hence why i compared a 19” CRT to a 18” LCD, as both have 18” viewable areas.
Vertical Refresh on CRTs: The key statistic for choosing a monitor or a resolution to run at on a given monitor is the vertical refresh rate. Measured in hertz (Hz), the number of Hz is the number of times the screen refreshes per second. For example, my AOC 19” does 1280×1024 resolution at 85Hz. This essentially means that my monitor refreshes 85 times per second at 1280×1024. Now, this may be meaningless to you, but at a refresh rate of 60Hz, you will get very sore eyes in a very short time. Given this info, you are aiming to get a balance between resolution and refresh rate. I could run my monitor at 1600×1200, but then i would only have a refresh rate of 72 Hz, which is getting a bit borderline for how much i use my computer. I could also run it at 1024×768 at 100 Hz, but i feel that 85 Hz is perfectly fine for what i need. Higher quality CRTs offer higher refresh rates at higher resolutions, for instance a Philips 109P 19” flat screen does 1600×1200 at 85 Hz, but costs double what my AOC cost. Get it?
Response time on LCDs: This is the time it takes for a pixel to respond to a command to change colour. Measured in milliseconds (ms), the most common one around today is 25ms. High-end LCDs have since moved to 16ms response time, which dramatically reduces ghosting so much that it becomes barely noticeable. This seems to be a good solution for gamers, and many gamers now swear by their LCDs being just as good as CRTs at 3D gaming. Of course, these 16ms displays are expensive, but well worth it. The higher the response time, the worse the ghosting will be.
Outputs?: LCD’s typically use the DVI input on your graphics cards, but can use the old analogue VGA port as well, usually with an adaptor. CRTs all use the VGA port.
Manufacturers/Brands
Main players: Sony, Philips, Samsung, Viewsonic, AOC, Mag, Proview
Characteristics of each manufacturer:
Sony: The highest quality and clearest displays are made by giant electronics manufacturer Sony. There is no doubt about the performance and quality of Sony’s monitors - they are the pinnacle of monitors, and with good reason. With their Trinitron displays which are perfect flatscreens, they offer the best image quality of any monitor around. But, this comes at a cost - they are very expensive - you’re looking at about $1000 for a 19” Sony monitor, a good $200 - $300 more than any other brand.
Philips: Long renowned for their displays, Philips produce high-quality displays that produce insane resolutions - their 109P 19” can run up to 1920×1440. They are not quite as expensive as Sony monitors, but are not quite up to the 2D quality either. Their low-mid range of monitors are very competitive too, with a good selection of 17” monitors to complement their 19” and LCD displays. Good quality monitors overall.
Samsung: The name for LCD displays. Samsung’s CRTs are quite good too, but have an anomaly of having low refresh rates for their monitors. For example, the 955DF 19” flatscreen, which retails at around $550, has a refresh rate of 75 Hz @ 1280×1024, and 68 Hz @ 1600×1200. This makes it that you virtually have to use 1024×768 to have any chance for your eyes to be of any use after a couple of hours gaming or surfing the web. Still, if its LCD you want, the first direction you should look in is Samsung.
Viewsonic: Solid stable monitors. Viewsonic are often the first choice for institutions such as schools and universities, as Viewsonics are usually built tough, are cheap, and command a range of CRTs ranging from 15” CRTs to 20” LCDs. A good choice if you don’t want an AOC or AOC don’t have exactly what you want.
AOC: Cheap. If you’re looking for a good monitor at a reasonable price, AOC are where its at. Who can argue with a 19” flatscreen CRT for $350? Hey, the image quality isn’t up there with Sony, and the refresh rates aren’t as high as Philips, but you’re comparing apples with oranges if you do that. AOC’s boxes say “I Need Value”, and thats what AOC deliver.
Mag: A manufacturer that I would steer away from. Their 19” monitors are absolute lemons, with their brightnesses needed to be set to max to get any definition out of them. They are very cheap, around the same price as AOC, but just don’t have the quality to be really any good.
Proview: Much the same as Viewsonic monitors, except on a cheaper scale. :mad:
The veins and nerves of your computer system, the motherboard is the most important part of any system, bar none. Surprisingly, motherboards also tend to be the cheapest part of the system, and a place where many people either don’t care, or choose to upgrade something else (like a bigger monitor, faster CPU or better graphics card) at the expense of the humble motherboard. I have covered most of the technology thats included in a motherboard today (from CPUs to ATA/SATA to AGP), so I’ll just make a quick note of north/southbridge functions and a list of currently available chipsets. :p
Those chipsets that get noted twice are available in both SD RAM and DDR variants, depending on motherboard manufacturer.
Intel (Socket 478 only, I will not include server-only chipsets)
Intel SD RAM: 845GL, 845, 845G, 845GV
DDR RAM: 845GL, 845, 845G, 845E, 845GV, 845PE, 865P (Springdale-P), 865PE (Springdale-PE), 865G (Springdale-G), 875P (Canterwood), E7205 (Granite Bay)
RD RAM: 850, 850E
SiS SD RAM: 645, 648, 650GL,
DDR RAM: 645, 645DX, 648, 650GX, 651, 655, 648FX,
VIA DDR RAM: P4X266A, P4X400
AMD (Socket A only)
AMD SD RAM: 750
DDR RAM: 760
VIA SD RAM: KLE133
DDR RAM: KT266A, KT333, KT400, KT400A, KT600
nVidia DDR RAM: nForce, nForce2
SiS SD RAM: 740
DDR RAM: 740, 745, 748
Chipset Brands/Manufacturers
Intel: Intel’s own designed chipsets are the most popular chipsets on the market, incorporating all the latest technology that Intel’s Pentium 4 (and although not included, Xeon and Itanium) CPUs bring to the market. They are reknowned for their stability and reliability. Although not always the fastest, many people swear by Intel’s chipsets because of their legendary reliability. Intel also caters to a wide range of people through both SD and DDR boards, and by being the only producer of a RD RAM chipset.
SiS: Intel’s main competition in Socket 478 chipsets, well known for producing cheap, fast chipsets. They follow quickly on Intel’s latest technology as well, releasing new chipsets to cater. They are known for their cheap cost and high performance (similar to AMD’s situation in CPUs), and also cater for SD and DDR RAM. No RD RAM chipset though. Their chipset numbering is quite strange though ;p As for AMD chipsets, they produce few, and aren’t big players in Socket A chipsets (much like VIA is to Socket 478 ).
VIA: Not a big player at all in the Socket 478 scene, VIA has released just 2 chipsets, both for DDR. Small players. Completely different for Socket A though, VIA were the king of the hill from KT266A right through until the release of nForce2 recently. Most of the boards produced between the release of VIA’s mainstay chipsets, KT266A and KT333, and nForce2, were VIA boards. VIA also released KT400 (essentially a revision of KT333 to take DDR400 RAM), KT400A (added AGP8x and SATA), and has just released KT600 as competition for nForce2.
nVidia: Most known for their graphics cards, nVidia have with just 2 chipsets had a massive impact on Socket A’s chipset market. nVidia produced much innovation with both nForce and nForce2, with nForce2 being the first and only chipset (until SiS’s 748 chipset) to run dual-channel DDR. Extreme performance has been the name of nVidia’s game in producing Socket A chipsets.
AMD : AMD have released few chipsets, relying on 3rd party manufacturers to power their CPUs. AMD is more focused on producing server and workstation chipsets, designed to run dual-CPU and 64-bit computing to corporate partners.
Northbridge/Southbridge Functions
Northbridge: This is the central nervous system of a motherboard. The Northbridge is where the RAM and AGP slot communicate directly with the CPU, and where the Southbridge functions connect with the CPU as well. If you’ve read the section on RAM, the memory bandwidth is the road, but the onramp onto the road is the Northbridge, with the final offramp being the CPU itself. As you can see, this is a very important part of the system, and the chipset names above are actually the names of the Northbridge, as the Southbridge doesn’t change much. The Northbridge has a direct link with the AGP Bus, and the System RAM, and contains the memory controller which controls all the action. Its placement in the centre of the motherboard is no coincedence either - the shorter the “tracks” (the linking lines you can see on the motherboard), the less latency/interference with the data travelling to and from the northbridge. If you look now, you’ll see that the main communicators with the Northbridge (the CPU, AGP slot and RAM) are very close by. Thats all for the geography lesson I think :p
Southbridge: This is where all the peripherals meet. The PCI bus, any audio controllers, onboard LAN, USB, Firewire, RAID, ATA and SATA all connect to the Southbridge. The BIOS ROM also connects to the Southbridge. Hence why the Southbridge is placed next to all these components on the motherboard itself. The architecture of the chipset becomes important in the link between the Southbridge and the Northbridge - the better designed and the more efficient the link, the better it’ll perform.
Major players: Abit, Asus, Soltek, Epox, Gigabyte, MSI, Albatron, Leadtek, Iwill, Soyo, Intel
Manufacturers Characteristics
Abit: The overclockers boards of choice. Abit are well-known for their overclocker-friendly focus, offering all the controls enthusiasts demand of their motherboards. These are high-performance boards, and Abit come to the party with nice pricing and features too. Top-quality boards, many people swear by them. And did I mention overclocking?
Current top-boards:
AMD : NF7-S, NF7 and NF7-M (all nForce2), KV7 (KT600)
Intel Socket 478: IS7 and IS7-G (865PE), IC7 and IC7-G (875P)
Asus: Top-quality feature packed boards are Asus’ speciality, and they hardly fail to deliver. They deliver the stablest boards in the world, especially for Socket A. Asus are also known for legendary performance and out-of-box functionality with onboard everything. Asus are also regarded as having the best build-quality in the world. They are more expensive than other boards, but many believe that it’s worth the extra money.
Current top-boards:
AMD : A7N8X series (nForce2)
Intel Socket 478: P4P800 series (865PE), P4C800 series (875P)
Soltek: Smallish players until recently, Soltek have had problems with products in the past, namely the DRV5. The DRV2 really brought Soltek onto the map with nice performance and cheap, and Soltek’s latest 75FRN series of nForce2 boards have really brought them into the limelight, with their maximum-performance minimum-price and be damned all the extra whistles attitude. Their 75FRN boards may not have SATA or Firewire, but they are damn fast and offer good overclocking options, as well as being cheap.
Current top-boards:
AMD : SL-75FRN2 series (nForce2)
Epox: Epox’s 8RDA+ board has been the clincher in Epox’s Socket A boards - its fast, reliable, and has good options and features. It really sums Epox up - feature packed, overclocker friendly, fast, and reasonably cheap. Epox’s older boards also featured lots of ATA ports, for lots of hard drives, and the best KT333 performer in the form of the 8K5A3+. Epox is Socket A only - it does not produce any Intel boards.
Current top-boards:
AMD : 8RDA3+ (nForce2)
Gigabyte: Gigabyte are known for their well-built and sturdy boards, but having fewer overclocking options on them than say, Asus or Abit. This has made them less popular amongst enthusiasts who demand total control over voltages and timings. Gigabyte produce a huge variety of boards across many chipsets, making them popular amongst system builders who don’t necessarily want their customers to have access to overclocking controls. Gigabyte are also well-known producers of high quality graphics cards for ATi.
Current top-boards:
AMD : 7NNXP (nForce2 Ultra400)
Intel Socket 478: 8KNXP series (875P), 8IG, P, or K series (865 series chipsets), 8S648FXM series (SiS648FX)
MSI: Boards that are reknowned for stability and ease-of-use. MSI are large manufacturers of motherboards and graphics cards, and produce some fine equipment. It isn’t always the first board to market, but you can be sure that when you buy MSI, you buy tested stability. MSI’s latest Pentium 4 offering isnt the best overclocker, and reflects MSI’s overclocking potential - not great. Similar to Gigabyte really.
Current top-boards:
AMD : K7N2G-L (nForce2)
Intel Socket 478: 875P-Neo series (875P), 865PE-Neo series (865PE)
Albatron: New players to the New Zealand market, Albatron have been around a while in the US. They produce good cheap boards from what I can tell.
Current top-boards:
AMD : KM18G Pro (nForce2)
Intel Socket 478: PX865PE Pro II (865PE)
Leadtek: Producers of highest-quality video cards, Leadtek are the latest addition to motherboard manufacturers. They seem to be reasonably fast boards, but they have been a bit lacking, not quite up their with the top-tier manufacturers just yet. People have also had problems with the onboard GF4 on their nForce2 boards.
Current top-boards:
AMD : K7NCR18 series (nForce2)
Iwill: Ultimate overclocking boards. These boards offer all the options, and with the XP333-R the only board in the world to offer a 1/6 divider, aren’t afraid to test the waters with overclockers. The main downfall for Iwill are their reliability - they are known as boards with “quirks”. Perhaps Iwill would have done a lot better without ALi Magik’s chipsets on their AMD boards, as this chipset gives bad 3D performance until you tweak the registry a bit. Iwill boards usually use weird chipsets too.
Current top-boards:
AMD : K7S3-N (SiS748 )
Intel Socket 478: P4S series (865 series chipsets)
Soyo: Produce visually stunning boards, with high-performance options on their boards. Soyo came, and seemed to disappear again just as quick. Soyo have been producing motherboards for a long time (my 486 has a Soyo in it), and produce quick boards with lots of bells and whistles. Soyo’s Dragon series of boards have been very successful, especially overseas.
Current top-boards:
AMD : SY-KT400 (KT400)
Intel Socket 478: SY-P4I875P Dragon 2 (875P), SY-P4I865PE Dragon 2 (865PE)
Intel: Intel reference boards, built and designed by Intel, so they are guaranteed to be stable and reliable, and to have all the features that make the Pentium 4 as good as it is.
Current top-boards:
Intel Socket 478: D875PBZ (875P), D865 series (865 series chipset)
First off, a video\graphics card is a display adapter which is a dedicated piece of machinery for processing of 3D images. It has its own Central Processing Unit (called a GPU, or Graphics Processing Unit) and its own store of onboard memory. This memory is used for storing textures (i.e. Just about everything you SEE in a game/3D application). As most parts of a computer follow, this memory goes up in powers of 2, from early cards with 256kb/512kb of memory, up to today’s standards of 64mb/128mb/256mb.
Early on, these cards used ISA or PCI slots. As programs became more and more 3D intensive, the limitations of these busses were found to be insufficient for heavy users of 3D. Hence, Accelerated Graphics Port (AGP) was born. With a bus of 66MHz and memory bandwidth of a shade over 250mb/s, AGP was more than twice as fast as PCI, and it was a dedicated 3D graphics card slot. The number (2x, 4x and 8x currently) you see after the AGP on motherboards and graphics cards are simply extensions of the original AGP specs (ie 8x means 250mb/s x 8 = 2.0 GB/s). The AGP slot is now the standard slot used by virtually all graphics card produced.
PCI-Express is a new technology that should be appearing very shortly that will improve graphics technology quite substantially. Watch out for the coming R420/500 ATi and NV40 nVidia cards, which will be PCI-E native, with AGP adaption at first.
Here is a list of nVidia’s and ATi’s Radeon cards (as far back as my memory can reach ), starting with the latest offerings. (I do not include SiS cards as these were virtually all integrated solutions, or were negligible in their impact. 3Dfx cards are also not included as 3Dfx no longer exists, and not many people have 3Dfx cards in their PCs anymore)
nVidia(Code : Family : Models : Variants)
NV40: GeForce : 6800 : non-Ultra and Ultra
NV38: GeForce FX: 5950: Ultra
NV36: GeForce FX: 5700: non-Ultra and Ultra
NV35: GeForce FX: 5900 : non-Ultra, Ultra, XT and LX (Leadtek)
NV34: GeForce FX: 5200 : non-Ultra and Ultra
NV31: GeForce FX: 5600 : non-Ultra, Ultra and XT
NV30: GeForce FX : 5800 : non-Ultra and Ultra
NV28: GeForce4 Ti: 4200-8x, 4800SE, 4800: none
NV25: GeForce4 Ti : 4200, 4400, 4600: none
NV20: GeForce3: standard, Ti200, Ti500 : none
NV17: GeForce4 MX : 420, 440, 440-8x, 460 : SE
NV15: GeForce2 Ti : none : none
NV14: GeForce2 : Ultra, Pro, GTS : none
NV11: GeForce2 : MX, MX-100, MX-200, MX-400
NV10: GeForce: 256, DDR : none
NV5: Vanta/TNT2: LT, M64, M64 Pro, standard TNT2, Aladdin, Pro, Ultra : none
NV4: TNT : none : none
NV3: RIVA : 128, 128ZX : none
Brands to Buy: Leadtek, Visiontek, Gainward
Brands to Avoid: Powercolor, Creative, some of the Chaintechs, Compro
ATi (Code : Model : Variant)
R350: 9800, 9600 : Pro, non-Pro and XT
R300: 9700, 9500 : Pro, non-Pro
R250: 9000 : Pro, non-Pro
R200: 8500, 9100 : Pro, non-Pro, LE
R150: 7500 : DDR, SD
R100: 7000 : DDR, SD, VE
Brands to Buy: Gigabyte, Made by ATi
Brands to Avoid: Powercolor, HIS, HighTech
NOTE:All 9500/9600/9700/9800 series Radeon cards are made by Sapphire or by ATi itself, thus they have all the same components. The only difference between manufacturers is the box, and maybe some cosmetic differences like the heatsink. When buying look at RAM ratings especially - these are often the only differences between manufacturers, as well as the software bundle that comes with the card.
now, history lesson over onto the questions.
I can’t choose between nVidia or ATi, help me!
The best way to choose is to go to some review sites, and look at the cards that are available in your budget area. Look at any comparisons between cards based on performance, image quality and compatibility/driver support. Choose the card that best meets your criteria, regardless of who manufactures it. That said, if you are using a Linux-based system, nVidia is virtually your only choice.
What’s the difference between an XT, a Pro and a non-Pro 9800/9700/9600/9500/9100/9000?
Mainly its just clock speed. But, on some cards (like the 9500/9600 series) some features may be disabled on the non-Pro. Generally speaking, most non-Pro cards should get up to Pro speed anyway. An XT card is simply an overclocked Pro card. This applies to the 9600XT (500 MHz core vs. 400 MHz core) and the 9800XT (412/730 MHz core/memory vs. 380/680 MHz core/memory)
How do I overclock my Radeon?
Some require a BIOS hack, google it to find out what is best for you. I also just discovered a neat little tool called RadClocker which brings up a Clock Rate tab on your adapter properties. It’s available here
Will my AGP4x motherboard be able to run an AGP8x card?
Yes, but it will only run the card at AGP4x bandwidth.
What’s with these DirectX versions on the cards?
The higher the number, the later the version of DirectX. DirectX is a tool used by developers to take advantage of hardware special features, like Pixel Shaders and the like. In turn, graphics card manufacturers often like to “future-proof” their cards by making them compatible with the next version of DirectX out, incorporating the new features into the cards.
I have a slow CPU. Is it pointless to upgrade my graphics card, or should I save for a new CPU/motherboard/RAM combo?
In a nutshell, save. Having a fast graphics card is pointless if you have a slow CPU because your CPU will limit the graphics cards potential. Save up for a new system, by then you will be able to get the best card out at the time at the right price and be better off in the long run, with a properly balanced system.
Alright, i want to overclock my graphics card now. What programs/tools should I use?
I personally recommend PowerStrip and Rivatuner. Both these programs come with excellent tweaks for graphics cards as well.
What are Anti-aliasing and Anistrophic filtering?
Without going into too much technical detail, these two things make your games look better by smoothing out those jagged edges and blending colours better. They also require a lot of GPU processing power. These are generally only viable on later series cards (the GeForce FX5700 and above, and R300 and above ATi cards). They can be enabled in the Display Control Panel of either manufacturer.
Let’s talk about power.
Later series cards have higher power requirements than earlier cards - as cards get faster, they get more hungry for power. Before buying your video card, it’s a good idea to check any power requirements on the manufacturers website. That said though, there are no cut-and-dried guidelines in terms of power supply. Read some review sites for a better indication.
Those extensions (MMX, 3DNow!) are pretty much irrelevant nowadays as distinguishing factors in gaming, as AMD’s processors support all of the MMX/SSE/SSE2 (well, the A64 supports SSE2) extensions anyway. The main differences are architectural rather than these extensions.
Basically, the XP/A64 architecture is better suited to gaming. In the XP’s, the shorter execution pipeline (12-stage compared to 20-stage of the P4) made the CPU less reliant on cache. I’ll assume you don’t know what I mean by this, so I’ll explain this, as it is very important.
Basically, the onboard cache (L1, L2, L3 in order of speed - then it’s onto system memory) of the processors is where the common/recent instructions are “in memory” for use. The execution pipeline pulls these instructions from the cache when needed. However, if the instructions needed are not in cache, they must be pulled from system memory. Meanwhile, the entire pipeline stalls, is emptied, and has to be refilled. A 20-stage pipeline takes longer to refill than a 12-stage one, and also has a higher probability of stalling due to its length (i.e. if something enters the pipeline and causes it to stall, 19 other instructions are discarded for the P4 compared to only 11 for the XP). Thus, the need for greater amounts of cache to lessen the effect of the longer pipeline - driving up production costs and likelihood of failures (those with damaged cache are often sold on as Celerons with the damaged cache disabled).
Consider as a prime example the Centrino laptop CPUs. Due to their huge cache sizes (2MB L2 cache), they have low clock speeds (top speed is 1.8 GHz). However, put a 1.6 GHz Centrino next to a 2.8 GHz 800 FSB P4, and the Centrino will run circles around it. Also take note that the Centrinos are only 400 FSB (due to heat constraints), and you can see how important cache is when you talk about P4 performance. A lesser example is the P4 EE CPUs. Although they have 2MB of L3 cache rather than 2MB L2 cache of Centrinos, they still perform significantly better than their lesser Prescott and Northwood counterparts at the same clock speed.
The FSB increases over the last few generations have lessened the impact of this, as well as increases in cache (256kB for Willamette P4’s, 512kB for Northwood, 1MB for Prescott). However, for the Prescott, Intel made what I consider a serious mistake in that they lengthened the pipeline to 31-stages - thus, the increase in cache was made pretty much redundant by the lengthened pipeline. This is one reason why the Prescott is slower than a Northwood CPU of the same clock speed.
This is the major contributing factor to why early P4’s are hideously slow compared to their XP competitors - 256kB of cache is inadequate for the 20-stage pipeline. An example of why AMD’s are better in this department can be seen from the change from Thoroughbred to Barton CPUs - even though they doubled the L2 cache from 256 to 512kB, the performance increase was very small, much in constrast to the difference between 256kB and 512kB P4’s. Coupled with the XP’s increased calculations per clock, it amounts to a better suited architecture for gaming.
One further point of interest is the memory bandwidth. P4’s require lots of memory bandwidth to be able to function their capacity, something which early DDR struggled to do. Rambus was the solution to the problem - but it was and still is very expensive, and also has high latency. DDR was cheap and low latency, but didn’t provide nearly as much bandwidth as Rambus. However, with the advent of Dual-channel DDR, this problem has been much alleviated. The jumps in FSB from 400 to 533 to 800 are proof of this - basically, DDR266 (2.1 GB/s bandwidth) to DDR333 (2.7 GB/s) to DDR400 (3.2 GB/s) to dual-channel DDR400 (6.4 GB/s). Those are all theoretical totals, but they show how big the difference in bandwidth is. With the advent of dual-channel, it has made increases in FSB much less important (as shown by the lack of performance increase in the 1066 FSB P4 EE models).
This quad-pumping of the FSB stands in stark contrast to XP CPUs. Relying on double-pumped FSB designed specifically for use with DDR, the impact of memory bandwidth is much less. Whereas the quad-pumped architecture can soak up and benefit from increases in memory bandwidth like a sponge, double-pumped means that the CPU only utilizes memory bandwidth up to it’s FSB speed (i.e. a 266 FSB XP CPU only utilizes 2.1 GB/s bandwidth maximum, any extra bandwidth simply isn’t used). Hence why even dual-channel provides insignificant performance gains - the CPU simply cannot use all the bandwidth supplied by the RAM.
Athlon 64/FX CPUs are a different kettle of fish altogether. They don’t have an “FSB” as such (FSB is basically the link between the CPU and Northbridge of the motherboard (the Northbridge controls AGP and RAM, the Southbridge controls the PCI/PCIe buses, I/O, sound, networking), but rather HyperTransport between the CPU and motherboard chipset. I say “motherboard chipset” rather than “Northbridge” because the A64/FX CPUs have integrated memory controllers - thus there is only need for one bridge on the motherboard. What the integrated memory controller does for increasing performance is it eliminates the Northbridge link, reducing latency and overhead, and also providing an ultra-fast link to the main processing cores of the CPU. What does this mean for gaming? CPU-intensive games (such as UT2004, Doom 3, Half-Life 2 and Far Cry) will benefit from the integrated memory controller and the overall superior processing power of the A64/FX chips. The P4 chip has remained relatively untouched (apart from the cache, FSB and pipeline stage increases) since it’s release, and it’s now showing it’s age. It cannot keep up with what is a superior chip in almost every way that counts for gaming.
The reason why you may not see increases in performance is because you’re playing old games - DoD uses the HL engine (obviously), which is a modified Quake/Quake2 engine, and is over 6 years old already, designed to work on Pentium-class CPUs, which are many many times slower than today’s chips.