Mega Sticky and FAQ!!

[Strange1]

This Mega sticky will be undergoing some changes. Firstly, the FAQ linked below will slowly be moved into this thread. More changes will come as they're thought of. lol.

Firstly, welcome one and welcome all to the Newbie lounge of Overclockers Australia.

You may post in any forum on OCAU, but if your just starting to learn the ins and outs of computing then this is a great place to start asking questions. If your an experienced user and have a question your a little embarrassed to ask in a specific forum then feel free to ask here Its a no-flame zone. If someone is having a go at you feel free to report the post and we will sort it out ASAP.

Ok now down to business....

The search function is your friend. Remember many a newbie have been through these forums, there is a good chance your question has already been asked and solved. That means a quicker reply for you
You'll find that button at the top right of every single page in the forums.

If you're looking for advice on what computer hardware or software you should buy or where you can buy them from, then please use the What / Where Should I Buy forum.
Offending threads will be closed.

If you came here to be able to buy and/or sell goods, you must be a member for more than 90 days to gain access to the trading forums.
More info here.

Here are some threads covering the most frequently asked questions :

The FAQ. Feel free to add questions and answers to this as well

Computer temperature question/problem? Then this is the link for you!

A handy dandy list of Abbreviations and Terms

Got a 100Mhz or 133Mhz fsb instead of 200Mhz or 266Mhz fsb? or CPU running below rated speed? Then look here.

Lots of tutorials beginner guides at PCTechguide.com

How to post images on the forums

How to tell if it is a Palomino, Tbred, or Barton. Look in here to find out.

If you have a thread that you think is worthy to be linked here then feel free to PM either looktall or myself and we can check it out.

So bye for now, enjoy your stay.

Adam

[looktall]

just a comment on the no flaming rule.

you may find your post edited or deleted in other forums, but it may also be left alone. that's taken on a case by case basis for me personally. i can't speak for other admins.

in here it will be quickly and harshly dealt with.
ie. you'll cop a right spanking.

in here, for me personally, what normally wouldn't be considered a flame in any other forum (like "OMG liek search n00b!!!!1111" etc) will be treated the same as if you called their mother a nasty name.


i've removed the bulk of this thread as it was getting off topic.

[anlashok]

What does defrag mean?

Defrag: Defragmentation, to unfragment.

Your Hard Drive is made up of clusters ( think little boxes that hold info ) ( an average cluster holds 4kb ).

Everytime you write data to your Hard Drive it fills those boxes.

Now say you transfer 40x5Mb files from 1 Hard Drive to another you will have on you Hard Drive something that looks like this:

# = clusters

################################################## ################################################## ################################################## ##

Now say 1 week later you delete 12 of those 5Mb files.
It will now look like this:

_ = Newly created free space

########______##############____######## ______##########
#######################___________####________________
###########_________________ ##########################

2 Days later you transfer some more data across on to the Hard Drive 10x20MB files.
The Computer will write the data in the gaps and else where:

= = New Data

########====##############===#######=======
#################################==========
####==============###########================
########################## ================================================== ============================

This now leaves some of your files broken up arcoss the Hard Drive, so when the hard drive seeks ( looks for your data ) it will read part 1 of the broken file then must seek again to find the next bit. This slows down your performance as you are kept waiting.

After a Defrag the fragmented data is rearranged so all the data is kept in one piece.

################################################## ################################################## ### ================================================== ================================================== =================================

As for how often you should do it everybody has a different answer.

Yes Defraging is normally a good thing.

Basically if you use FAT32 you will need to Defrag more than if you use NTFS ( NTFS handles fragmentation better ).

If you shift alot of Data around often you should do a Defrag more often, people with Large Hard Drives generally don't like to Defrag too often as it takes quite sometime.

On a normal home pc that gets a fair bit of day to day use, I would recommend a Defrag every 3-6months about the same time you should give your PC a service ie check all cables are still tight clean out dust etc.

[looktall edit: fixed the colouring and formatting and also the "what is.." question at the top of the post.]

[Quadbox]

FROM THE FAQ STICKY:

As per this thread, I'm starting an FAQ.

Most definitely under construction. If an admin could sticky it, that would be great.

Please feel free to contribute.


How do I unlock my [type] cpu?

Slot A Athlons - See this guide here.

Socket A Athlon "Thunderbirds" (and pre-morgan Durons -

Socket A Athlon XP "Palominos" (and Duron "morgans") - There is a good guide here

Socket A Athlon XP "Thoroughbreds" - Check out the guide here

All Intel CPU's - Sorry, you're out of luck. They can't be unlocked.

(If someone could point me at some guides for the ones I don't have, that'd be great.)


I've installed linux in a dual-boot setup, and now I can't get rid of it. How do I boot into windows?

For Windows 95/98/ME: Get into a dos prompt, and type the following:
C:>fdisk /mbr

For Windows 2000/XP:

1. Boot from your Windows CD.
2. When asked, choose "Repair a Windows Installation"
3. Choose "Repair a Windows Installation Using the Recovery Console"
4. Log onto your windows installation, by following the prompts.
5. When you get to a dos-like prompt, type "fixmbr". Follow the prompts.
6. Reboot.

I want to try linux. Which distribution should I use?

It's a matter of opinion. Check out the Perpetual Linux/BSD Distribution thread here.


I've only been registered on the forums for less than 90 days, and I can't read/post in the Trading Forums. Why?

You can't access the trading forums until you've been a member for 90 days. See Agg's explanation of the current situation here.


What case should I buy?
What sort of power supply do I need?

See Wolfy's thread here.


What temperature should my cpu be?

AMD - There's a guide on overclockers.com here. Also see the thread here.

Intel - Intel chips don't really have a problem. If you're using an approved heatsink/fan, don't worry about it.

There's also a great thread here.


Where can I buy [foo] online in Australia?

There are many online shops in Australia. Check here for a good list (sorted alphabetically by state, no bias given towards or against any vendor listed. The list is provided as is, with no condonement by OCAU of any vendor featured).

I've overclocked my FSB. How do I know what speed my PCI bus is running at?

It depends on what the PCI Divider is set to. This varies from motherboard to motherboard. If you've got an FSB of X MHz, and a multiplier of Y, then your PCI bus is running at X * Y MHz. An 'in spec' PCI bus is one that is running at 33 MHz.

EDIT by Manaz: I replaced the vendor URLs with a link to a thread containing a long and detailed list.

[Ashendragon]

RMA - Return Material Authorisation - (Faulty Component Replacement)

Obviously most companies do things differently, this is just a general guide to follow if you need to return something and don't know where to start.

Before you return a faulty component, you must get authorisation from the manufacturer you are returning it to. So you need to request a Return Material Authorisation.
The process will vary, but generally this is done by going to the manufacturers webpage and filling in an RMA form (with some companies this cannot be done online, they will send a paper copy to you). To do this you just need to fill in your personal details and the serial number of the component you're returning. Once you submit this form a page will be displayed that contains your personal information and your issued RMA number. This needs to be printed out, signed\dated and attached to the package you're returning. All manufacturers do things slightly differently so follow their instructions. You're responsible for the costs of shipping etc. generally it will cost around $30.00 - $40.00. The time it will take for you to recieve the replacement product depends upon the manufacturer. It can take 1 week or up to 3 months, but more than likely it will be around 2-3 weeks.

Some tips

* Find out if the retailer you bought the component off can replace it for you before you request an RMA through the manufacturer. A retailer with good support will do it free of charge depending on how long ago you bought the part. Even if they do charge you, you will usually get a replacement sooner than if you went through the manufacturer and you won't have to worry about packing and sending it yourself. The manufacturer may also have a service centre in Australia that you can return it to, if you can't get the retailer to do it, ask them if they can put you onto an Australian service centre, it may be worth your while going through one of these.

* If you return something without a RMA number the package will be sent back to you unopened. Technically this will also happen if the RMA form you printed out is not viisble on the outside of the package.

* Pack your gear well, follow the manufacturers' instructions. You don't want to be denied a replacement for a legitmately faulty component because you skimped on the packing. There is no such thing as overkill!

* Test the component thoroughly before returning it, remember that you're the one paying for the shipping etc. If you don't have a working computer to test the component in, then ask the retailer you purchased it from to test it for you. It's the least they can do . If it's a HDD, run the manufacturers' test on it, if it turns up an error code make sure to include this when filling out the request RMA form. It is a lot easier to get a replacement when even their own software tells you the drive's faulty. Otherwise, where applicable, always include a detailed, but to the point, fault description.

[SLATYE]

Video Cards
Video cards are a very interesting area, because while they're not as highly regarded as CPUs, they're a huge amount more complex. I think the P4 core has something like 55 million transistors, and that's including 512kb cache (that should account for a bit over 3 million). The Radeon 9700, in comparison (a reasonably old card now) has (I think) about 110 million transistors. The NV40 from Nvidia is expected to have around 210 million. The core of a video card is generally referred to as the GPU.

There are a few main manufacturers here - Intel, ATI, and Nvidia. There's also Volari, who have tried to make a decent card with the V8 Duo (using two GPUs). However, with cheating disabled (they did a bit of that) it falls behind even the mid-range cards from ATI and Nvidia, and makes a huge amount of noise.

There are six basic 'ranges' for video cards:
Office work: This is dominated by Intel, with their Intel Extreme graphics. While pretty much useless for anything 3d, Intel Extreme graphics is fine for office work, and it comes integrated on many mainboards. It's also common in notebook computers.

Low-end gaming: This generally comprises the Geforce4 MX440SE cards through to the GeforceFX 5200s. It also includes the Radeon 9200, 9200SE and 9600SE. These are fine for older games, and often included in cheap pre-built systems. However, in modern games they can't cope.

Mid-range gaming: This generally includes the Radeon 9600XT and the GeforceFX 5700 Ultra. The Geforce4 Ti4200, the GeforceFX 5600/5600 Ultra/5700, and the Radeon 9600 non-Pro are the very bottom of this segment, and don't really count. The 'standard' cards here (Radeon 9600XT and GFFX5700 Ultra) are quite good cards. Both will be fine at most resolutions in modern games, and should do a decent job in new games (like HalfLife2 and Doom3). The Radeon 9600XT has stronger DirectX 9 performance, while the GFFX5700 Ultra is faster in current DX8 stuff. These cards should also be quite capable of using anti-aliasing in most games without a significant performance drop (however, the older GF4Ti cards took a big performance drop).

High-end gaming: As far as I'm concerned, this is the Radeon 9800 Pro, the GFFX5900XT and the GFFX5900. These are cards which were, at one point, pretty much as fast as you could get, but have been superceded. They're still very fast, in many cases only a few frames-per-second behind the top cards.

Very high-end gaming: This belongs to the Radeon 9800XT and the GeforceFX 5950 Ultra. Both cards should have little trouble in current games at high resolutions with all the detail options as high as they'll go. These cards are also the only current ones which actually can make use of 256Mb of RAM. However, for the slight speed increase over the older generation, you pay a big premium.

Workstation: This comprises cards which won't do very well in gaming at all - they're designed specifically for 2d image quality, and extreme precision. Matrox has always been well-known here. It is one place where the ill-fated Parhelia has been able to stay. Nvidia cards are also good. However, ATI's cards tend to be slightly modified versions of their desktop cards, so they don't do as well.

More info on video cards
SE cards:
Any card with 'SE' on the end of it's name will generally be slow. Most have either 64-bit RAM, or much slower RAM than usual. This results in absolutely terrible performance. Note that some of Gigabyte's cards (their Radeon 9600 and GF4MX ones) are actually 'SE' cards, despite not being listed as that.

Also be aware of 'XT' cards. In ATI cards, 'XT' is a good thing. XT cards are faster than the other types. However, in Nvidia cards, XTs are slower than the others (a GFFX 5900XT has slower RAM than a GFFX 5900).

Speed measurement:
The speed of a graphics card is measured in two ways:
(1) Frames per second (fps) - used to determine minimum and maimum framerate in games. Different people like it at different speeds. Personally, I like to make sure it stays over 40 fps at all times.

(2) 3dmarks - this resulted from Futuremark's excellent benchmark program 3dmark. This is used to compare cards of all types. It's a free download too. 3dmark 2001SE test DirectX 8 and DirectX 8.1 speed, while 3dmark 2003 also tests DirectX 9 speed. A while ago, Nvidia was found cheating on 3dmark 2003 tests, but I think that's pretty much fixed now.

Card design:
There are a few things which are very important to any video card:
(1) Pipelines - this is the number of things that a card can work on at any one time. I think low-end cards use 2, mid-range ones have 4, and high-end ones have 8. The NV40 is expected to have 16.

(2) Shaders - these are used to add effects to pixels. If a card has slow shaders, these will bottleneck the whole card. This is what happens to Nvidia cards in DX9 applications.

(3) RAM speed and width - this is very important, because video cards need a huge amount of RAM bandwidth. The top cards have over 30Gb/s available from the onboard RAM. High-end cards use 256-bit RAM, while cheaper ones use 128-bit RAM. 'SE' cards often have 64-bit RAM. The speed of the RAM is also important, as it changes the memory bandwidth too.

(4) DirectX versions - old cards only support DirectX 7 (GF4 MX cards have this). That means no pixel or vertex shaders. These are required for some games. Other games will run, but they won't be as pretty. DirectX 8 adds Pixel Shader and Vertex Shader support (GF3 and GF4Ti cards use this). DirectX 8.1 adds better Vertex Shader support, which is used on the Radeon 8500, 9000, 9100, and 9200. However, I don't think many games use these. New cards (Radeon 9500 and better, GFFX cads) support DirectX 9. This adds support for Pixel Shader 2.0 and Vertex Shader 2.0. Some new games use these.

Special cases
I might just add this here, for reference:

Volari V8 Duo - this seemed like a good idea. Use two GPUs, get twice the speed. However, it requires a huge and noisy heatsink, costs a lot (due to the expensive PCB and two GPUs), and needs twice as much RAM as any other card (because each GPU needs it's own RAM). In practice, it's slower than a Radeon 9600XT.

GeforceFX 5800 Ultra - this gets referred to quite a lot. It was meant to be 30% faster than the Radeon 9700 Pro, but it came out very late, and when it did appear, it sounded like a vacuum cleaner (really, really loud). The performance wasn't as good as a Radeon 9700Pro, especially in DirectX 9. Nvidia has since removed almost all references to this card from their website.

NV40 - Nvidia's new GPU. 16 pipelines, on a new manufacturing process. Heat output will be huge, but should be a very quick card.

Nvidia driver cheats - a while ago, Nvidia took to cheating in 3dmark, by stopping their cards from drawing the whole image. I think they've fixed this one now. ATI previously cheated in Quake 3.

Nvidia vs ATI in DX9 - when Microsoft was drawing up the DX9 specifications, they decided to follow what ATI said. This included using 24-bit colour. Nvidia decided to create their own standard, using a combination of 16-bit and 32-bit rendering. Nvidia failed, and their cards don't do well in DX9 because they have to render everything in 32-bit. Prformance is much better in 16-bit, which is why their drivers now select certain things which will be rendered in 16-bit, and other things to be rendered in 32-bit. ATI's cards just run 24-bit all the time, and have no performance problems at all.




RAM
RAM is where the CPU stores all it's information that it is working on, but not right now. For example, if you run Windows on it, much of Windows will get loaded into the RAM, and the CPU can work on that information. The video card also takes a bit of RAM for itself, to store textures in.

Currently DDR400 SDRAM is about the 'standard' RAM. SDRAM refers to it being Synchronous Dynamic Random Access Memory. You can safely ignore that - the term SDRAM is normally used to refer to 'standard' SDR SDRAM (the old type, which is used for PIIs, early PIIIs, and early Athlons).

DDR means Double Data Rate - in each clock cycle, this RAM transfers information twice. This means that at a given speed (say 200Mhz) it will transfer twice as much information as SDR (Standard Data Rate) RAM would. Almost all RAM now is DDR. The 400 refers to the fact that it transfers data 400 times per second. However, since it's actually transferring data twice per clock cycle, it's only really doing 200 clock cycles per second (200Mhz).

The PCxxxx numbers on RAM reflect the theoretical bandwidth. For example, DDR400 (PC3200) RAM transfers data 400 times per second, and each time it transfers 64 bits of data. You then divide by 8 to give the number of bytes of data (8 bits = 1 byte). So, the theoretical bandwidth of DDR400 RAM is 3200Mb/s or 3.2Gb/s.

You will often hear references to 'dual channel RAM'. Most P4s and some Athlons can make use of this. Pretty much what it does is transfer data to or from two RAM sticks at the same time. The result is that there's a 128-bit RAM bus instead of the normal 64-bit. This doubles the effective memory bandwidth.

On P4 systems, where the CPU has an 800Mhz FSB (capable of moving 6.4Gb/s), and DDR400 RAM (capable of moving 3.2Gb/s), dual-channel RAM is very useful. It doubles the RAM bandwidth, so suddenly the RAM can transfer data fast enough to completely saturate the FSB.

On AthlonXP systems, it's not that useful. AthlonXPs have a 400Mhz FSB, capable of moving 3.2Gb/s. DDR400 RAM can already saturate this in single-channel mode. There's no point in having dual-channel there, and the performance differences are small. There is one place where dual-channel RAM can help an AthlonXP. If you have a mainboard with onboard video, the onboard video relies completely on the system RAM. By doubling the effective RAM bandwidth, the video card gets a huge performance boost. However, this only applies to systems with Nforce2 mainboards, because only they support dual-channel RAM.

On Athlon64 systems, you can ignore dual-channel RAM because the A64 has a memory controller built into the CPU, and it can't take dual-channel.

The Athlon64-FX has dual-channel support (as does the Opteron), but so far performance increases seem small.

Dual-channel RAM: To run RAM in dual-channel mode, you normally need almost identical RAM sticks. Preferably ones which are manufactured on the same day, by the same company, in the same place. Running dual-channel can also significantly lower the maximum overclock. For these reasons, some companies have produced special RAM which is made for dual-channel work. Corsair TwinX is the most common type - a pack of TwinX contains two identical sticks which have been tested at their rated speed, in dual-channel mode, and are guaranteed to work.

The numbers that people use to refer to RAM represents the speed of certain functions of the RAM. There's a decent explanation here which probably explains it all better than I could.


SATA
Serial ATA. This is a new type of connection for HDD. Pretty much, it provides more speed than the old 'PATA' (Parallel ATA) connections, uses much thinner cables, and is quickly becoming the standard way of connecting HDDs to the mainboard.

RAID
Redundant Array of Inexpensive Disks. Pretty much, this allows you to connect two identical HDDs and make them look like one. The two most common connection methods are RAID0 and RAID1.

RAID0 means that the RAID controller sends half the data to each HDD. That means that each drive only has to read or write half as much as it would normally, so transfers happen much faster. It also gives you a drive which is the total capacity of both disks. So, if you want a 500Gb drive, you can connect two reasonably cheap 250Gb drives together, which gives excellent performance and it doesn't cost too much. The disadvantage is that if one drive dies, then you lose all the data from the array.

RAID1 is a simple way of making sure you don't lose any data. Pretty much, all data is stored on both drives. So, if you connect two 40Gb drives to the RAID controller, it'll send the same data to each disk. If one fails, the other one will be able to continue working normally until you get a replacement. However, from the set of two 40Gb disks, you only get 40Gb of space in total (because everything is stored in two places). The read speed is faster (since it can read half the data from each drive), but the write speed is no better than having a single drive.

There are lots of other types of RAID, but RAID0 and RAID1 are the ones which are normally supported by onboard controllers. SATA RAID simply means that the RAID controller uses the SATA ports on a mainboard, and so can use RAID on SATA drives.

Firewire
Firewire is a connection for external devices, similar to USB (but faster). It'll be around the same speed as USB2. It's normally used for digital video (due to the high speed), but it's also used for lots of external devices like drive boxes. Some mainboards have this onboard.

[looktall edit: added more content from another one of SLATYE's excellent posts]

[SLATYE]

A bit more:

Mainboards
The mainboard is pretty much what holds the whole system together. It provides a link between the CPU, RAM, and all the other components. Many modern boards also have many of the components already built in.

The Northbridge deals with only a few things - it connects the AGP, CPU, and RAM together, while also providing a link to the Southbridge. Since the Northbridge has to work very fast (at whatever speed the FSB is) and it does a lot of work, it does tend to get very hot. The NB is also in charge of handinlg the RAM, converting requests from the CPU into requests that the RAM can deal with. In systems using dual-channel RAM, it has to handle the 128-bit 400Mhz to 64-bit 800Mhz conversion.

The Southbridge connects all the slower stuff together. It connects the PCI slots to the northbridge, and most current ones also include IDE and sound hardware. Since it runs slower, the SB doesn't get too hot (although on some mainboards it still needs a heatsink).

FSB
Front Side Bus. This is the main communications channel from the CPU to the northbridge (and to the rest of the system). The CPU speed is determined by multipying the FSB by a given value. On Athlons, the FSB is a DDR based system. The real speed is doubled to give theoretical speed (so 100Mhz real FSB = 200Mhz theoretical FSB). On P4s and Athlon64s, there is a QDR (quad-data-rate) system, so all speeds are multiplied by 4 (100Mhz real FSB = 400Mhz theoretical FSB).

Generally, it's a good idea to make sure the RAM runs at the same speed as the FSB. That way, all of their clock cycles are synchronised, and there's less work for the northbridge to do. P4s don't really mind having the RAM running at a different speed to the FSB (just as well, since there's no 800Mhz RAM available right now). However, Athlon chipsets aren't nearly as good at dealing with asynchronous FSB/RAM, and there tends to be a massive performance drop.

All current CPUs use a 64-bit FSB. P4s make use of dual-channel RAM to get 128-bit 400Mhz memory access, which then gets converted to 64-bit 800Mhz for transfer to the CPU.

[SLATYE]

CPUs
CPUs are in a period of rapid change right now. The AthlonXP and P4 Northwood, which have both been around for quite a while, are just getting phased out, and new sockets are being introduced.

AMD
AMD has three major product lines right now, each with it's own socket:
Athlon XPs: These are still by far the best value for money CPUs around. There are also Durons around which are Athlon XPs with some cache disabled. Compared to P4s, they tend to be good at gaming but pretty bad at encoding/decoding due to the low core speed. They use Socket A, which has been around for ages. Unfortunately, the stock heatsink for AXPs is pretty terrible (the first thing I'd change in any AthlonXP system).

There are two main cores around now - Bartons and Tbreds. Bartons have 512kb L2 cache and are somewhat more efficient than Tbreds. They come as 2500+, 2600+, 2800+, 3000+ and 3200+ types. Tbreds are an older design using 256kb L2 cache. They have a higher core speed for a given speed rating. Tbreds can be 1700+, 1800+, 2000+, 2100+, 2200+, 2400+, 2600+, 2700+, and 2800+ CPUs.

The CPUs slower than the 2500+ use a 133Mhz clock-doubled FSB (normally referred to as 266Mhz). A few very rare 2600+ CPUs also use this. The 2500+ to 3000+ CPUs use a 166/333Mhz FSB, and the 3200+ uses a 400Mhz FSB.

Currently all AthlonXPs are multiplier-locked, so you can only overclock using the FSB. However, there are a few Mobile AthlonXPs around, which not only use a lower core voltage, but have an unlocked multiplier. These are quite excellent for overclocking. Otherwise, the 2500+ is a very popular CPU, because by simply increasing the FSB to 200Mhz, you get a CPU which is in all ways identical to a 3200+.

The 3000+ and 3200+ are currently poor value, because they don't compare well to the equivalent P4s. Below that, AXPs are excellent value.

Athlon 64s
Athlon64s are one of AMD's new CPU types, supporting 64-bit instructions. They're significantly faster than AthlonXPs at a given core speed, and they tend to be given pretty good ratings when compared to P4s (the 3000+, at 2000Mhz, tends to beat a P4 3.2 in many tests). The lower core speed helps them run pretty cool, which is nice. These CPUs currently use Socket 754, which will take CPUs up to the 3700+. The highest current A64 is the 3400+.

Athlon64s have the memory controller on the CPU, which reduces the cost of mainboard by a bit, since the Northbridge is almost eliminated. The current ones use single-channel DDR400 RAM.

AMD has an interesting strategy with Athlon64 CPUs. The multiplier can be adjusted to be lower than stock, but not higher. This means that if you've got a mainboard and RAM capable of running well above 200/400Mhz, you can increase those speeds and drop the multiplier. The CPU stays at the same speed, but benefits from greater RAM bandwidth.

Athlon64s currently use 1Mb of cache RAM, except for the 3000+ and 2800+ types (the 2800+ isn't in Australia yet). Those two use 512kb cache.

Unlike the AthlonXP with the core exposed, AMD has finally put a heatspreader on their CPUs to protect the core.

Athlon64-FX and Opteron
These are AMD's high-end CPUs. The Opteron quite thoroughly beats the Intel Xeon CPUs in a lot of dual-CPU tasks, which has led to many companies starting to adopt it. The Athlon64-FX is just like the Opteron, but without the ability to run in a dual-CPU system.

Both of these CPUs are completely multiplier-unlocked, but you pay quite a bit for that privelidge. They also both use Socket 940 at the moment.

Both the A64-FX and Opteron require Registered DDR RAM. Some of them can take DDR400, others need DDR333. It's actually very hard to find Registered DDR400 at the moment. Both CPUs can also run dual-channel RAM to double memory bandwidth. On some dual-CPU mainboards (like Tyan's Thunder K8W), each Opteron CPU has it's own bank of RAM. If one CPU needs information fast, it can use not only it's own memory but also the other CPU's Hypertransport links to get it done. This helps a lot in Operating Systems which actually understand how it works.

Currently, there's not much point in buying an Athlon64-FX. The normal Athlon64 is very close in performance at any given clock speed, and it costs much less. There's also no problem with finding Registered RAM, since the A64 doesn't need that.

Socket 939
Socket 939 is the new standard for AMD. Within a few months, both Athlon64 and Athlon64-FX CPUs will use this. Both types will use non-registered DDR RAM in dual-channel. This should lead to a decent performance increase. As far as I know, the Hypertransport frequency is also increasing from 800Mhz (on current CPUs) to 1000Mhz. This socket will be AMD's main socket for quite a while. The Opterons will continue to use Socket 940 and Registered DDR RAM.

Intel
I don't know much about Intel's server solutions, so I'll ignore those. Apart from them, Intel has four main CPU types:

Celeron
This is Intel's budget CPU. It's very slow, due to the small (128kb) cache (Intel CPUs suffer a lot from having a small cache). Generally, a Celeron will be about half as fast as an equivalent-rated AthlonXP. For example, an XP1700+ would need something like a Celeron 3400 to equal it in performance. That said, Celerons do have some uses. In applications where clock speed matters, they're excellent value, and they are also good for budget business systems (where performance isn't essential). The Intel stock heatsink for these is quite excellent. There's a new set of Celerons coming out which are based on the Prescott core and have 512kb cache. This makes them reasonable value for money, unlike the current Northwood-based Celerons.

All current Celerons do not use Hyperthreading, use a 400Mhz FSB, and use Socket 478.

P4 - Northwood
This is Intel's main P4 core. It's pretty good, but it doesn't process nearly as many instructions per cycle as AMD CPUs do (that's why an Athlon64 3200+ only needs to run at 2Ghz to equal a P4 at 3.2Ghz). These CPUs are built on the 130nm manufacturing process - the same as AthlonXPs and Athlon64s. The stock heatsink is excellent. All Northwood CPUs have 512kb cache.

The Northwoods currently come in two types: 533Mhz FSB, and 800Mhz FSB. The 533Mhz ones are older, and are referred to a "P4-B" CPUs. The top model, the P4-B 3.06Ghz, was the first CPU where Hyperthreading was enabled. I'll explain Hyperthreading a bit later. All current 800Mhz FSB ones use Hyperthreading.

All current Northwoods use Socket 478.

P4 - Prescott
This is the new P4 core. It's built on the 90nm manufacturing process, and has a few advantages over the Northwood. These include SSE3 (instructions for some multimedia applications), updated Hyperthreading, and the possibility of 64-bit (this is a rumour. Intel won't confirm it yet). However, the Prescott has some problems. It's got about 10 more pipeline stages, which means that when it needs data that isn't in the cache, it slows down a lot (the Northwood did too, but this does it worse). Intel has partially fixed this by using a 1Mb cache.

Performance is still below the Northwood, but it looks like after about 3.4Ghz the Prescott will be faster per clock cycle.

The other great problem for Intel right now is heat. The Prescott outputs a fairly ridiculous amount of heat, around 103w for a 3.2Ghz part. This results in temperatures about 10 degrees higher than Northwoods normally. For overclockers, who can deal with this heat, the Prescotts are excellent.

Prescotts also need a mainboard capable of supplying enough power to run properly. This isn't too common - most mainboards weren't build for this sort of power draw. Check before you buy.

All Prescotts currently feature Hyperthreading and an 800Mhz FSB, except for the P4 2.4A.

The P4 2.4A (pretty much a Prescott Celeron) is looking very nice right now. They should be cheap, and yet overclocking looks very good. Unfortunately, this CPU doesn't have Hyperthreading and only uses a 533Mhz FSB.

Prescott CPUs are referred to as P4-E CPUs. Don't confuse that with the P4-EE CPUs.

P4-EE
The Pentium 4 Extreme Edition. This is a Northwood core, but with an extra 2Mb of cache added as L3 cache. The performance increase can be quite large. The core isn't actually a Northwood - it's a Xeon core modified to fit in Socket 478. This was Intel's answer to the Athlon64-FX51 CPU, and they're pretty much equal in many things. The CPU alone is over $1000, so personally I wouldn't buy one. Unlike the Athlon64-FX, these are multiplier locked. They are available in 3.2 and 3.4Ghz versions currently.

Hyperthreading
The big thing which stops P4s performing well is that they have very long pipelines. One result of this is that most of the CPU spends all it's time doing nothing. Hyperthreading is meant to fix this. It makes the CPU act like two processors, so it gets fed twice as much stuff at once. The result is that more of the CPU actually spends time doing useful work. In some tasks it can also reduce performance, but not by much. This even works with WindowsXP Home Edition, which isn't meant to support two CPUs.

P4 FSBs
It may seem strange that I keep referring to the P4s as having a '400Mhz FSB' or '800Mhz FSB', while the AthlonXP is still using a 200Mhz one (clock doubled to 400Mhz). The fact is, both CPUs actually use FSBs from 100Mhz to 200Mhz. The AthlonXP transfers data twice in each cycle, leading to a clock-doubled FSB (so at 100Mhz true speed, it's actually transferring data at 200Mhz). TheP4 does the same, except with four transfers in each cycle. I think the Athlon64 does this too.

Dual-channel RAM
RAM bandwidth is always a problem for CPUs. When the CPU is running at 3.2Ghz and the RAM is only running at 200Mhz (clock doubled to 400Mhz), there's a bit of a problem when the CPU actually needs data from the RAM. AthlonXPs just live with this problem normally. Some chipsets like the Nforce2 offer dual-channel RAM, but this doesn't help much.

P4s can use dual-channel RAM to a great advantage. The FSB is 800Mhz on current P4s, and it is 64-bits wide (bandwidth of 6.4Gb/s). DDR400 RAM is 400Mhz, 64-bits wide (bandwidth of 3.2Gb/s). The result is that the RAM can't hope to fill the FSB. However, what many chipsets do is connect two RAM sticks together, to get a 128-bit RAM bus. This now gives 6.4Gb/s RAM bandwidth, which can nicely fill the P4's FSB. It gives a decent performance increase.

The reason that it doesn't help AthlonXPs is that they only have a 400Mhz, 64-bit FSB, so even with RAM capable of transferring 6.4Gb/s, the FSB can only ever move 3.2Gb/s to the CPU.

64-bit
This is something quite new. The last time that all deskops did a move like this was when we moved from the 80286 to the 80386. Performance should be very nice when 64-bit applications actually become available. The Athlon64 definitely supports 64-bit processing, as do the Athlon64-FX and Opteron. The P4 Northwood doesn't, and we're still debating about the Prescott. Intel was originally going to include support soon, but then Microsoft decided to support AMD's 64-bit extensions instead of Intel's. The result was that Intel pretty much had to start again.

LGA775
This is Intel's new socket, to replace Socket 478. In this design, the CPU legs are actually part of the socket - the CPU doesn't actually have legs attached. However, a lot of mainboard manufacturers are very annoyed at this because the socket is fragile and it's going to lead to a lot of RMAs. I don't think anyone knows how we'll deal with this yet.

[SLATYE]

Copied from this thread. I'll re-write it and the previous few posts when I've got time (and when I remember).

Originally Posted by heyimsu

So the Q6600 runs at 1066FSB. Since there are 4 cores, that means 1066/4 = 266 ??

Quick explanation:

[EDIT: Fine, so it's not actually a 'quick' summary. Unfortunately I can't think of any way to make it much shorter]

All Intel CPUs since the first P4 have used a QDR FSB. What this means is that in each clock cycle of the FSB, data gets transferred four times. That's where the 266 comes from. The actual FSB frequency is only 266MHz. However, because it transfers data four times in each clock cycle, the effective data frequency is four times 266MHz (1066MHz, or close enough).

Naturally Intel advertise this number (1066MHz), because most consumers think that higher numbers are always better. I perfer to use the real FSB frequency (266MHz in this case).

The FSB transfers 64 bits (8 bytes) of data at a time, so the total bandwidth for the Q6600's FSB is:
[code]1066 (million transfers per second) * 8 (bytes per transfer) = 8533MB/s, or 8.533GB/s[code]

Now we'll have a look at RAM. Current RAM is DDR, which means that it transfers data twice in each clock cycle. Again, the manufacturers almost always quote the effective data frequency instead of the actual clock speed ("DDR2-800" is really 400MHz; "DDR2-1066" is really 533MHz).

A single RAM module can transfer 64 bits (8 bytes) of data at a time. However, almost all current mainboards can use dual-channel. This links two modules together to double the amount of data transferred to 128 bits (16 bytes). This does require two RAM sticks, which is why everyone goes for 2x 1GB sticks instead of one 2GB stick.

Note that it doesn't apply for more RAM - you can't transfer 256 bits (32 bytes) at a time with four RAM sticks unless you've got a very expensive server mainboard. Four RAM sticks hurt overclocking anyway, so just stick with two.

Getting back on topic, the bandwidth of a pair of DDR2-800 sticks in dual-channel mode is given by:

800 (million transfers per second) * 16 (bytes per transfer) = 12800MB/s or 12.8GB/s

As you can see, the RAM bandwidth is actually much higher than the FSB bandwidth. There's not much point having this extra bandwidth - the CPU can never access more than 8.533GB/s because all the data going to or from it has to pass through the FSB.

In fact, for a Q6600 at stock speed, you can saturate the FSB with DDR2-533 RAM (not that it's a good idea now, since faster RAM is so cheap).

The simple rule of thumb is that you need to keep the RAM frequency greater than or equal to the FSB frequency (266MHz in this case), or alternatively keep the RAM data frequency at half the FSB data frequency (so DDR2-533 and a "1066MHz" FSB will go together well).

For overclocking this changes, because you overclock by increasing the FSB. This creates more bandwidth, so you can use faster RAM to take advantage of that. Some (most) boards will also force the RAM to run at the FSB frequency or faster, so you really do need decent RAM.

For example, if you get your Q6600 up to 3.6GHz (9x multiplier, 400MHz FSB frequency, 1600MHz effective FSB data frequency) then you'll want to have DDR2-800 RAM (really 400MHz) to go with that.

There's no harm in having faster RAM, except for the cost. If you can get DDR2-1333 RAM cheaply (or if you move straight to DDR3-1333) then that'll still be fine with a low FSB - you just won't be fully using the available bandwidth.

[jayjones1]

thanks for the tips

[Daft_Munt]

Any how to upload pixs to OCAU? Daft need to learn how to read!

[cnd]

Originally Posted by anlashok

What does defrag mean?

Good explanation, up to here:-

This slows down your performance as you are kept waiting...
Yes Defraging is normally a good thing.

which is wrong.

Basically - do some timing, defrag, give it a day or 2, then do more timing - you'll see it's visibly slower.

Or - in detail - "performance" isn't usually affected by fragmentation, since not all files are read in entirety, and most "reads" are of multiple files - so - doing a "defrag" will intentionally move things around to make multi-file and partial reads all SLOWER, because the original "fragmented" situation more than likley had most files that are used most often more-or-less "close" to one-another.

Spreading them all out (which is what a defrag does) will absolutely guarantee that as soon as you start using your machine, everything will get a lot slower, a lot faster, since the "worst case scenario" (having to write/read a file that's all the way at the other end of the disk) will always happen.

Sad but true. DeFrag is bad. In fact - the "perfect" hard disk would be comprised of files that are deliberately arranged and pre-fragmented such that the most-often-read-all-at-once sequence of files and file parts are all alongside each other. Sadly - nobody's written a tool to do this yet :-(

[anlashok]

Defraging is normally a good thing. However the subject your refering to is "file placement" these types of things can now be done, see UltimateDefrag for example.

You will find there are many variables that you would need to take into consideration if you wish to dive into the finer details of storage system optimization and performance. This can be disscussed in depth in the "Storage & Backup" section.

[SLATYE]

Originally Posted by cnd

Or - in detail - "performance" isn't usually affected by fragmentation, since not all files are read in entirety, and most "reads" are of multiple files - so - doing a "defrag" will intentionally move things around to make multi-file and partial reads all SLOWER, because the original "fragmented" situation more than likley had most files that are used most often more-or-less "close" to one-another.

I don't think this is correct. No defragmenting means that any one file is more likely to be closer to part of any other file - but that's pretty pointless (since that part of the other file probably won't be the part you want).

[cnd]

Originally Posted by SLATYE

...any one file is more likely to be closer to part of any other file

Actually - it would be more likely to be closter to part(s) of other files that were written to disk around the same time. This being my point: those are the things more likely to be requiring to be read more-or-less together/at-once in the future.

It's all a moot point anyhow - SSDs are already replacing mechanics - it won't be long before hard discs are as common as floppies.