Updated CPU/GPU List - 6/8/2012
A thread by mahnini and I to inform you about computers!
+ Show Spoiler [Want a list of parts/Build?] +
Important questions you should ask yourself
If you request a build, please answer these questions. We will spend as much effort on your build as you spend on your answers to these questions! I urge people to abstain from providing builds unless all the questions are answered.
What is your budget?
This should be an obvious one, you should determine a price range that best describes how much you want to spend on your build. It makes it easier for people to make recommendations.
What is your resolution?
This is extremely important especially for a gaming machine. I see people all the time asking for a video card recommendation and 5 people reply "5850" and the person hasn't even given their gaming resolution; it turns out the person games at 1280x1024 and could have ended up wasting $200 on a card that is total and complete overkill for his resolution.
What are you using it for?
Mostly gaming? Some gaming with video encoding mixed in? Photoshop? Streaming? All this matters, you can end up spending a lot of money and not be able to do anything you wanted to or you can spend a lot of money on stuff you don't even want to do.
What is your upgrade cycle?
Knowing the answer to this question makes it easier for people to recommend CPU sockets and GPUs so you get the most for your money. A person with a longer upgrade (2+ years) cycle will probably end up spending more initially but end up saving more at the expense of some performance. On the other hand a person with a shorter upgrade cycle (1 - 2 years) wants to spend less initially so they can put more money toward their next build.
When do you plan on building it?
Computer parts shift in prices often. With the release of new products, exist prices may drop further. Even without the release of new prices, changes in supply can affect the price of a specific part. The HD 5850 was supposed to retail at the $260 mark, but in a mere 4-5 months after its release it climbed to top $300 for basic reference models.
Do you plan on overclocking?
If you plan on overclocking, the motherboard choice and heatsink choice will be affected. Enthusiast overclocking requires enthusiast heatsinks and at least a stable motherboard. A simple budget Foxconn motherboard will not be adequate for your overclocking needs!
Do you need an Operating System?
An OS is around $100. It affects how much of your budget we can spend on the actual hardware.
Do you plan to add a second GPU for SLI or Crossfire?
CrossFire and SLI are powerful solutions that allow higher performance, but in order to utilize them, we have to choose motherboards and PSUs that can support the increased demands that Crossfire and SLI place on your components.
Where are you buying your parts from?
If you have a nearby Fry's or Microcenter you can utilize their deals. If you're in Europe you obviously cannot use Newegg/Fry's/Microcenter so we need to know where you are from to put together a list of parts from a website or area you can actually buy from.
If you can't answer these questions, make sure you can answer them before you ask us to help you put together a list of parts for you. Every single question affects the build we will put together for you greatly so make sure you can answer them accurately!
If you request a build, please answer these questions. We will spend as much effort on your build as you spend on your answers to these questions! I urge people to abstain from providing builds unless all the questions are answered.
What is your budget?
This should be an obvious one, you should determine a price range that best describes how much you want to spend on your build. It makes it easier for people to make recommendations.
What is your resolution?
This is extremely important especially for a gaming machine. I see people all the time asking for a video card recommendation and 5 people reply "5850" and the person hasn't even given their gaming resolution; it turns out the person games at 1280x1024 and could have ended up wasting $200 on a card that is total and complete overkill for his resolution.
What are you using it for?
Mostly gaming? Some gaming with video encoding mixed in? Photoshop? Streaming? All this matters, you can end up spending a lot of money and not be able to do anything you wanted to or you can spend a lot of money on stuff you don't even want to do.
What is your upgrade cycle?
Knowing the answer to this question makes it easier for people to recommend CPU sockets and GPUs so you get the most for your money. A person with a longer upgrade (2+ years) cycle will probably end up spending more initially but end up saving more at the expense of some performance. On the other hand a person with a shorter upgrade cycle (1 - 2 years) wants to spend less initially so they can put more money toward their next build.
When do you plan on building it?
Computer parts shift in prices often. With the release of new products, exist prices may drop further. Even without the release of new prices, changes in supply can affect the price of a specific part. The HD 5850 was supposed to retail at the $260 mark, but in a mere 4-5 months after its release it climbed to top $300 for basic reference models.
Do you plan on overclocking?
If you plan on overclocking, the motherboard choice and heatsink choice will be affected. Enthusiast overclocking requires enthusiast heatsinks and at least a stable motherboard. A simple budget Foxconn motherboard will not be adequate for your overclocking needs!
Do you need an Operating System?
An OS is around $100. It affects how much of your budget we can spend on the actual hardware.
Do you plan to add a second GPU for SLI or Crossfire?
CrossFire and SLI are powerful solutions that allow higher performance, but in order to utilize them, we have to choose motherboards and PSUs that can support the increased demands that Crossfire and SLI place on your components.
Where are you buying your parts from?
If you have a nearby Fry's or Microcenter you can utilize their deals. If you're in Europe you obviously cannot use Newegg/Fry's/Microcenter so we need to know where you are from to put together a list of parts from a website or area you can actually buy from.
If you can't answer these questions, make sure you can answer them before you ask us to help you put together a list of parts for you. Every single question affects the build we will put together for you greatly so make sure you can answer them accurately!
+ Show Spoiler [CPU] +
-New build? Make it Intel.
-AMD upgrade? Bulldozer sucks, just get a cheaper Phenom II X4/X6.
Sub $100
Celeron G530
Athlon II X4 640
Pentium G620
Athlon II X3 445
$100-$150
Core i3 2100
Phenom II X4 955 AM3
$150-$200
Core i5 3450 LGA 1155
Core i5 2400 LGA 1155
Phenom II X6 1090T
$200-$300
Core i5 3550 LGA 1155
Core i5 2500K LGA 1155
Core i5 2500 LGA 1155
Core i5 3570K LGA 1155
$300+
Core i7 3770K LGA 1155
Core i7 2600K LGA 1155
Core i7 3820 LGA 2011
Core i7 3930k LGA 2011
Core i7 3960X LGA 2011
you have too much money
2011 vs 1155
There is almost no reason to build on LGA 2011. The only reasons you would build on it is if you need 32GB+ of Quad-Channel RAM, x16/x16 Dual Graphics or like wasting money.
1155 vs 1156 OBSOLETE
LGA 1155 CPUs are not compatible with LGA 1156 motherboards and vice versa. However, many LGA 1155 motherboards will also support the heatsinks with LGA 1156 brackets. Contrary to popular belief, this change was not avoidable due to the new architecture of the Sandy Bridge CPUs.
As far as performance goes, there isn't much reason to upgrade from a higher end Core 2 Quad/Phenom II X4/Core i5/i7 as the performance benefits from the Sandy Bridge CPU will not be as noticeable, especially if your CPU is overclocked.
1156 vs 1366 OBSOLETE
There are pretty much only 3 reasons that a person should be going for Socket 1366: you absolutely NEED triple channel or a dual graphics setup, you want to overclock your CPU at stock voltages, or you enjoy benchmarking and value your e-penis. Socket 1156 provides the same CPU performance but costs less, draws less power, and runs cooler. If you are looking at LGA 1366 for upgradability you are looking at the wrong socket. Both LGA 1156 and LGA 1366 WILL be replaced by next year. The new Intel CPUs will use a different socket.
+ Show Spoiler [Motherboard] +
AMD Chipset:
x16/x16
890FX
790FX
x8/x8
890GX
790X
x16
880G*
870
785G*
770
Intel Chipset:
x16/x16
X58
P55**
P67**
X79
x8/x8
P55
P67
Z68^
Z77^
x16
H55***
H57***
H61^"
H67^"
H77^"
Z68^
Z77^
X79
*Integrated graphics on motherboard but can be cheaper than non-IGP counterparts.
**Some P55 boards can achieve the full x16/x16 bandwidth through use of the Nvidia NF200 chip. This chip however is expensive, and generally it isn't worth the money to buy such P55 motherboards.
***Integrated graphics when paired with an i3 dual core or i5 dual core
^Integrated graphics when paired with 2nd/3rd generation Intel Core i3/i5/i7 series CPU.
"Does not support overclocking
800Chipset vs 700Chipset
The 8xx chipset on the AMD motherboards is a new chipset that AMD released in 2009. The most notable of the additions to the 800 series is the SB850 which added native SATA 6GB/s support. Instead of using a third party chip that would occupy PCIe Express lanes, the SATA 6GB/s is processed straight through the chipset itself which allows the full bandwidth for PCIe expansion cards like graphics cards and the full 6GB/s bandwidth for SATA.
x16/x16 vs x8/x8 vs x16
This denotes PCI-E bandwidth, x16/x16 provides full bandwidth to dual GPU configurations, whereas x8/x8 does not. x8/x8 is generally fine for cards less powerful than the HD 4870. The single x16 denotes that the motherboard only supports a single GPU configuration.
x16/x16
890FX
790FX
x8/x8
890GX
790X
x16
880G*
870
785G*
770
Intel Chipset:
x16/x16
X58
P55**
P67**
X79
x8/x8
P55
P67
Z68^
Z77^
x16
H55***
H57***
H61^"
H67^"
H77^"
Z68^
Z77^
X79
*Integrated graphics on motherboard but can be cheaper than non-IGP counterparts.
**Some P55 boards can achieve the full x16/x16 bandwidth through use of the Nvidia NF200 chip. This chip however is expensive, and generally it isn't worth the money to buy such P55 motherboards.
***Integrated graphics when paired with an i3 dual core or i5 dual core
^Integrated graphics when paired with 2nd/3rd generation Intel Core i3/i5/i7 series CPU.
"Does not support overclocking
800Chipset vs 700Chipset
The 8xx chipset on the AMD motherboards is a new chipset that AMD released in 2009. The most notable of the additions to the 800 series is the SB850 which added native SATA 6GB/s support. Instead of using a third party chip that would occupy PCIe Express lanes, the SATA 6GB/s is processed straight through the chipset itself which allows the full bandwidth for PCIe expansion cards like graphics cards and the full 6GB/s bandwidth for SATA.
x16/x16 vs x8/x8 vs x16
This denotes PCI-E bandwidth, x16/x16 provides full bandwidth to dual GPU configurations, whereas x8/x8 does not. x8/x8 is generally fine for cards less powerful than the HD 4870. The single x16 denotes that the motherboard only supports a single GPU configuration.
+ Show Spoiler [RAM] +
Frequency vs Cas Latency and Timings
In general CAS Latency can be seen as the amount of time it takes for memory to do whatever you tell it to do. Generally the lower the latency, the better. However, CAS latency is dependent on clock cycles because instructions can only be sent at a certain clock frequency (1066MHz, for example). When looking at RAM the higher the frequency and lower the CAS latency the better.
Don't need 8GB of RAM
4GB of RAM is the standard now. There is almost no need at the current moment to get more or less than 4GB. There are certain cases in which extra RAM may be useful such as heavy Photoshop use. Like many parts it's heavily dependent on what you and what applications you use.
Triple Channel vs Dual Channel
The X58 is currently the only chipset available that supports triple channel memory. The difference between triple channel and dual channel is very marginal and may only see performance increases in RAM heavy applications. Triple channel memory addresses memory bandwidth when dealing with very memory intensive tasks and is oriented more towards Intel Xeon platforms, not i7 platforms.
Native frequency support
As of 2008, all new CPUs have onboard memory controllers that can specify how fast RAM can run by default. Intel LGA 1156 CPUs have a stated max of 1333MHz, LGA 1366 have a stated max of 1067MHz, and AMD AM3 CPUs have a stated max of 1333MHz. The higher speeds of 1600MHz and above are allowed by the motherboard manufacturers and also by overclocking. At one point, the AMD memory controllers had a hard time coping with the 1600MHz RAM and would force RAM rated at 1600MHz to run at 1333MHz. The problem was caused by the memory controller in the 2nd revision AMD CPUs. This has supposedly been fixed as of the new 3rd revision CPUs which were released in 2009.
All new Intel Sandy Bridge CPU (2nd generation Core i3/i5/i7) have a base clock of 100MHz. This base clock is tied to other frequencies on the CPU (such as the PCIe bus frequency) so only very small changes can be made to it.
Low Voltage RAM
Often times you will see RAM modules marketed as being 'Low Voltage'. This is much more common with DDR3 because the memory controller on the i3/i5/i7 CPUs cannot handle more than 1.65V while the Core 2 Duo/Quad front side bus was able to handle up to 2.1V on the memory. Generally it is better to purchase RAM that operates at a lower voltage, so if two specs of RAM are exactly the same except for the voltages, the lower voltage sticks will be better. However, beware of the 1.35V and 1.25V modules of RAM that are marketed as 'extra low voltage'. While it is true that they operate at lower voltages, many motherboards will not allow them to operate at such low voltages AND in general, you do not save much more electricity. If you are concerned about saving electricity, you should first spend money on a more efficient power supply, more efficient motherboard, or more efficient graphics card because those are the parts that will make the most difference in power consumption.
In general CAS Latency can be seen as the amount of time it takes for memory to do whatever you tell it to do. Generally the lower the latency, the better. However, CAS latency is dependent on clock cycles because instructions can only be sent at a certain clock frequency (1066MHz, for example). When looking at RAM the higher the frequency and lower the CAS latency the better.
Don't need 8GB of RAM
4GB of RAM is the standard now. There is almost no need at the current moment to get more or less than 4GB. There are certain cases in which extra RAM may be useful such as heavy Photoshop use. Like many parts it's heavily dependent on what you and what applications you use.
Triple Channel vs Dual Channel
The X58 is currently the only chipset available that supports triple channel memory. The difference between triple channel and dual channel is very marginal and may only see performance increases in RAM heavy applications. Triple channel memory addresses memory bandwidth when dealing with very memory intensive tasks and is oriented more towards Intel Xeon platforms, not i7 platforms.
Native frequency support
As of 2008, all new CPUs have onboard memory controllers that can specify how fast RAM can run by default. Intel LGA 1156 CPUs have a stated max of 1333MHz, LGA 1366 have a stated max of 1067MHz, and AMD AM3 CPUs have a stated max of 1333MHz. The higher speeds of 1600MHz and above are allowed by the motherboard manufacturers and also by overclocking. At one point, the AMD memory controllers had a hard time coping with the 1600MHz RAM and would force RAM rated at 1600MHz to run at 1333MHz. The problem was caused by the memory controller in the 2nd revision AMD CPUs. This has supposedly been fixed as of the new 3rd revision CPUs which were released in 2009.
All new Intel Sandy Bridge CPU (2nd generation Core i3/i5/i7) have a base clock of 100MHz. This base clock is tied to other frequencies on the CPU (such as the PCIe bus frequency) so only very small changes can be made to it.
Low Voltage RAM
Often times you will see RAM modules marketed as being 'Low Voltage'. This is much more common with DDR3 because the memory controller on the i3/i5/i7 CPUs cannot handle more than 1.65V while the Core 2 Duo/Quad front side bus was able to handle up to 2.1V on the memory. Generally it is better to purchase RAM that operates at a lower voltage, so if two specs of RAM are exactly the same except for the voltages, the lower voltage sticks will be better. However, beware of the 1.35V and 1.25V modules of RAM that are marketed as 'extra low voltage'. While it is true that they operate at lower voltages, many motherboards will not allow them to operate at such low voltages AND in general, you do not save much more electricity. If you are concerned about saving electricity, you should first spend money on a more efficient power supply, more efficient motherboard, or more efficient graphics card because those are the parts that will make the most difference in power consumption.
+ Show Spoiler [GPU] +
2560x1600
GTX 680
GTX 670
HD 7970
HD 7950
GTX 580
HD 6970
1920x1200
GTX 670
GTX 580
HD 6970
HD 7870
HD 7850
GTX 570
HD 6950 1GB
GTX 560 Ti
HD 6870
GTX 460 1GB
HD 6850
1680x1050
GTX 460 1GB
HD 6850
GTX 550 Ti
HD 7770
HD 6770
1280x1024
GTX 550Ti
HD 7750
HD 6770
GTS 450
HD 6750
HD 6670
Integrated vs Dedicated
The main difference between an integrated GPU and a dedicated GPU is that the dedicated GPU will have its own supply of VRAM while the integrated GPU will have to borrow System RAM, which is why your computer may appear to have less RAM than you thought! The integrated GPU is sharing (stealing) your system RAM!
Why is that important? All RAM is the same right...? No! In general, GPUs use very fast RAM which has a wider bandwidth, and faster transfer speeds. The faster a GPU can send information to the rest of the system, the better it performs. Typically, Dedicated GPUs will also have more dedicated RAM than the integrated can borrow from your system!
The other difference which is simply a result of the above is that integrated GPUs will be much weaker than their dedicated brothers! Because integrated GPUs do not have the memory bandwidth to support much data, they are made to process very little data. While a powerful integrated GPU can have as many as 16 or so pipelines (shaders), dedicated GPUs can have tens of hundreds! If you plan on gaming on your integrated GPU, don't!
Why does integrated exist then? Integrated GPUs are cheap to produce, are light on power consumption, run cool, and are generally sufficient for non-gaming, non-workstation machines. Many users will probably never need an HD 4870 or even an HD 4550!
HD 4770 vs 4850 OBSOLETE
The HD 4770 is a slightly newer card based on a 40nm manufacturing process as opposed to the 55nm of the HD 4850. The 4850 is the better card when it comes to absolute performance and is slightly better than the 4770. However, if temperature and power draw is a concern the 4770 provides a majority of the performance at lower temps and lower power consumption.
HD 4X00 vs HD 5X00 OBSOLETE
The HD 4X00 series is based on a 55nm manufacturing process while the HD 5X00 is based on a 40nm process. As a result, the 5X00 series is runs much cooler and consumes less power. However, the new 5X00 series underperforms when compared to similarly priced 4X00 series GPUs.
GTX 470 explanation OBSOLETE
Some of you may be surprised that there is an Nvidia Fermi card being recommended! When the GTX 470 was first released, it lacked driver support and performed fairly poorly compared the the ATi HD 5850 which was $50 or so cheaper. In addition to its heat and noise problems, this caused the GTX 470 to be largely disregarded along with the GTX 480. However, with the release of the new Nvidia drivers, the GTX 470 has been performing at around the level of the 5870 in many games and routinely performs above the HD 5850. Along with the price cuts in the GTX 470 stock (there are many GTX 470s at around $300-$320 at the moment), and the lack of price cuts in the HD 5850 line, this makes the GTX 470 much more attractive.
Why HD 5750 sucks
When the 57XX GPUs were first announced, the 5750 looked like a better buy than the 5770 because the 5750 would routinely beat the cards it competed against (HD 4850 and GTS 250), while retailing at around the same price and having the new features like DX11 and Eyefinity while the HD 5770 struggled against the HD 4870 and the GTX 260. However, while the price of the GTS 250 and the HD 4850 have dropped considerably over these 7-8 months where they can be found at $80, the HD 5750 has actually increased in price to around $120-140. At $140 the 5750 simply cannot compete against the HD 5770 which has routinely dropped in price to around $150.
Why the HD 5830 sucks
When it was first announced, the HD 5830 was supposed to be the card to bridge the HD 5850 and the HD 5770 at the $200 price range. However the HD 5830 turned out to be a disappointment because of its high retail price of $230, and because of its low performance. It failed flopped against the cheaper HD 4890 at $200, and showed very small increases over the HD 5770 which could not justify the $80 price premium. Due to the fact that it was based on a 58XX chip, it was also fairly large and shared the higher power consumption of the 58XX series while offering little of the performance. While the HD 5830 has dropped down to $200 recently due to the release of the GTX 460, we still cannot recommend it because the performance of the GTX 460 at around $200-$230 compeltely outstrips the HD 5830 in every respect.
Why GTX 465 really sucks OBSOLETE
The GTX 465 is a severely crippled Fermi that lacks everything good about Fermi, and retains everything bad about Fermi. It performs around the level of the HD 5830 yet sucks up the power of a GTX 470 and outputs the same heat as the GTX 470. It is an overcrippled GPU that has no chance of competing with the competitively priced HD 5850 or even the relatively disappointing HD 5830. Along with its extremely high price of $280 at launch, there was no reason to purchase the GTX 465. While GTX 465 prices have dropped to more reasonable $240 or $250 ranges, the recently released GTX 460 1GB at $230 often provides the same if not better performance without the heat output and power requirements at a lower price point.
SLI and CrossFire
SLI and CrossFire solutions are multi-GPU configurations that split the work between two GPUs instead of having just one GPU doing all the work. They serve the same purpose, but they are implemented differently. SLI is a technology that is restricted to the motherboards that Nvidia supports and Nvidia chipsets whereas CrossFire will function on any motherboard that has two PCIe x16 slots and does not have an Nvidia chipset. In general, SLI will scale better than CrossFire due the superiority of Nvidia drivers and game support.
Both SLI and CrossFire can use a 'bridge' that will allow the GPUs to communicate with each other. While SLI will function without the bridge and communicate across the PCIe bus, if you SLI two high powered cards such as the GTX 480, the PCIe bus will often be saturated, which will actually cause performance decreases much of the time. In general, you should use the SLI bridge whenever possible. While CrossFire bridges come with the graphics card when you purchase it, normally SLI bridges should come with your SLI Certified motherboard.
Update 1/25/2011
AMD has been working on their driver support on many of their games, and now CrossFire configurations can easily boast that they can top SLI configurations of similar prices. The two technologies are now relatively equal and choosing between the two can be heavily game dependent.
HD 6850 and HD 6870
First to explain the naming, the Barts 6850 and the 6870 are slower than their respective 58xx parents. The 6870 performs like a 5850 (sometimes better), and the 6850 performs under the 5850. The obvious purpose of these cards are to bridge the less expensive Juniper (57xx) based GPUs to the newer up and coming Cayman (69xx) GPUs as well as the older Cypress (58xx) GPUs, an area dominated recently by the GTX 460 768MB and GTX 460 1GB. This is an area where ATI has been lacking due to the failure of the HD 5830, and the lack of stock of the HD 4890.
Where it gets interesting is in pricing. The 6850 and 6870 have no clear counterparts in Nvidia's lineup in terms of price and performance. The GTX 460 768MB is between the 6850 and 5770 in terms of performance and price, the GTX 460 1GB is between the 6850 and 6870, and the GTX 470 is a bit above the 6870. So when you are asked to choose between ATI and Nvidia, it becomes more complicated because it is no longer a 'this or that' situation where you are paying the same amount. Choosing ATI or Nvidia can mean spending an extra $20 for that extra performance.
OBSOLETE
In my honest opinion, I believe that the GTX 460 768MB/1GB and the GTX 470 still occupy the sweet spot in terms of performance. The overclocking ability of the GTX 460 is unmatched, and the GTX 470 is still a very very good option at its new, lower pricepoint. However, anybody could argue the opposite and still be very correct.
HD 6970 vs HD 6950 vs GTX 570 vs GTX 580
Note: The 6970 and the 6950 in this article are the 2GB versions.
As far as the current 4 high end cards go, they are all great performers and in general performance ranks like this:
GTX 580 > HD 6970 >= GTX 570 >= HD 6950.
The GTX 580 is the only card that is for the most part clearly superior to the others, and that is reflected in its very high price tag. The HD 6970, GTX 570, and the HD 6950 are much closer together when it comes to performance.
As far as I'm concerned, the HD 6970 does not offer the performance to justify it being so much more expensive than the HD 6950. The HD 6950 along with the GTX 570 occupy the current sweet spot. In multi-GPU configurations, two HD 6950 often outperform a pair of GTX 570, however, the GTX 570 outperforms a single HD 6950 most of the time. Both are very effective GPUs at their price point while the two most expensive GPUs are not very reasonable. The HD 6950 comes out on top for me.
HD 6950 2GB vs HD 6950 1GB
As far as the two cards go, the only difference between them is that one is short 1GB of frame buffer. For most of us, this does not mean much because as long as you are at or under 1920x1200, the frame buffer does not matter as much as the other details such as the GPU clock, memory clock, or bus width. However, at higher resolutions than 1920x1200, and when high levels of anti-aliasing are applied, the 2GB version comes out on top.
Most of us are at a lower resolution and do not need/want the high level of quality of AA offers, so the HD 6950 1GB can be the better choice even though the savings aren't too great (around $20 or so at the time of writing).
HD 6950 1GB vs HD 6870 vs GTX 560 TI
All three are great cards that perform extremely well at their price points. Of the three, the HD 6870 probably offers the best cost/performance ratio for the Starcraft 2 player who doesn't want any extra anti-aliasing, plays at a comfortable 1920x1200 resolution, and has Ultra graphics. However, there will be times when the HD 6870 will slow down. That is why there are two other cards here, the HD 6950 1GB and the GTX 560 TI. Both are amazing cards at what they do, and are easily worth the $30-60 premium that you end up paying.
As far as choosing between the two, the GTX 560 TI outperforms the 6950 1GB in just as many games as it loses out to its competitor, so the choice is mostly based on which games you plan on playing, or based on cost alone. The GTX 560 TI is very overclockable and shows very good results when overclocked (rivaling that of a GTX 570).
Drivers
Wait a few days ( Nvidia ;_; ) and then..
INSTALL THE LATEST DRIVERS IMMEDIATELY BEFORE YOU COMPLAIN ABOUT YOUR GPU SUCKING OK?
OK? PLZ
+ Show Spoiler [PSU] +
12V Rail and Wattages
Most PSUs today use their 12V rail to power the most demanding pieces of hardware. The CPU, GPU(s), hard drives, and fans all use the 12V rail for power. When selecting a PSU, you want to make sure that your 12V rail has enough amps to support your hardware. Wattage does not mean everything when selecting a PSU. If a 1000W PSU has merely a 70A 12V rail, it can only supply 12 * 70W of power to your peripherals. You want the product of 12 and the number of amps on the 12V rail to be fairly close (around 0-50W) to the wattage of the PSU.
80+ Ratings and PSU ratings
PSUs are rated by the wattages that they can deliver at max output safely. This means that a 750W PSU will not be pulling 750W of electricity out of the outlet at all times. It will only pull enough electricity to support the components. 80+ ratings are standards that some PSUs adhere to. If a 400W PSU is 80+ certified, that means at the test points of 20% load, 50% load, and at 100% load at room temperature it is above 80% efficient. There are 4 different levels of certification that demonstrate different levels of efficiency. There is 80+ Certified, 80+ Bronze, 80+ Silver, and 80+ Gold. The advantage of purchasing an 80+ PSU is that it will of course be more power efficient. If you need 400W out of a 400W 80+ PSU, it will pull 500W out of the outlet, but if you use a PSU with a lower efficiency PSU, it would pull more than 500W. Another benefit is lower heat output. Because the PSU is more efficient, less power is wasted as heat so the PSU runs cooler, which increases its life span and in turn reduces noise because there is less heat to be exhausted.
Be careful when choosing 80+ PSUs however, because while many shady PSUs may pass the requirements at room temperature, at higher temperatures that occur during longer operation or during summer, they may not be able to hold their certification. 80+ Certification is a way for PSUs to market their PSUs so you should be aware of it when purchasing a PSU.
Reputable brands
Corsair, Antec, Seasonic, Silverstone, Enermax, XFX
Reliable brands Less so than the above in general
Thermaltake, Zalman, Cooler Master, Xigmatek
Reputable OEMs
Seasonic, Channel Wells Technology, Enhance Electronics, Delta
Most PSUs today use their 12V rail to power the most demanding pieces of hardware. The CPU, GPU(s), hard drives, and fans all use the 12V rail for power. When selecting a PSU, you want to make sure that your 12V rail has enough amps to support your hardware. Wattage does not mean everything when selecting a PSU. If a 1000W PSU has merely a 70A 12V rail, it can only supply 12 * 70W of power to your peripherals. You want the product of 12 and the number of amps on the 12V rail to be fairly close (around 0-50W) to the wattage of the PSU.
80+ Ratings and PSU ratings
PSUs are rated by the wattages that they can deliver at max output safely. This means that a 750W PSU will not be pulling 750W of electricity out of the outlet at all times. It will only pull enough electricity to support the components. 80+ ratings are standards that some PSUs adhere to. If a 400W PSU is 80+ certified, that means at the test points of 20% load, 50% load, and at 100% load at room temperature it is above 80% efficient. There are 4 different levels of certification that demonstrate different levels of efficiency. There is 80+ Certified, 80+ Bronze, 80+ Silver, and 80+ Gold. The advantage of purchasing an 80+ PSU is that it will of course be more power efficient. If you need 400W out of a 400W 80+ PSU, it will pull 500W out of the outlet, but if you use a PSU with a lower efficiency PSU, it would pull more than 500W. Another benefit is lower heat output. Because the PSU is more efficient, less power is wasted as heat so the PSU runs cooler, which increases its life span and in turn reduces noise because there is less heat to be exhausted.
Be careful when choosing 80+ PSUs however, because while many shady PSUs may pass the requirements at room temperature, at higher temperatures that occur during longer operation or during summer, they may not be able to hold their certification. 80+ Certification is a way for PSUs to market their PSUs so you should be aware of it when purchasing a PSU.
Reputable brands
Corsair, Antec, Seasonic, Silverstone, Enermax, XFX
Reliable brands Less so than the above in general
Thermaltake, Zalman, Cooler Master, Xigmatek
Reputable OEMs
Seasonic, Channel Wells Technology, Enhance Electronics, Delta
+ Show Spoiler [Case] +
Aluminum vs Steel
Aluminum's advantage over steel is that it is much lighter than steel in general. However, poorly made aluminum cases can cut your fingers when you work in them because aluminum is generally thinner and sharper than steel. Steel is generally a stronger material and consequently it will be heavier.
Case Choice
Case choice should be based on the looks, the airflow, cable management, and space. You must be able to have enough space for your components in your case. Whether you need space for an unorthodox motherboard form factor or an extra long graphics card, you need to be able to put the components in. Cable management makes the computer easier to work with, and can enhance airflow. Airflow is important in a case because without good airflow, the air in the case remains stagnant and the temperatures of your parts will increase.
Some companies that generally produce good, solid cases are Silverstone, Antec, Corsair, Cooler Master, Lian Li
Aluminum's advantage over steel is that it is much lighter than steel in general. However, poorly made aluminum cases can cut your fingers when you work in them because aluminum is generally thinner and sharper than steel. Steel is generally a stronger material and consequently it will be heavier.
Case Choice
Case choice should be based on the looks, the airflow, cable management, and space. You must be able to have enough space for your components in your case. Whether you need space for an unorthodox motherboard form factor or an extra long graphics card, you need to be able to put the components in. Cable management makes the computer easier to work with, and can enhance airflow. Airflow is important in a case because without good airflow, the air in the case remains stagnant and the temperatures of your parts will increase.
Some companies that generally produce good, solid cases are Silverstone, Antec, Corsair, Cooler Master, Lian Li
+ Show Spoiler [HSF] +
Generally an aftermarket HSF is not needed if you do not plan on overclocking. Purchase one only if you want a silent computer or are planning on overclocking.
Lower End:
Corsair A50
Cooler Master Hyper 212+
Cooler Master Hyper N
Scythe Mugen Rev B
Cogage True Spirit
High End:
Prolimatech Megahalems
Prolimatech Super Mega
Thermalright TRUE
Cogage Arrow
Corsair A70
Corsair H60
Corsair H70
Antec Kuhler
Noctua NH-D14
Thermal Compounds
When talking about cooling, many people get overeager about thermal compounds. Thermal compounds fill the gaps of air between the surface of the CPU and the surface of the heatsink. While thermal compound is extremely important, it is generally not as important as the heatsink and fan configuration itself. Good thermal compounds cost about $10 for a tube, and can probably help your temperatures by 2-3C. Unless you are doing extreme overclocking, such meaningless expenses should be shaved off and invested in a better motherboard or GPU. If you buy a high end heatsink like the Megahalem or H50, they will normally come packaged with very good thermal compounds so there is no need to buy more.
Lower End:
Corsair A50
Cooler Master Hyper 212+
Cooler Master Hyper N
Scythe Mugen Rev B
Cogage True Spirit
High End:
Prolimatech Megahalems
Prolimatech Super Mega
Thermalright TRUE
Cogage Arrow
Corsair A70
Corsair H60
Corsair H70
Antec Kuhler
Noctua NH-D14
Thermal Compounds
When talking about cooling, many people get overeager about thermal compounds. Thermal compounds fill the gaps of air between the surface of the CPU and the surface of the heatsink. While thermal compound is extremely important, it is generally not as important as the heatsink and fan configuration itself. Good thermal compounds cost about $10 for a tube, and can probably help your temperatures by 2-3C. Unless you are doing extreme overclocking, such meaningless expenses should be shaved off and invested in a better motherboard or GPU. If you buy a high end heatsink like the Megahalem or H50, they will normally come packaged with very good thermal compounds so there is no need to buy more.
+ Show Spoiler [HDD/SSD] +
Why the Velociraptor Sucks
At the moment, you pay a high premium to get the 10,000 RPM of the Velociraptor, but the increases in random access, and sequential read/write do not justify the premium. A Velociraptor might have an access time that is at best twice as fast as a traditional HDD, but it costs more than 20 times as much per GB. A SSD's access time will be 100 times as fast as a traditional HDD, but will cost around 35 times as much per GB. The sequential performance of the SSD will also be far superior to the performance of the Velociraptor.
7200RPM vs 54M00RP
Pretty simple, the faster the HDD spins the faster the HDD will read, write, or access. For your system drive (the drive you install your OS and applications on) you want it to run at 7200RPM because it will improve drive performance. For storage drives when speed isn't generally an issue, 5400RPM is acceptable, even preferred, for less power consumption and heat.
Random Performance SSD vs HDD
A SSD excels in random performance. A SSD can access a file in 0.1ms while an HDD can take anywhere from 4ms to 13ms. This allows the SSD to process many different requests faster than the HDD. Launching a program like Chrome or Starcraft II is dominantly random and opening 5 programs at once may stress an HDD, but a SSD will do it near instantly due to the high random reads, and low random access times. If you have ever had a computer start slowly, you have seen the result of the HDD being stressed. Each program takes a long time to load, and does not respond to clicks. With a SSD, the system start will be near instantaneous and you will not experience the slowdown that a HDD would cause.
RAID 0
RAID 0 is often stated as a way for HDDs to equal SSDs in performance or simply just to increase performance. RAID 0 is very much a double edged sword. While it does increase your sequential read and write speeds, your data is split across two hard drives which means that if one fails, you lose the information on both. While RAID 0 increases sequential read and write, it does not increase random read/write or random access, which are the points where the SSD is the strongest. Sequential reads and writes are not all that useful compared to random access and random read/write because most of your actions are random.
RAID 0 with SSDs also fairly pointless. While you do gain performance and you don't suffer from potential data loss, you do lose access to TRIM (on non-Intel motherboards). SSDs at the moment still need TRIM to keep performance high. Without TRIM, your SSD's write speeds will generally decrease over time.
xmShake has informed me that Intel motherboards can now support TRIM on RAID 0 and RAID 1 SSD configurations!
At the moment, you pay a high premium to get the 10,000 RPM of the Velociraptor, but the increases in random access, and sequential read/write do not justify the premium. A Velociraptor might have an access time that is at best twice as fast as a traditional HDD, but it costs more than 20 times as much per GB. A SSD's access time will be 100 times as fast as a traditional HDD, but will cost around 35 times as much per GB. The sequential performance of the SSD will also be far superior to the performance of the Velociraptor.
7200RPM vs 54M00RP
Pretty simple, the faster the HDD spins the faster the HDD will read, write, or access. For your system drive (the drive you install your OS and applications on) you want it to run at 7200RPM because it will improve drive performance. For storage drives when speed isn't generally an issue, 5400RPM is acceptable, even preferred, for less power consumption and heat.
Random Performance SSD vs HDD
A SSD excels in random performance. A SSD can access a file in 0.1ms while an HDD can take anywhere from 4ms to 13ms. This allows the SSD to process many different requests faster than the HDD. Launching a program like Chrome or Starcraft II is dominantly random and opening 5 programs at once may stress an HDD, but a SSD will do it near instantly due to the high random reads, and low random access times. If you have ever had a computer start slowly, you have seen the result of the HDD being stressed. Each program takes a long time to load, and does not respond to clicks. With a SSD, the system start will be near instantaneous and you will not experience the slowdown that a HDD would cause.
RAID 0
RAID 0 is often stated as a way for HDDs to equal SSDs in performance or simply just to increase performance. RAID 0 is very much a double edged sword. While it does increase your sequential read and write speeds, your data is split across two hard drives which means that if one fails, you lose the information on both. While RAID 0 increases sequential read and write, it does not increase random read/write or random access, which are the points where the SSD is the strongest. Sequential reads and writes are not all that useful compared to random access and random read/write because most of your actions are random.
RAID 0 with SSDs also fairly pointless. While you do gain performance and you don't suffer from potential data loss, you do lose access to TRIM (on non-Intel motherboards). SSDs at the moment still need TRIM to keep performance high. Without TRIM, your SSD's write speeds will generally decrease over time.
xmShake has informed me that Intel motherboards can now support TRIM on RAID 0 and RAID 1 SSD configurations!
+ Show Spoiler [Overclocking] +
Overclocking is the act of pushing a component of the computer past what it is normally rated to operate at. CPU, RAM, GPU, and GPU RAM can all be overclocked past the normal clock rates, and under many circumstances, the GPU and GPU memory can be pre-overclocked when you buy it. CPU/RAM and GPU/GPU RAM overclocking are both quite different. As you are pushing the component beyond what it is rated for, you never know what to expect when you attempt to overclock. No two pieces of silicon are exactly the same so while somebody may be able to get 4GHz out of their i7, that does not mean you can achieve the same results in the fashion they did. Because overclocking pushes your components further, they will require more power, and in turn they will produce more heat. As such, overclocking will require a good stable motherboard, a stable PSU, and in most cases, an aftermarket heatsink and fan. The case can also be important in controlling the temperatures (especially with GPU).
CPU/RAM
CPU and RAM overclocking depends upon the CPU, RAM, motherboard, heatsink, and PSU. The CPU and RAM clock rates depend upon 3 core values on Intel boards (I can't give advice on AMD, though it should be very similar). Your CPU speed is determined by the BCLK (baseclock), and the CPU multiplier. Your memory clock depends upon the BCLK and the RAM multiplier. The BCLK is typically measured in MHz, and your CPU speed is the product of the BCLK and CPU multiplier. Your RAM speed is the product of the BCLK and the RAM multiplier. As such, raising your BCLK will raise both the speed in which the CPU operates at and the speed in which the RAM operates at.
A typical Intel CPU like the i7 860 has a BCLK of 133, a CPU multiplier of 21, and a default RAM multiplier of 10 which means that the CPU will operate 133MHz * 21 (2.80GHz), and the RAM will operate at 133MHz * 10 (1333MHz). If I changed the BCLK to 160, the CPU would operate at 160MHz * 21 (3.36GHz), and the RAM at 160MHz * 10 (1600MHz).
In most CPUs, you will not be able to change the CPU multiplier, only the BCLK and RAM multiplier. The exceptions are the Intel K series processors, Extreme Edition processors, and the AMD Black Edition processors. Those CPUs have unlocked multipliers and are generally easier to overclock because it is possible to overclock just by changing the CPU multiplier without changing the BCLK. It also offers more flexibility in the ways you can overclock a CPU.
If you want to learn more about overclocking, there are many guides online that tell you about how you can begin. If you are curious, you can enter your BIOS before Windows boots and change the BCLK by 5MHz either up or down. It shouldn't affect your stability and you'll be able to see the results of your overclock (or underclock) in CPU-Z.
I'm not going to get into voltages here. There are better articles out there for that.
After your overclock, you should download a program called Prime95. The purpose of Prime95 is to stress all of your cores. Prime95 uses all of your cores and pushes them to 100% load. If after 8-10 hours of Prime95 running, you still have not gotten an error, then your overclock can be considered stable. If however, Prime95 shows that your CPU made an error, it is time to go back to BIOS and adjust your clock!
For memory, you can change the clock through the BCLK/RAM multiplier or you can attempt to change the timings on the RAM. The timings as mentioned above are a measure of time in clock cycles. The lower your latency, the faster your memory can be accesed and used. As for memory, you can use Memtest86+ to stress your memory. If you get a errors or a BSOD then your memory timing/overclock is too aggressive.
Update 1/25/2011 LGA 1155 and future Intel CPU Overclocking
As of LGA 1155, all new Intel CPUs feature a unified base clock which is tied to every other frequency that is controlled by the CPU. What this means is by changing the base clock on a second gen Core i3/i5/i7 CPU, you not only change the memory frequency and the CPU frequency (which is what it was limited to for the most part before), you also change much more sensitive frequencies like the PCIe frequency. The more sensitive frequencies can not handle the overclocking that the memory frequency or CPU frequency can handle.
As such, the only way to overclock new Intel CPUs is to buy an unlocked CPU (denoted by the K suffix at the end of the CPU's name), along with a chipset that supports overclocking. P67 currently supports overclocking, but H67 does not.
When overclocking Intel CPUs, it will simply be a matter of changing the CPU multiplier and the memory multiplier instead of the base clock of the CPU.
GPU/GPU RAM
As with CPU and RAM, you can overclock your GPU/GPU RAM, but overclocking the GPU is generally much easier and is done through software rather than BIOS. Many graphic cards come with a CD that includes software from the company which can be used to overclock the GPU. Generally, GPU/GPU RAM overclocking is as simple as moving a slider. Many programs exist that allow you to overclock your GPU, and most of them are based off of RivaTuner. Many companies have their own software bundled with the GPU, EVGA has EVGA Precision, ASUS has Smart Doctor, MSI has MSI Afterburner, etc. Generally they serve the same purpose and many are cross compatible with different GPUs, though some features may not work with all GPUs.
Unlike with CPUs, often times you do not need to replace the heatsink on your GPU in order to get a decent overclock. Frequently, the reference designs provided by ATI/Nvidia will be sufficient for mild overclocking, and after the graphics card has matured a bit, you will often find non-reference designs that can cool more efficiently. Still, a limiting factor of GPU overclocking is still the heat, and you want to monitor the temperatures that your GPU reaches. Though temperatures in the high 80s and low 90s aren't exactly uncommon, it would be best to avoid those temperatures for the safety of your GPU.
However you still do need to check the stability of your overclock! Fortunately, there is a great tool similar to Primers for the GPU called FurMark which will stress your GPU to 100% by drawing fur. If you see 'artifacts', objects on the display that do not belong like black strips and flashing colors, your GPU overclock is unstable and needs to be lowered. If however, after around 15-20 minutes, you do not see any artifacts and your GPU is sufficiently cool, your overclock can be considered stable.
CPU/RAM
CPU and RAM overclocking depends upon the CPU, RAM, motherboard, heatsink, and PSU. The CPU and RAM clock rates depend upon 3 core values on Intel boards (I can't give advice on AMD, though it should be very similar). Your CPU speed is determined by the BCLK (baseclock), and the CPU multiplier. Your memory clock depends upon the BCLK and the RAM multiplier. The BCLK is typically measured in MHz, and your CPU speed is the product of the BCLK and CPU multiplier. Your RAM speed is the product of the BCLK and the RAM multiplier. As such, raising your BCLK will raise both the speed in which the CPU operates at and the speed in which the RAM operates at.
A typical Intel CPU like the i7 860 has a BCLK of 133, a CPU multiplier of 21, and a default RAM multiplier of 10 which means that the CPU will operate 133MHz * 21 (2.80GHz), and the RAM will operate at 133MHz * 10 (1333MHz). If I changed the BCLK to 160, the CPU would operate at 160MHz * 21 (3.36GHz), and the RAM at 160MHz * 10 (1600MHz).
In most CPUs, you will not be able to change the CPU multiplier, only the BCLK and RAM multiplier. The exceptions are the Intel K series processors, Extreme Edition processors, and the AMD Black Edition processors. Those CPUs have unlocked multipliers and are generally easier to overclock because it is possible to overclock just by changing the CPU multiplier without changing the BCLK. It also offers more flexibility in the ways you can overclock a CPU.
If you want to learn more about overclocking, there are many guides online that tell you about how you can begin. If you are curious, you can enter your BIOS before Windows boots and change the BCLK by 5MHz either up or down. It shouldn't affect your stability and you'll be able to see the results of your overclock (or underclock) in CPU-Z.
I'm not going to get into voltages here. There are better articles out there for that.
After your overclock, you should download a program called Prime95. The purpose of Prime95 is to stress all of your cores. Prime95 uses all of your cores and pushes them to 100% load. If after 8-10 hours of Prime95 running, you still have not gotten an error, then your overclock can be considered stable. If however, Prime95 shows that your CPU made an error, it is time to go back to BIOS and adjust your clock!
For memory, you can change the clock through the BCLK/RAM multiplier or you can attempt to change the timings on the RAM. The timings as mentioned above are a measure of time in clock cycles. The lower your latency, the faster your memory can be accesed and used. As for memory, you can use Memtest86+ to stress your memory. If you get a errors or a BSOD then your memory timing/overclock is too aggressive.
Update 1/25/2011 LGA 1155 and future Intel CPU Overclocking
As of LGA 1155, all new Intel CPUs feature a unified base clock which is tied to every other frequency that is controlled by the CPU. What this means is by changing the base clock on a second gen Core i3/i5/i7 CPU, you not only change the memory frequency and the CPU frequency (which is what it was limited to for the most part before), you also change much more sensitive frequencies like the PCIe frequency. The more sensitive frequencies can not handle the overclocking that the memory frequency or CPU frequency can handle.
As such, the only way to overclock new Intel CPUs is to buy an unlocked CPU (denoted by the K suffix at the end of the CPU's name), along with a chipset that supports overclocking. P67 currently supports overclocking, but H67 does not.
When overclocking Intel CPUs, it will simply be a matter of changing the CPU multiplier and the memory multiplier instead of the base clock of the CPU.
GPU/GPU RAM
As with CPU and RAM, you can overclock your GPU/GPU RAM, but overclocking the GPU is generally much easier and is done through software rather than BIOS. Many graphic cards come with a CD that includes software from the company which can be used to overclock the GPU. Generally, GPU/GPU RAM overclocking is as simple as moving a slider. Many programs exist that allow you to overclock your GPU, and most of them are based off of RivaTuner. Many companies have their own software bundled with the GPU, EVGA has EVGA Precision, ASUS has Smart Doctor, MSI has MSI Afterburner, etc. Generally they serve the same purpose and many are cross compatible with different GPUs, though some features may not work with all GPUs.
Unlike with CPUs, often times you do not need to replace the heatsink on your GPU in order to get a decent overclock. Frequently, the reference designs provided by ATI/Nvidia will be sufficient for mild overclocking, and after the graphics card has matured a bit, you will often find non-reference designs that can cool more efficiently. Still, a limiting factor of GPU overclocking is still the heat, and you want to monitor the temperatures that your GPU reaches. Though temperatures in the high 80s and low 90s aren't exactly uncommon, it would be best to avoid those temperatures for the safety of your GPU.
However you still do need to check the stability of your overclock! Fortunately, there is a great tool similar to Primers for the GPU called FurMark which will stress your GPU to 100% by drawing fur. If you see 'artifacts', objects on the display that do not belong like black strips and flashing colors, your GPU overclock is unstable and needs to be lowered. If however, after around 15-20 minutes, you do not see any artifacts and your GPU is sufficiently cool, your overclock can be considered stable.
+ Show Spoiler [Useful Software] +
General System Monitoring
HWMonitor
SpeedFan
RealTemp
CPU
CPU-Z
Prime95
SuperPI
IntelBurnTest
GPU
GPU-Z
FurMark
(Choose One) MSI Afterburner, EVGA Precision, RivaTuner,
Catalyst Control Center (ATI)
Memory
Memtest86+
HDD
DiskCheckup
HD Tach
Benchmarking
FurMark
SuperPI
Unigine
3D Mark
Fraps
Futuremark
and others so google
HWMonitor
SpeedFan
RealTemp
CPU
CPU-Z
Prime95
SuperPI
IntelBurnTest
GPU
GPU-Z
FurMark
(Choose One) MSI Afterburner, EVGA Precision, RivaTuner,
Catalyst Control Center (ATI)
Memory
Memtest86+
HDD
DiskCheckup
HD Tach
Benchmarking
FurMark
SuperPI
Unigine
3D Mark
Fraps
Futuremark
and others so google
+ Show Spoiler [Websites that RULE] +
Tech sites
http://www.anandtech.com
mahnini and FragKrag agree that anandtech is best ez (come here for everything, especially SSDs)
http://www.hardwaresecrets.com
http://www.jonnyguru.com
http://www.xbitlabs.com
http://www.hardocp.com
http://www.bit-tech.net
http://www.tomshardware.com
http://www.notebookreviews.com
Laptops
http://www.fudzilla.com
rumors and shit
http://www.frostytech.com
http://www.hardware-revolution.com/
buy stuff from
USA/CANADA
http://newegg.com
(http://newegg.ca)
http://woot.com
http://NCIX.com (canada?)
http://tigerdirect.com
Fry's Electronics (http://www.frys.com)
Microcenter (http://www.microcenter.com)
Europe (maybe?)
http://www.pixmania.com
http://www.komplett.ie
http://www.ebuyer.co.uk
http://www.hardwareversand.de
http://www.mindfactory.de
Australia
http://pccasegear.com.au
Hong Kong
http://www.jumbo-computer.com/pricelist.asp
http://www.anandtech.com
mahnini and FragKrag agree that anandtech is best ez (come here for everything, especially SSDs)
http://www.hardwaresecrets.com
http://www.jonnyguru.com
http://www.xbitlabs.com
http://www.hardocp.com
http://www.bit-tech.net
http://www.tomshardware.com
http://www.notebookreviews.com
Laptops
http://www.fudzilla.com
rumors and shit
http://www.frostytech.com
http://www.hardware-revolution.com/
buy stuff from
USA/CANADA
http://newegg.com
(http://newegg.ca)
http://woot.com
http://NCIX.com (canada?)
http://tigerdirect.com
Fry's Electronics (http://www.frys.com)
Microcenter (http://www.microcenter.com)
Europe (maybe?)
http://www.pixmania.com
http://www.komplett.ie
http://www.ebuyer.co.uk
http://www.hardwareversand.de
http://www.mindfactory.de
Australia
http://pccasegear.com.au
Hong Kong
http://www.jumbo-computer.com/pricelist.asp
Made with Google Wave
RIP ;_;
[/spoiler]