Buying a Video CardI. Introduction – What is a video chip and where is it at?
Video processing or the Graphics Processing Unit (GPU), comes in two forms. One is located in the northbridge of a motherboard. The other is located in a dedicated video card typically plugged into the PCI-Express x16 slot of a motherboard.
a. Onboard Chipset
This form of GPU has historically been very poor at doing anything other than watching videos through a browser, DVD playback, or other light uses where something more powerful is not required. However, in recent years, these chips have become much more powerful, some can play back blu-ray movies and even games. This is due to modern GPUs being more powerful then ones in the past, and game engines being older then the GPUs available.
b. Dedicated Video Card
A dedicated video card has much more uses. In addition to everything an onboard chip can do, dedicated graphics cards can handle much more demanding games, HD streaming video, 3d modeling and drawing, rendering, video creation and editing, and blu-ray playback with all of its features.
II. Stream Processors
Video cards typically contain what is known as stream processors. They act as the “army” of the video card. The more “soldiers” you have the more powerful and quicker a video card can do things. In reality, stream processor act as the processor or CPU of the video card where each of them works together to calculate things and perform specific tasks that you require of them. This is what is called parallel computing or processing.
Typically video cards have much faster memory then what a normal motherboard contains, and acts in just the same way. It is a very fast way or storing and reading things that the card’s stream processors use. What memory mostly effects is textures, those being everything you see on the game. Each brick, tire, or piece of trash is part of a texture, the more ram you have the more textures you can load into the video cards ram, and the faster that ram is the quicker it can display it on the screen. In some cases, more memory allows you to display much larger textures resulting in a more detailed image.
Video cards can be limited by several factors. Certain things can limit the performance of a video card while others can drastically helping a card last much longer then intended.
Video cards used to come in 2 older forms, PCI slots and AGP slots. Both of these have been replaces by PCI-Express Slots which operate at varying speeds, but typically the x16 speeds are reserved for dedicated video cards.
PCI-Express slots have different versions, those being:
· 1.0 – Initial release, contains a transfer speed of 2.5 billion data transfers per second (2.5 GT/s)
· 1.1 – Moderate fixes
· 2.0 – Double the bandwidth of its 2.0 counterpart (5 GT/s)
· 2.1 – Same bandwidth as v2.0 but has the power advantages of 3.0 slots
· 3.0 – Bandwidth of 8 GT/s and removal of encoding overhead
Keep in mind that a 2.0 slot video card will work perfectly fine when used in a v1.0 slot or a v3.0 slot. All of them are compatible, but most modern motherboards only support v2.0, but older ones that have not been upgraded in a while will be v1.0. You cannot even buy a v3.0 card until late 2010 or early 2011, and there will not be a motherboard to support it until then.
Graphics chips come in different versions. Let’s suppose it is a year from now and you are buying a video card upgrade. If you buy a v3.0 card for your now v2.0 motherboard, when you later build a new machine, the old video card in the new slot will operate much better because it is no longer limited by the slot bandwidth. You may be surprised at how well it performs and get a few months of life to allow for that new release of video cards or price drop.
Memory bandwidth is a 3 tiered system currently, and will eventually have more and more tiers as technology progresses. Right now the three tiers are 256-bit memory, 128-bit memory, and 64-bit memory. 256-bit has 2x the bandwidth of the 128-bit memory and 128-bit memory has 2x the bandwidth of 64-bit memory. These bits refer to the size of the data chunk capable of being sent and stored on the GPUs ram.
A video card developer will release their top tier card, containing all the bells and whistles, the highest amount and quickest memory, highest amount of stream processors, and highest memory bandwidth of that line of graphics cards. After those cards are released 2 things will happen, they take things out and lower the power requirements of newer video cards by releasing versions with slower and less memory, less stream processors, and lower memory bandwidth to allow those cards to sell for less money. They will also take chips from the top performing single GPU cards and place two of them on a single board, almost doubling the performance and the price.
One thing that limits some users selection is the physical dimensions of a video card, in particular its length. Video cards come in two forms, single slot and dual slot. This refers to the mounting bracket required which is used to supply extra room inside the case for the video cards expanded cooling. Most high performing video cards will be dual slot design.
V. Multiple GPUs – SLI / Crossfire
There are two ways to accomplish having multiple GPUs, the first is to buy a single card with two GPUs on one PCB or board and the second is to purchase multiples of the same card, typically same vendor, and connect them on the motherboard through SLI cables (nvidia) or Crossfire cables (AMD/ATI). In either case you will not get the same performance, but there are differences and advantages to each.
a. One Board Scenario
The main difference between one board and multiple boards is that the two GPUs are not limited to a connector speed. The two GPUs can talk to each other on the same board without bandwidth limitations. This results in slightly better performance when compared to two cards setup in SLI/Crossfire, but also has its own drawbacks. Heat is a much bigger issue when you have two high powered GPUs right next to each other as opposed to with dedicated coolers on each. This reduces or limits overclocking ability and may result in liquid cooling being necessary in some instances.
b. Multiple Card Scenario
In the second case, you have distance between each card, typically not very much, but each card still has its own heatsink and fan. The main advantage of SLI/Crossfire is that the limit to the number of GPUs you can use is double that of the single board method. You can go crazier by mixing the two methods and having two boards with multiple GPUs in SLI/Crossfire. Unfortunately there is a catch. Typically with most motherboard, when you use more than one card the bandwidth of the PCI-Express slots is limited. Typically a x16 single card scenario will become two x8 slots sharing the available bandwidth between each other. However, in high end motherboard, sever x16 slots are used and the full speed of a card can be utilized.
Most middle and high end cards require a connector from the power supply to be directly connected to the video card. There are three varieties, but only two are now used, 6-pins and 8-pins.
When buying a power supply you will need to make sure you have these, and when buying a video card you will need to check what the power requirements of a particular graphics card is and compare it with your current systems power output and availability. If you do not have these connectors it does not mean you can use higher end graphics cards, simply that you will need to buy a 4-pin Molex to 8-pin/6-pin adapter.
Simply having available connectors does not mean that you can run any graphics card. You need to have the wattage as well. Going by modern PSUs, you typically need 650 watts or better to be able to have SLI/Crossfire capable setup. I have even found in some instances, 650 watt PSUs or better are the only ones that have enough of these connectors to allow for SLI/Crossfire.