The current graphics cards are composed of several “basic” components arranged on a piece of printed circuit forming the card itself (the PCB for Printed Circuit Board). We find in particular the graphics processor (the GPU for Graphics Processing Unit), a number of memory chips (between 0 and 24). There are also connectors to connect the displays, power connectors if the card requires a lot of power, the PCIe port used to connect the graphics card to the motherboard and a number of other components serving as floor power.

Note: Sometimes the terms GPU and best graphics cards for gaming are used, wrongly, as synonyms, it is a shortcut very often used in the specialized press, including by your servant.

The graphics card memory is usually GDDR5, but there are still some low-end cards equipped with GDDR3, or worse DDR3. The latter is absolutely not suitable for the GPU, especially in terms of speed, but remains less expensive than the GDDR5. The latest high-end AMD best graphics cards for gaming (the Fury series) have HBM memory replacing the commonly used GDDR5. This type of memory is directly included in the GPU, hence the lack of memory chips on the PCB.

The whole thing is capped by the cooling system. It can be passive (for cards with little power), but most often active (associated with one or more fans, ed). There are very high-end graphics cards equipped with Watercooling or Hybrid Air / Water cooling systems.

The board power is shared between the PCIe port (which delivers up to 75 W) and the additional connectors at the end of the board. These connectors are either 6 pins (max 75 W) or 8 pins (max 150 W). For example, to have a card of 300 W, it is enough of a 6 pins and an 8 pins (75 W by the PCIe port + 75 W on the 6 pins, + 150 W on the 8 pins = 300 W) .

The multi GPU, SLI and CrossFireX

It is possible to use several best graphics cards for gaming in a computer, they will work either independently (this is the case of Mac Pro OSX for example) or “together”. To run multiple graphics cards “together” it is necessary to go through technologies like SLI at nVidia or CrossFireX at AMD.

These technologies are unfortunately only available on Windows. Thus, a Mac Pro on OSX has two independent AMD graphics cards, while when running on Windows (via Bootcamp) its two graphics cards can work in CrossFireX (because it is not a requirement, the option is activatable / can be deactivated in the card control center).

These two technologies work in much the same way. You need 2, 3 or 4 times the same model of graphics card, and a motherboard supporting this kind of configuration. Of course, you need a Windows computer running DirectX. In the case of old AMD cards and all nVidia cards, it is necessary to connect all the cards together using a connector: the SLI / CrossFire bridge. Since two generations of cards, AMD no longer uses a bridge, in this case it is the PCIe bus which is in charge of the communication between the cards.

If you have four SLI / CrossFireX cards each with 4GB of RAM, you’ll get … 4GB of RAM, that’s great, is not it? And yes, because each card contains the same information in RAM, that’s how it works up to DirectX 11, because DirectX 12 will give more freedom to developers on this point.

Having multiple graphics cards is not necessarily always better, but the raw calculation will be faster, but there may be a number of small bugs (the higher the number of cards, the more bugs there are) can become embarrassing. It is also necessary to have updated and optimized drivers for EACH software in which one wants to use several GPUs. However, the SLI and CrossFireX have the merit of being “centralized” in the driver, so it is not up to the developers of each application to design themselves this kind of optimization (developers at Apple, do you understand where I want to come here?!).