Learn About Desktop GPUs

What is it?

It stands for graphics processing unit (GPU). It is a processor that is specialized in computing graphics very fast. With a GPU in your computer, you will be able to run graphic intensive programs smoothly such as 3D games. Lately, GPUs have been specializing in many more things like accelerating the performance of many programs that can take advantage of the GPU's highly parallel structure (Embarrassingly Parallel), and deep learning neural networks.

How does it work?

GPUs work by doing the following (over simplified) steps:

1. Receive instructions from CPU
2. Execute instructions (Graphics or Compute)
3. Send graphics to display or store compute results in memory

Look for

• General Benchmarks

  Check out general benchmarks so that you can get a feeling of how the GPU performs

  One example of such benchmarks is Geekbench

• Specific Benchmarks

  Check the benchmarks that are specific to the game or software that you will be using

  Let's say you only play 1 game, and that is the only thing you do on your computer, then only check how the GPU performs specifically to the game you intend on playing. If you found a GPU runs your game to your desired level, then no need to over spend on a faster GPU.

• Compatibility

  Check Motherboard PCIe, power supply wattage, PC case clearance, and display ports.

• Display Resolution

  Check if GPU performance with the display resolution

  Displays have many different resolutions based on their size, use, and quality. If you have a display with 1080p resolution, and you have a powerful GPU, your GPU will not be fully utilized because it is overpowered for 1080p resolution. You need a display with higher resolution to take advantage of your powerful GPU. The best thing to do is to make sure the GPU can handle your display.

• Display Refresh Rate

  Check if GPU performance can take advantage of the display refresh rate

  Displays have many different refresh rates, most commonly 60Hz. Higher refresh rates are mainly for better gaming experience. If you have a display with 120Hz refresh rate, you need to make sure you get a GPU that can run your games at your display's native resolution and native refresh rate.

• Power Consumption

  Choose a GPU with the best performance per Watt for your requirements

  Less power consumption mean less money paid on electricity bill. It also makes cooling the GPU easier and quieter.

• Noise Level

  Make sure the GPU has acceptable noise levels

  This is directly correlated with power consumption and cooling headroom. You need less power consumption and more cooling headroom to reduce noise levels.

Avoid

• Future Proofing

  Do not buy an overpowered GPU for your needs just for future proofing

  Let's say you got the top-of-the-line GPU X today for $1500, and it performs 20% more than the mid-range GPU Y that cost today $500. The price of GPU X is 200% more then GPU Y, yet only 20% faster. Later next year, the mid-range GPU Z performs the same as the top-of-the-line GPU X from last year with newer features yet costing third at $500. This would make your GPU X depreciate in value considerably. If you were to get the mid-range GPU every year it would cost much less, and you will have access to newer features. In addition, if you happen to sell your old GPU for a good price, you will save more. In short, your performance/$ is much less (diminishing returns) if you buy top-of-the-line. However, if top-of-the-line is what it takes to cover your needs, then you have no choice but to get it.

• Overclock

  Avoid overclocking your GPU

  For the most part, overclocking your GPU will give you negligible results, while increasing power consumption dramatically and could void your warranty.

• Spec Comparisons

  Do not compare GPUs based on their specs

  Every GPU is based on an architecture, and every architecture is designed differently. For example, a 1280 Shaders GPU can be faster than 2048 Shaders GPU, hence, it is pointless to say that the 2048 Shaders is more than 1280 Shaders, so it must be better. Same thing with cache size, and any other metrics. The best thing to do is check the benchmarks and real world use cases.

• Clock Speeds Comparisons

  Do not compare GPUs based on their clock speeds

  All modern GPUs adjust their clock speeds based on usage and power/cooling headroom to save on power and maximize performance. As a result, if you have a fast GPU clocked at 1.5GHz, but your power/cooling headroom is only enough for it to run sustainably at 1GHz, then it will run at 1GHz, even though it is a 1.5GHz GPU. Also, a GPU from architecture X with clock speed of 1GHz can out perform a GPU from architecture Y running at 1.5GHz. Instead, you should check out how much your GPU power/cooling headroom is.

• Memory Size Comparisons

  Do not compare GPUs based on their memory size

  Usually, modern GPUs come with an appropriate GPU memory size relative to their performance level. Unfortunately, some manufacturers combine less powerful GPUs with more memory to deceive consumers with the assumption that more memory means more performance. It is rarely the case where getting more memory will get you more performance. Instead, focus on GPU benchmarks, not memory size.

• Multi GPU Gaming

  Do not get more than 1 GPU for gaming

  Most games do not take advantage of multiple GPUs. On top of that, the games that do support it do not improve their minimum FPS, which can make the games stutter.

• Bottlenecks

  Make sure there is no bottleneck from your CPU

  Let's say you have the fastest GPU and a slow CPU, your GPU will not work at full capacity because it will not receive instructions from the CPU fast enough to keep it busy. Because of that, the GPU will be mostly idle.