What is a Graphics Card? How Do You Choose a Good One?

If you’re anything like us, then the graphics card is your favorite part of a computer. Graphics cards let your computer do awesome things like super complex computations, physics processing, and most importantly, producing shiny graphics in games.

So, what is a graphics card, and how do you figure out which one you should buy?

A graphics card is basically a mini computer inside your computer. It consists of the Graphics Processing Unit (GPU) and supporting hardware, including its own power supply, RAM, and heatsink.

Similarly to how the CPU is the brain of the computer, the GPU is the brain of the graphics card. The GPU is a specialized electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images.

In short, it makes these pretty pictures on your screen.

When people think of GPUs, they usually think of two major manufacturers: NVIDIA, and AMD. Intel is a third GPU manufacturer, because they integrate GPUs into their CPUs. There are also a bunch of other manufacturers that make GPUs for mobile phones, tablets, and embedded systems.

But for our purposes, we’re going to be discussing discrete GPUs, meaning the dedicated graphics cards you buy if you want to do really fun stuff with your PC. When it comes to discrete graphics cards, the two big names are NVIDIA and AMD.

Here in our office, we’ve pulled out two graphics cards from AMD’s venerable 200 series. We’ve got a Radeon R7 270X, and an R9 290.

Both AMD and NVIDIA launch new generations of graphics cards every one to two years. With every new generations, they offer a bunch of different cards at a range of power levels and price options. Let’s take a closer look so that we can discuss some important numbers that will help you compare graphics cards.

First, let’s discuss the R7 270X. It has 2 GB of video RAM, 1,280 stream processors, and a clock speed of 1,050 megahertz. It’s currently considered a solid mid-range graphics card.

Next, the R9 290. It has 4 GB of video RAM, 2,560 stream processors, and a clock speed of 947 megahertz. This is currently considered an upper-mid-range graphics card.

So, video RAM, stream processors, clock speed… what do these numbers all mean? Should you understand the specs, or just go with the most expensive graphics card you can afford?

Generally speaking, the best way to decide the right graphics card for you is to look at real-world gaming benchmarks, but we’ll get that in a few minutes. For now, let’s talk specs.

First, video RAM, also known as VRAM. This is usually the first spec you’ll notice when shopping around for video cards, and often it’s the only spec that people really pay attention to. This is a bit of a mistake, because while VRAM is essential, having more than you need doesn’t increase performance at all.

Video RAM is mainly important for loading the textures and images that make up the game you’re playing on your screen. All of that data takes up VRAM, and it gets especially taxing on VRAM as you increase your game resolution, which we’ll also touch on in a minute.

If your game begins to load more data than your VRAM can handle, your PC moves the extra data over to your normal RAM.

Sharing is great, right? Well, not always. Sharing is bad when it comes to VRAM trying to share data with system RAM.

Any time you’re playing a game and your graphics card runs out of available VRAM, it causes huge hits to performance. If you’ve ever been playing a game that turns in – to – slide – show – mode, your graphics card may not have had enough VRAM.

There are a few factors that especially impact your VRAM allocation: Resolution, anti-aliasing, and texture quality. Basically, the higher these settings are, the more VRAM you’re going to need. That’s why the higher-end cards pack in larger amounts of VRAM. If you even want to consider gaming at 4K resolution, you’re going to want a lot of VRAM.

Oh, right. We should probably talk about resolution.

Every computer display is made up of tiny elements called pixels. Your resolution determines the number of pixels that will appear on your screen. More pixels equals more visual detail, resulting in crisper, clearer images.

Here’s a size comparison of common computer resolutions:

 

resolution

The standard resolution for modern PC gaming is 1920×1080, commonly known as 1080p. For a slightly bigger pixel count, an increasingly popular resolution is 2560×1440, or 1440p. Currently, at the upper end of resolutions for consumer-grade computer screens is 3840×2160, or ultra-high definition 4K. You might be able to tell from this visual that 4K resolution packs in four times as many pixels as 1080p, resulting in about four times the demand on your graphics card while playing PC games.

 

Producing large numbers of game pixels at resolutions like 4K demands an enormous amount of VRAM, meaning you’re going to need a high-end graphics card to play at 4K resolutions. In short: 2 GB is enough to play most PC games at 1080p, 4 GB is better for playing really demanding games at 1080p or most games at 1440p, and you’ll probably want 6 or more GB of VRAM to even attempt a demanding PC game at 4K resolution, unless you don’t mind your framerate resembling a slideshow.

These are just rough guidelines. Newer, faster memory technologies such as High Bandwidth Memory, or HBM, will help alleviate the need for more memory in the future.

Stream processors (AMD) or CUDA cores (NVIDIA) are the processors in the GPU that perform the computations to produce images on the screen. They can also be used to do other types of complex mathematics and number-crunching, such as mining cryptocurrency or modeling weather patterns.

GPU clock speed refers to the speed of the GPU. Generally speaking, higher speeds mean that the card performs tasks faster and produces a higher framerate when gaming. But like with CPUs, the clock speed is relative to each family of GPUs, so you can’t compare the clock speed of this GPU and assume it will get things done faster or slower than a GPU from NVIDIA with a different clock speed.

To better understand a GPU’s performance and qualities, you should check all of its specs, and there are a lot of them.

But if you want to get a very rough estimate of how one GPU performs compared to others, you can multiply the number of stream processors by the clock speed. So, for example, if we multiply the stream processors and clock speed of the 270X, we get (1,280 x 1050) 1.3 billion. Compare that to the 290, at (2560 x 947) 2.4 billion, and we can pretty safely assume that the 290X is considerably more powerful than the 270X, even though the 270X has a higher clock speed.

So, a higher clock speed and a higher number of stream processors are both good things, but you can’t look at any single specification to gauge the performance of your graphics card.

Often, the best and simplest way to determine the right video card for you is to consult real-world game benchmarks. Benchmark measurements are based on frames per second, or FPS, which is the number of game images the system is able to render every second. Higher framerates are better, and 60 FPS is a good framerate to target when PC gaming, although slower paced games can be comfortable to play down to around 20 FPS, and fast-paced games can benefit from even higher framerates than 60 FPS.

To demonstrate how you can compare graphics cards through gaming benchmarks, let’s look at some benchmarks from TechSpot for the Witcher 3, currently a very demanding game.

Here are the framerates at 1080p on Ultra quality. We can see the NVIDIA GTX 980 averages 56 FPS, while our AMD R9 290 averages 36. The 270X can only squeak out 14 FPS, which is essentially unplayable.

1080_Ultra copy

Moving up to 1440p resolution, framerates drop to 40 FPS on the GTX 980. For the 290, we see a drop to 28, which is still playable, but far from the ideal 60-plus FPS. The 270X, meanwhile, is in full slideshow mode at 10 FPS.

1440_Ultra copy

For 4K benchmarks, we’re going to switch over to another source, GamersNexus, because they tested slightly more powerful setups. 

witcher-bench-4k-u

4K is a demanding resolution for The Witcher 3. Two GTX 980s averages 42 FPS at 4K resolution. Even the GTX Titan X, with its 12 gigabytes of VRAM, can only manage 36 FPS at 4K, while a single GTX 980 gets 26.

That reminds us of something else we should mention: Multi-GPU setups.

In those 4K Witcher benchmarks, the best performance came from two GTX 980s which were operating in the same system. Assuming you have a compatible motherboard and power supply, a PC can have as many as 4 graphics cards working together to split up the graphical workload.

Two GPUs can be almost twice as fast as a single GPU. Adding a third GPU, however, usually only gives you a small advantage over two, and four GPUs is rarely any better than 3.

Here’s an example of Battlefield 4 performance at 1440p and 4K with one, two, three, and four GTX 980s (courtesy of PCPer.com). You will notice that four GTX 980s actually perform worse than three, because the complexity of adding a fourth card outweighs the extra horsepower it provides.

bf4

The name for this connection of multiple GPUs is different depending on which brand you’re using. For NVIDIA cards, it’s called SLI. With AMD cards, it’s called CrossFire.

 

There are a few more practical considerations to make when picking out your graphics card: Power, heat, and noise.

Typically, your graphics card is the single-most power-hungry component in your computer. A single GPU like this 290 can easily take 300 watts from your power supply. Adding multiple GPUs can quickly rack up the power consumption and is the quickest way to requiring a power supply upgrade.

Graphics cards, like all of the other components in your computer, waste about 90 percent of the power they use as heat, meaning that good airflow is important as you upgrade to large graphics cards or build a multi-card setup. Keep that in mind while selecting a case for your build.

And when they’re under a high performance load, the fans on graphics cards can quickly become the noisiest thing in your computer. If you would like to keep your computer quiet, you’ll want to compare noise levels of different cards. Noise is a tricky thing to compare, because it’s difficult to measure consistently. You’ll have to find third party benchmarks to get useful noise measurements, and they won’t be directly comparable with noise benchmarks from different people.

Here’s a comparison from hardware.info showing load noise levels of different GTX 970s and 980s. Lower numbers mean less noisy. The rule of thumb is that a 10 decibel difference sounds about twice as loud, so the loudest GTX 970 is more than twice as loud as the quietest GTX 970.

noise levels
In conclusion, if you’re shopping for a new graphics card, we highly recommend familiarizing yourself with each card’s specs, comparing benchmarks, and checking out recommendations from reliable sources.

Like I mentioned, NVIDIA and AMD release new generations of graphics cards every year or two, so it’s good to make sure your recommendations are up to date.

In conclusion, if you’re shopping for a new graphics card, we highly recommend familiarizing yourself with each card’s specs, comparing benchmarks, and checking out recommendations from reliable sources.

The GPU chip in each graphics card is made by either NVIDIA or AMD, but the card itself is manufactured by another company, such as EVGA, ASUS, or Gigabyte.

The main differences between these manufacturers will be customer service, warranty, and the noise level and quality of the heatsink.

Helpful links:

Graphics cards featured or mentioned: