The graphics processing unit (GPU) has the greatest influence on a gaming PC’s performance, second only to the central processor unit (CPU). The GPU is made up of an extra processor that converts data from the CPU into visuals that may be displayed on your display. In other words, the GPU handles the majority of the computationally intensive tasks while you play games. A GPU’s (also known as a graphics card’s) processing capability determines how much data can be calculated and presented in a given amount of time and, ultimately, how enjoyable your gaming experience will be.
The CPU was in charge of converting data into graphics in the early PC era. In particular memory areas known as “frame buffers,” the data was kept and then sent to the display. As many general-purpose CPUs struggled to do these tasks well, “graphics accelerators” were developed to take over some of the CPU’s more specialized tasks. This became more crucial as graphical user interfaces (GUIs), which are present in more contemporary operating systems like Windows, gained popularity.
Modern GPUs are exceptionally quick at processing complex 3D visuals needed for cutting-edge gaming experiences. They excel at processing enormous volumes of picture data in parallel and presenting text and graphics in windowed GUIs. GPUs are handy for some applications outside of gaming since they can handle other tasks that require handling large amounts of data in parallel effectively.
GPUs are crucial, but how do you choose the right one? There are numerous GPU alternatives available from different manufacturers, so it might not be immediately obvious which one best suits your requirements. Making that choice can be made simpler if you understand the fundamentals of how they work and how they differ from one another.
Table of Contents
AMD or Nvidia?
Even though there are several manufacturers and hundreds of graphics cards, only Nvidia and AMD actually produce the GPUs that power these devices; however, Intel’s Xe Graphics has started shipping for laptops and will likely be available for desktops in the near future as well. AMD is more competitive than it has been in recent years with Nvidia and its current-generation Ampere cards, including the GeForce RTX 3080, thanks to its RX 6000 cards.
But real-time ray tracing is the unseen, realistically lit elephant in the room that we have been ignoring up to this point. “Team Green” is now using its second generation of RTX with 30-series GPUs, which was first introduced as a significant new feature with Nvidia’s then-current-generation RTX 20-series cards. With its RX 6000 cards, AMD (“Team Red”) made a significant entry into this market in 2020. However, because real-time ray tracing is still new to AMD, it trails behind Nvidia in this area.
However, the development of games that use ray tracing effectively has lagged behind. There’s no denying that more games are adding ray tracing functionality, and there will be a lot more in the future since the Sony PlayStation 5 and Microsoft Series S|X consoles also support it. There are currently just a select few video games that make use of ray tracing in a manner that we would consider to be outstanding.
Games that simply make use of one RT effect, such as reflections, typically have lower requirements and lower overall quality. Therefore, consider how engaged you are in these games, how significant the greatest visuals are to your enjoyment, and how much future-proofing you want baked into your GPU before deciding how crucial ray tracing performance is.
DLSS, Nvidia’s AI-assisted resolution upscaling, should not be overlooked either. It can give better performance with a smaller impact on frame rates than is normal when you maximize the resolution of your monitor. Only a small number of comprehensive ray tracing games, though they are increasing in number, support this functionality. Although DLSS is more frequently supported in games that actually require upscaling, AMD offers its own open source alternative to it called Fidelity FX Super Resolution (AMD FSR), and FSR 2.0 should make things even better.
How to buy a GPU: Which specs matter and which don’t?
Memory on the graphics card: Very important. For 1080p gaming, get a card with at least 6GB and ideally 8GB or more. If you install high-resolution texture packs or play with all the settings turned up, you’ll need more memory. More than 8GB is great if you’re playing games at very high resolutions, like 4K.
Form: Very significant. Make sure there is space in your case for your card. Look at the width, depth, and height. Graphics cards can be found in single-slot, dual-slot, triple-slot, and even half-height (slim) varieties (or more). With current-gen cards being thicker and larger than many previous-gen models, the majority of gaming-focused cards will be full-height and occupy two or more expansion slots. A card can obstruct an adjacent slot even if it only physically occupies two slots in your chassis due to its large heatsink and fan shroud. Look for a “mini” card if you have a compact Mini-ITX motherboard; these cards are typically 8 inches (205mm) or shorter in length. Check the specs because some cards with this name are longer.
TDP: Significant. The term “thermal design power,” or TDP, refers to a measurement of heat dissipation that also provides an estimate of the amount of power required to run your graphics card at its default settings. (Both AMD and Nvidia appear to be moving toward TBP, or typical board power, which refers to the total power of the card. In any case, that is what the majority of us assume when discussing graphics power.) If you have a 95-watt overclocked CPU and a 400-watt power supply unit (PSU), and you wish to add a card with a 250-watt TDP, you almost surely require a PSU upgrade. Generally speaking, many older-generation cards were good with a 600W PSU. However, it’s recommended to use a higher-wattage PSU if you’re going with an RTX 3080/RX 6800 XT or higher, especially if overclocking is on the agenda. Extreme users will likely need a 1200–1600W PSU with cards like the RTX 3090 Ti and reports of upcoming 600W GPUs.
Power connectors are crucial. All serious gaming cards consume more power than the x16 PCIe slot’s typical 75W limit. Additional PCIe power connections, available in 6- and 8-pin types, are needed to connect these cards. The RTX 30-series cards made by Nvidia have 12-pin connectors but also come with 8-pin to 12-pin adapters. There can be both 6- and 8-pin ports on the same card, and some cards have one of these connectors, others two or even three. Upgrade your power supply if it lacks the additional connectors you require; adapters that rely on a few SATA or Molex connectors are not advised as long-term solutions.
Important: ports. Some monitors use DisplayPort, while others use HDMI, and some older models only have DVI. A few monitors can also route DisplayPort signals over USB Type-C, however these are currently very uncommon. In order to avoid having to purchase an adaptor or even a new display, make sure the card you intend to purchase has the connectors you require for your monitor(s) (unless you want to). If given the option, which port would you like to use? For additional information, see our story comparing HDMI and DisplayPort.
Timekeeping: Slightly significant. A tiny 3-5% variance in frame rates may result from manufacturer overclocking of cards with the same GPU (for example, an RTX 3060 Ti). Clock speed is important, but memory speed, the number of cores, and the architecture also need to be taken into account. On cards with the same GPU, better cooling frequently triumphs over clock speed as well.
CUDA Cores / Stream Processors: Slightly significant, similar to clock speed, as it only provides a portion of the information you require to estimate the performance level of a GPU. Comparing core counts between architectures is less significant than comparing core counts within the same design. Therefore, comparing Ampere and Turing CUDA cores (or Streaming Multiprocessors) is less helpful than comparing Ampere and Turing. Likewise, comparing Navi to Vega or Polaris Stream Processors (or Compute Units) for AMD isn’t especially useful. Even less helpful is comparing the architectures of AMD and Nvidia just on the basis of core counts.
TFLOPS and GFLOPS: Vital. Trillions of floating-point operations per second, or TFLOPS, is a measure of a GPU’s maximal potential capability. (Alternatively, it can be written as GFLOPS, or billions of FLOPS.) The TFLOPS for a GPU is calculated by multiplying the core count by the clock speed in GHz and then by two (for FMA, or Fused Multiply Add instructions). TFLOPS often indicates how much faster one on-chip is in comparison to another when comparing chips that have the same architecture. It is less helpful to compare different architectures (such as AMD Navi 10 vs. Nvidia Turing TU106 or AMD Navi 10 vs. AMD Vega 10).
Memory bandwidth and speed: Slightly significant. Faster memory can make one card faster than another, similar to higher clock speed. Because to the improved memory bandwidth, the GTX 1650 GDDR6 is, for instance, 15% quicker than the GTX 1650 GDDR5. The number of memory accesses can be decreased by features like AMD’s Infinity Cache on RDNA 2, thus bandwidth isn’t the only thing to take into account.