In modern pcs and laptops graphics cards are an essential component, because most applications need some level of graphics processing.
On desktop PCs we have the option of installing a discrete graphics card into the pcie slots. These graphics cards can be changed and upgraded in future.
For gaming pcs graphics cards are absolutely necessary as most modern 3d games require one. Other applications based on 3D graphics like modelling, animation etc also require a graphics card.
Besides specific applications even standard applications and OSes like windows and linux require some level of graphics capabilities for optimum performance.
Graphics Cards have a lot of technical specifications that determine its performance. If you are planning to buy a graphics card then its imperative to evaluate key specs like gpu, memory and power requirements.
Even though its true that more expensive graphics cards are more powerful than cheaper ones, they may not always have the best price to performance ratio.
So even if you have a big budget, its important to make sure that the graphics processing power of the graphics card is actually worth the price.
In this article we shall talk about the key features and specs of graphics cards that you need to know when buying one.
- GPU - AMD, Nvidia
- Core Count
- Core Clock Speed
- Memory Type
- Memory Size
- Memory Bandwidth
- Motherboard Interface
- Thermal Design Power
- Power Connectors
- Video Output Ports - HDMI, DisplayPort
- API Support - DirectX, Vulkan
- Compute Performance - TFLOPS
1. The GPU
There are only 2 brands that make GPUs, namely Nvidia and AMD. Their GPUs are then used by 3rd party manufacturers to make graphics cards. Both the brands offer a really large collection of GPUs in various prices points and feature set. The GPU is often referred to as the graphics co-processor or the graphics chipset, both of which mean the same thing.
There is a graphics card for every use case from basic gaming to high fps gaming and 3d modelling. The gpus have many similar technologies implemented under a different code name. For example Nvidia uses the term CUDA Cores where as AMD calls them Stream processors. Similarly nvidia uses the term SLI for multi gpu setup whereas AMD uses the name Crossfire for its multi gpu solution.
Dedicated graphics cards are available as discrete pci cards for desktops pcs and complete pre-installed inside laptops. On desktop pcs you can change the graphics card to upgrade to a newer one, whereas this might not be possible on laptops.
Some of the most popular GPUs include
- Radeon RX 5600 XT
- Radeon RX 550
- Radeon RX 580 GTS
- Radeon RX 570
- Radeon RX 6800 XT
- Geforce GTX 1050 Ti
- Geforce GTX 1650
- Geforce GTX 1660 Ti
- RTX 2080
- RTX 3080
- RTX 3090
In general more expensive GPUs are more powerful in terms of performance and deliver more graphics processing capabilities and functions.
2. Stream Processors/CUDA Cores
These terms refer to the same thing. Stream Processor is the nomenclature for AMD hardware and CUDA Cores for Nvidia. These cores can be thought of as the many individual computing units in the GPU that do the graphics computations and calculations. Having more cores will deliver more performance.
However, comparing cores across manufacturers may not give you an accurate idea of the difference in graphical power, as more variables can affect GPU performance, such as clock speed and architecture.
Even within the same gpu brand, architecture (the design or process upon which a GPU was built) can vastly alter the performance of cores. Comparing the number of cores across two cards in the same architecture will give a more straightforward comparison.
Example of core counts of some GPUs
- AMD Radeon RX 5700 - 2304 Stream Processors
- Nvidia GeForce GTX 1650 - 896 CUDA Cores
3. Core Clock Speed
Each of the aforementioned cores is similar to a core in a CPU, in that it operates at specific clock speed. This clock speed indicates the number of computations the cores does every second, and is measured in MHz.
Once again, simply putting core clock speeds head-to-head is a poor means of comparison, as several other factors can influence performance in general. However, if everything else is identical, then a higher clock speed will generally indicate better performance.
The clock frequency is not constant. AMD Radeon RX 5700 for example has a base frequency of 1465 MHz and a boost frequency of upto 1725 MHz. The base frequency indicates the minimum stable clock speed of the processing core and the boost frequency is the upper frequency limit that is achieved during heavy work load.
Besides this, a lot of GPUs also support overclocking which allows applications to increase the base and boost frequency to much higher values than the specs.
It should be kept in mind that higher clock frequency will produce more heat and heavily depend on thermal conditions. So if you plan to overclock your GPU make sure that there is ample cooling and that the GPU does not exceed the critical temperature thresholds.
4. Memory Type - GDDR
Memory in graphics cards works the same as regular RAM. It stores graphics data temporarily to be processed by the GPU.
RAM in graphics cards is referred to as VRAM, and these days you are likely to see cards that use either GDDR5, GDDR5x, or GDDR6 VRAM.
GDDR6 gives better power efficiency and performance than GDDR5X, which in turn does the same over GDDR5.
In general graphics memory of higher GDDR version will perform better than lower version numbers.
5. Memory Size
As with regular RAM, its size is measured in GB. More RAM is always better, as there is more space to store graphical information. It is important to note that performance might not be increased by increasing the RAM beyond a certain level, as it depends on having applications or games that can properly utilize it.
Commonly seen VRAM sizes are 4GB, 6GB, 8GB. It is worth knowing that VRAM on a graphics card cannot be changed or upgraded like regular RAM on motherboard. The VRAM is built into the hardware of the graphics card.
Most GPUs from Nvidia and AMD specify the amount of memory supported so most manufacturers use the the same amount of VRAM for the same GPU in their cards.
Higher amount of RAM is available on more powerful GPUs.
- AMD Radeon RX 5700 - 8GB
- Nvidia GTX 1650 - 4GB
6. Memory Bandwidth
Memory bandwidth can be seen as an overall assessment of the performance of the VRAM on a graphics card. Memory bandwidth is simply how fast the VRAM on your card can be accessed and utilized when in use.
Memory bandwidth is the product of three variables: memory clock speed, memory bus width, and the transfers-per-clock of the memory type.
- Memory clock speed: Measured in MHz, this variable refers to how quickly your VRAM can access the information it has stored. The higher the number, the better.
- Memory bus width: Bus width is similar to the lanes mentioned earlier. With each clock cycle, a wider bus width will allow for more information to be transferred. This is measured in bits, such as 128-bit and 256-bit.
7. Motherboard Interface/Connection
Whether you're building a PC from scratch or simply upgrading the graphics card in the PC you already own, you need to make sure that the graphics card you purchase is compatible with the motherboard.
In the past, an interface known as AGP (Accelerated Graphics Port) was widely used, but it started to phase out from 2004.
Now, all graphics card use the PCI Express (PCIe) interface to connect to the motherboard.
At this time, PCIe 4.0 is only getting its first few graphics cards, so most cards you see will be PCIe 3.0 based. It is very important to know that PCIe is backward compatible, which means that any PCIe graphics card will operate in any PCIe compatible motherboard.
However, a PCIe 4 card will not be able to reach its full potential in a PCIe 3 slot, while a PCIe 3 card in a PCIe 4 motherboard will not reach the motherboard's full capacity.
If you are planning to purchase a high end graphics card that supports PCI-E 4.0 then it is recommended to have a motherboard that has PCI-E 4.0 support. That way you get the maximum performance from the graphics card.
PCIe interface has an ‘x’ value, for example, x8 or x16. This refers to the number of lanes that the slot has. Think of these lanes as the lanes on an expressway, or pipes carrying water.
So, x16 will be able to work at a higher bandwidth than x8 or x4. Most graphics cards are x16 nowadays.
8. Thermal Design Power (TDP)
Thermal design power, or thermal design point, is a good way to assess the power consumption and thermal performance of a GPU. Like the term denotes, it indicates the power required to generate the highest amount of heat that the cooling system can handle.
This is measured in watts and it potentially affects the selection of other parts of your pc build. You must ensure that the output power rating for your PSU is sufficient to not only carry your graphics card but all other components in the system.
- AMD Radeon RX 5700 - 180 W
- GeForce GTX 1650 - 75W
If your graphics card has a high power rating like 180W and above then it is recommended to have a PC case that has good ventilation for maximum heat dissipation.
9. Power Connectors
A PCIe slot can provide power to the card slotted into it, but only 75W. Graphics cards have become so power-intensive that it did not take long for them to surpass that limit and require more power.
Due to this, modern GPUs have power connectors that allow them to draw additional power directly from the PSU. These connectors can be either six-pin or eight-pin connectors.
A modern graphics card can have up to 2 connectors, which can be any combination of these two. So when purchasing a PSU, besides the maximum output power, you should take note of the power connectors it has and ensures it will be able to power your graphics card.
10. Display Output Ports
Graphics Cards often have multiple different types of video output connectors.
Depending on the type of monitor you use, you will most likely be able to connect to the card via HDMI or DisplayPort, which are more common-place when it comes to displays.
Some newer cards support using USB Type-C to connect, though it is less common to find monitors that support that, as it is still an emerging technology. VGA and DVI are relatively older ports which you may only see on older displays.
If you hope to connect your PC to multiple monitors, it is important to take note of which ports are available, and the connectors that your monitors have access to.
Currently the most ubiquitous of the port options available, HDMI has been around for a long time and good reason. It can be seen on PCs, TVs, Blu-ray players, video game consoles, and set-top boxes.
HDMI is advantageous as it supports audio and video, both in their uncompressed forms. The newest revision, HDMI 2.0, has enough bandwidth to support resolutions of up to 4K at 60Hz, which can also allow 1080p at 144Hz.
HDMI 2.0 also supports up to 10-bit and 12-bit color, which allows for HDR (High Dynamic Range) content playback.
At this point, DisplayPort is as well known as HDMI, and on its way towards having the same reach that HDMI does. Similar to HDMI, it supports both audio and video output.
Achieving higher resolutions on DisplayPort has always been straightforward, even since the earlier revisions. DisplayPort 1.4 can render up to 4K at 144Hz, while even the 1.1 revision, which is relatively outdated, can support up to 1080p at 144Hz.
At lower refresh rates, DisplayPort can support up to 8K resolution, making it one of the only output options that can support this coveted resolution.
The newest of the group, USB Type-C, improved on the foundation that USB Type-A had established. It is smaller, fully reversible, and extremely versatile. USB Type-C can transmit data, but also audio, video, and even act as a charger.
USB Type-C can be found on laptops, tablets, and smartphones, and with its presence increasingly widening, monitors are starting to support USB-C.
USB Type-C can support resolutions of up to 4K refreshing at 60Hz. One downside is that USB-C monitors that do not support at least DisplayPort Alt Mode 1.2 are currently not able to support Adaptive-Sync technology.
DVI is a relatively older output type that is slowly being phased out in favor of hdmi and displayport.
There are 3 types of DVI, which are DVI-A (analog, and essentially obsolete), DVI-D (digital), and DVI-I (both analog and digital signals). For DVI-D and DVI-I, there are single and dual-link variants, of which the latter can support more bandwidth.
DVI-D is still capable, however, being able to support a maximum resolution of 1080p at 144Hz.
VGA is the oldest display output method of those mentioned here and was primarily used in the days of CRT displays. Newer output interfaces were developed as flat-screens and higher resolutions became more prominent, as analog signals of VGA could not support the resulting resolutions.
VGA can only support up to 1080p at only 60Hz. The VGA port can be seen on older graphics cards only. Most new and recent graphics cards and motherboards have completely removed VGA support.
Most newer monitors from top brands have also abandoned vga port and have either hdmi or displayport or both.
11. API Support - DirectX, OpenGL, Vulkan
Graphics cards are built to process graphical information for your PC, as they are specially engineered to be able to do this. However, to do this, the hardware and software must be able to communicate and send instructions to one another, and this is where a Graphics API comes in.
An application programming interface contains a set of instructions that tell the GPU how to resolve complex graphical tasks.
There are different APIs, which are all coded differently but can each achieve most graphical tasks required in this era.
APIs must be specifically supported by graphics cards drive and the hardware must be able to interpret the instructions provided by the API.
DirectX 12, OpenGL 4.6, and Vulkan 1.2 are the latest versions of the most popular APIs currently. Most popular graphics cards based on AMD or Nvidia GPUs support Vulkan and DirectX.
It should be noted that OpenGL is being replaced by Vulkan as a cross platform 3d graphics API.
Check out the wikipedia page to learn more
A gigaflop or a teraflop is a unit for measuring the theoretical performance of a processing unit, which can be either a CPU or a GPU. The FLOPS stands for floating-point operations per second, which refers to how many floating-point operations it can do in a second.
Using gigaflops or teraflops is one of the best ways to have an estimate of the relative performance of a processing unit over another, though it is not comprehensive. Differences across architectures may not give accurate estimations.
13. Vendor-Specific GPU Technologies
Nvidia and AMD have been competitors for years, and beyond the raw graphical power of their respective offerings, they are each constantly developing new technologies to give the consumer a better experience when using their graphics cards.
These technologies are specific to the manufacturer and can improve the gameplay experience for the consumer.
- Nvidia G-Sync: This is Nvidia’s take on adaptive sync technology for displays. With both a graphics card and a monitor which supports G-Sync, the refresh rate of the display can be adapted to match that of the GPU, thus preventing screen tearing.
- Nvidia DLSS: DLSS stands for Deep Learning Super Sampling. Images are rendered at a lower resolution and up-scaled with AI. This allows for higher graphical fidelity to be achieved with less performance cost.
- Nvidia Ansel: This is a software addition that makes it easy to capture in-game shots during gameplay, and even adjust positions and apply filters. The images can then be shared extremely easily on different social media platforms.
- Nvidia NVLink: This is an interface that allows for direct interconnection of multiple Nvidia GPUs simultaneously, with impressive bandwidth. This can allow for improved graphical performance, but typically only where supported.
- Nvidia GPU Boost: During gameplay, if an Nvidia GPU is running cool even at its base clock speed, it can intelligently overclock itself up to a certain speed to leverage more performance.
- Nvidia VR Ready: This is a tag used by Nvidia to show that the respective hardware has the technical capability to support VR applications.
- Nvidia Highlights: This software can detect important moments during gameplay, and automatically record them. These shots can easily be shared afterward.
- AMD FreeSync: This is AMD’s variant for adaptive sync. The graphics card and display must both support FreeSync. However, unlike Nvidia’s G-Sync, FreeSync can be used by either Nvidia or AMD GPUs.
- AMD CrossFire: CrossFire is AMD’s multi-GPU technology to increase graphical performance. It allows up to 4 GPUs to be interconnected in a single PC.
- AMD Eyefinity: This allows multiple displays to be used in sync with one another. Multiple monitors can be placed side-by-side, and Eyefinity software will distribute the whole image over each monitor to give you a bigger viewing area.
- AMD ReLive: This allows for stress-free capture of in-game shots and videos, which can then be effortlessly shared on social media platforms. It also has support for live streaming, making it easy to get started on platforms like Twitch.
- AMD VR Ready: This is AMD’s tag on its hardware which can support VR software and headsets such as the Oculus Rift.
- AMD PowerTune: This allows supported AMD GPUs to alter their clock speed dynamically to improve performance during work or play. It uses the power consumption and thermals of the GPU to limit the overclocking.
- AMD Radeon Boost: During moments where increased frame rates are required, like when moving the crosshair rapidly on-screen, resolution can be reduced intelligently to allow for increased FPS.
That was a brief overview of the technical specs of the graphics cards. Some of the specs like core counts and memory are similar across all cards with they have amd or nvidia gpus.
Besides that, each gpu manufacturer has their own proprietory technologies like G-Sync/FreeSync that may do similar things but have technical difference in their implementation.
Also keep in mind that the choice of graphics card also affects the power supply, pc case, the monitor and sometimes even the motherboard.
If you have any questions or feedback, let us know in the comments below.