What is a graphics card?
A graphics card renders images to the screen. It converts the binary data to images. Graphics Card is of two types -Integrated or Discrete.
Integrated graphics card are built in with CPU. It operates basic functions such as internet browsing, documents, movies etc.
Discrete graphics card is an extra hardware to the motherboard. It processes advanced games, AI, video editing in addition to the basic functions.
What are the main components of a graphics card?
There are four main parts to the graphics card – GPU (Graphical Processing Unit), RAM (Random Access Memory), External Bus Interface and Display Ports.
What are the market shares of Nvidia and AMD in discrete Graphics Card?
What is the function of RAM in Graphics Card?
Video RAM (VRAM) read, writes, stores data and images in its memory. It inputs binary data to the GPU and retrieves the digital images. It then processes resolution, frame buffers etc. and connects to DAC. DAC is Digital to Analog converter. All the monitors uses analog signals for putting on the screen.
What are the types of VRAM?
Currently most RAM uses DDR (Double Data Rate). Consider, clock as sinusoidal waves. Hence, in every crest and trough, data is transmitted. Therefore, it is DDR. Most common used VRAMs are GDDR6X, GDDR6, GDDR5X, GDDR5, HBM and HBM2. G denotes Graphics. Number denotes the generation. HBM is High Bandwidth Memory. It is stack able multi layer memory. It has higher bus width and bandwidth. A higher size gives better FPS and resolution.
Lets see how they compare below :-
A discrete graphics memory is much faster than regular RAM . And hence they are also very costly.
What is a GPU?
Graphical Processing Unit are the set of cores in graphics card. Image processing requires huge resources. General CPU cannot put enough resources to process this heavy data. For this purpose, GPUs are used. They are dedicated for image processing.
Each GPU is divided into hundreds of cores. Each core divides the processing among themselves. In Nvidia, each core is called CUDA core.
What is Ray Tracing and Tensor Cores?
Each image also has textures, shadows, objects, lighting. These are also handled separately in the GPU. Nvidia uses RT Cores (Ray Tracing) for simulating light source effect on each object. Their mirror images on glass, shadows, brighten areas etc. This makes them more realistic.
Tensor Cores performs deep learning and Artificial Intelligence for image processing.
What is Base Clock and Boost/Peak Clock?
Each memory process follows clock cycle. Every data is transmitted over clock cycle. This determines the speed of the GPU and Memory.
At normal operation, base clock is used. Boost Clock is the speed upto which the memory can run under heavy processing.
What is Thermal Throttling?
If GPU gets hot under heavy processing, this is the speed to which it drops to. Therefore, we need heavy coolers over a graphics card.
How to calculate Effective Memory Clock speed?
Each memory has own setup. For Example, GDDR5 has 8 memory chips with DDR capacity. But, GDDR6X, GDDR6, GDDR5X has QDR that is 4 times Single Data Rate or 2*DDR. So, they have double the DDR capcacity.
Effective Memory Clock = Memory Clock * Number of Chips / 2
Number of chips can be taken from above mentioned table. 2 denotes DDR, because speed needs to be converted to DDR as per standard.
For Example, RTX 2080 Super has GDDR6 VRAM with memory clock speed of 1936 MHz.
Now, from above VRAM comparison table, GDDR6 has 16 chips.
So, Effective Memory Clock = 1936 / 2 * 16 = 15,488 Mbps.
What is Bandwidth? How to calculate Bandwidth?
Bandwidth is the speed of the data transfer between VRAM and GPU. It is calculated as Effective Memory Clock * Bus Width.
For example, Nvidia GTX 1080 has 11008 Mbps of effective clock speed and 256-bit of bus width. This is in Megabits per second. For 1 MB = 8 bits and 1 GB=1024 MB. Therefore, its is equal to 11008/8=1376 MB/s (1.376 GB/s).
Now, Bandwidth = 1.376 GB/s * 256 = 352.26 GB/s
What is TFLOPS? How to calculate TFLOPS?
TFLOPS or Tera Flops are the way to measure the computational power of GPUs. It measures how many floating point calculations it can handle per second. Since, processors uses real numbers as data, floating point refers exactly that. Higher the number better power of the processor.
Formula = Number of Cores * Boost Clock * 2 / 1 million
2 denotes number of instruction set per cycle. Each cores process 2 instructions per cycle.
For example, Radeon RX 5600 has 2048 cores and 1560 MHz Boost Clock speed. So, it can process 2048 * 1560 * 2 / 10,00,000 = 6.39 TFLOPS or 6.39 trillion real numbers per second.
What are NVLink/SLI or Crossfire?
These are the protocols for any 2 or more graphics card to work together. Earliest protocol SLI or Scalable Link Interface has Master-Slave protocol. That means, Master Card (with best configuration) divides the work among its slave graphics card. And the processed data is given back to the Master Graphics Card. Maximum 3 slaves can attach to one Master Card.
In NVLink, there is no such Master Slave protocol. All graphics card process and share data bi-direction-ally acting as a single card. This improves the bandwidth and throughput. Example, RTX 2080 has NVLink support.
But, in general single graphics card with combined capability serves the best.
What is FPS? How it affects the gameplay?
FPS is frame rate per second. Also, Hertz is same as FPS. Human Eye can detect 25 frames per second. But, games need at-least 60 fps for smooth and faster scroll. Higher the FPS better the game-play and smoothness.