It comes to the attention to most consumers on how to purchase a card. In fact I have seen the majority of them only purchase cards because of the pricing, higher the price the better, right? Wrong!. What we and everyone should know is how to differentiate between the specifications that are provided on each card. What will be discussed here in this section is the key importance to gaming and price, what’s worth and not worth purchasing.
What is the Core (GPU)?
Core is used to manipulate and alter the memory to accelerate the building of images in the frame buffer and then output the information to a display. Like a CPU, the GPU also comprises of several cores within the same die, like a quad-core CPU has 4 cores on 1 die, the GPU contains several streaming-multiprocessors that handle the same calculations, as to how many cores are used depends on the code being processed.
nVidia has 16-groups with each group containing the 32 cores for a total of 512 cores or stream processors, while AMD has 6 cores.
AMD’s 6 cores handle several calculations with linked clock speeds so they cannot be adjusted. This keeps the hardware stable.
AMD is said to have linked Core/Shader Clock frequency and cannot be adjusted or overclocked independently to avoid performance loss or bricking/melting of the hardware. Some say with the reduced amount of cores used, performance and calculations are increased without the use of so many cores handling the job, but with the advantage of bundled calculations. Temperature and power usage should be stable as the shader/core clocks are linked.
nVidia’s Fermi is comprised of (16-groups * 32 cores each group) = 512 streaming-multiprocessors, you can throw any piece of code at it and it will run on as many of the stream processors as it needs, they all have the same functionality.
With nVidia the Core/Shader Clock are not linked together which allows for adjusting /Overclocking the frequency between the two. However the downside is of course, melting the hardware and instability. Some believe that though, any core can carry the same task, may be more accurate in its computations, with more cores the job gets done faster, but at the cost of power usage and increased temperature, this is just word of mouth. But since the shader/core clocks aren’t linked, this can be adjusted to decrease power usage which also decreases temperature.
Keep in mind this is not a fanboy discussion.
What is a Shader?
There are three (3) types of shaders within the GPU; Pixel, Vertex and Geometry.
Fragment shaders or pixel shaders calculate color and other attributes of each pixel. Fragment shaders range to applying a lighting value such as ambient lighting, brightness, contrast, light levels, hue and saturation, bump mapping, shadows, mid-tones, specular highlights, and translucency/alpha channels.
It’s purpose is to transform each vertex’s 3D position in void space to the 2D coordinate at which it appears on the screen. Vertex shaders can control the properties such as position, color, and texture coordinate. It’s like creating a map within Source SDK, through the compiling process of a map it draws the position on the screen in which is can be viewed while playing. In fact it will allow you to see a pre-rendered scene within a 3D viewport within the sandbox where you create the maps. See image below of all shader combinations to create the rendered output.
The geometry shader can generate primitives, points, lines, and triangles, from those primitives that were sent to the beginning of the graphics pipeline.
Combined shaders(Pixel, Vertex, Geometry) to create an objects position, lines, color, points and primitive on the screen: 3D-View Port
We know how the Shader cores work, so how is the Shader Clock important?
The higher the Shader clock speed, the faster the Pixel, Vertex and Geometry is constructed within the ALU and immediately sent to the core for output.
What is Memory Clock?
The memory speed determines how fast the data can be retrieved and sent to the core (GPU).
Shared memory plays a little role in storing instructions for the Core and Shader cores. The Core and Shader will have already completed the computation long before the shared memory has new instruction for either to process. With most modern PC’s most of the instructions are held within your computers physical memory rather than in the VRAM of the graphics card, you will notice during gameplay that very little shared memory is being used on the card while maybe a max of 500MB physical memory is used when you play games.
If you are running 8GB on Windows 7, with page-file enabled, your page-file won’t reach higher than 1.5GB - 2GB in page and your games shouldn’t go over 300MB-500MB of physical memory with intense games like Crysis 2 or Battlefield 3. At a low if you are playing World of Warcraft you will see an increase of 375MB of physical memory just to run the application. Always read the minimum requirements of a game before you play it as well.
Now as explained above the way the GPU/Shader cores work, not many instructions are stored in the shared memory which shows how fast the cores are doing their jobs, higher clock speed in the shader means a faster output through the GPU. The shared memory does not play much of a role but its clock speed does, keep an eye on high memory clock speeds and fewer eyes on shared memory. With most modern cards you will generally see 1GB shared, that is all you need or 768MB minimum.
The memory interface is also important, the wider the bus the faster the data transfer through the bus. However with a wider bus then clock speed is reduced, unless overclocked. But we know that overclocking can cause instability and excessive power usage that increase temperature and other problems as well, so be careful if this is a route you choose.
It is said by Tom’s that shared memory does not play a large role and if you choose a card get a card with low shared memory increased clock speed and wider bus, just remember as your bus gets wider the clock speed will decrease by a percentage.
The focus is low shared memory, larger memory interface, faster memory clock speed, faster core/shader clock speed, and at least GDDR5.
Choosing GDDR3, GDDR4 or GDDR5
GDDR5 is based on DDR3 memory which has double the data lines compared to DDR2. GDDR5 has an 8-bit wide prefetching buffer similar to GDDR4. However GDDR3 and GDDR4 has higher clock speeds but reduced latency, try to keep an eye for GDDR5 in memory.
I would like to point out that these are my thought on choosing a card for the best price without sacrificing an arm and leg. But also keep in mind that the hardware you have in your machine plays an important role in gaming as well, your HDD/SSD, your processor speed, core count and caching, your motherboard etc. Please, also before anyone makes and bashing/flaming statements if anything seems incorrect I am more than happy to learn and adjust my wording in this thread.. Possible sticky this thread?