When buying a video card, what specs should I compare/which are most important?

I'm currently deciding on if I should upgrade my current video card or not. I know I will find myself in this situation more times in the future, so I figured I would educate myself.

What should I be looking for as far as specs are concerned and which are most important? From a basic one monitor setup to a multi-monitor gaming PC setup. I would love a hard and fast way of comparing specs. Such as Card A has 123mhz of SpecA and 2gb SpecB, so it will perform better than Card B with 100mhz of SpecA and 1gb of SpecB.

As an example, how would I quantitatively compare the specs of the following cards:

  1. http://www.newegg.com/Product/Product.aspx?Item=N82E16814121660
  2. http://www.newegg.com/Product/Product.aspx?Item=N82E16814131544

I understand that sites like videocardbenchmark.net exist, but I would like to be able to compare specs even if I don't have access to these types of resources.

TLDR: What video card specs are most important and how can I compare them?

Thanks!


Solution 1:

Because performance is heavily application-specific, you cannot quantitatively compare two video cards with different GPU architectures based on specs alone. Different GPU architectures scale differently with various specs such as memory speed, memory size, memory type, and bus width; and the only way to divine the scaling ability is to look at benchmarks.

The only exception to this is if you are comparing two cards which are "obviously" not within the same performance class-for example, comparing a 2 MB S3 ViRGE to a 4 GB Radeon R9 290. The specs are so incredibly disparate (by an entire level of magnitude) that it's not difficult to guess which card is most likely many generations newer and should have better performance.

But for the two cards you list, you have to refer to benchmarks for the exact or very similar applications that you are going to run--and note that similarity is not just a class of applications (such as "games" or "bitcoin mining") because some applications within the same class may be better-optimized for a particular piece of hardware, or a given piece of hardware may have more mature drivers.

That said, there are several specs you can compare if you're looking at two video cards based on the same GPU architecture. For example, see the Nvidia and AMD tables at Wikipedia.

GPU architecture/code name

GPU clock speeds

  • Core (MHz)
  • Shader (MHz)

Number of shaders

  • Unified shaders
  • Texture mapping units
  • Render output units

Memory

  • Size (MB or GB)
  • Bus type (DDR3, GDDR5, etc.)
  • Bus width (bits--e.g., 64-bit, 128-bit, 256-bit)
  • Frequency (MHz)

Power consumption

  • TDP (watts)

There are many other specs that are derived by multiplying two or more of these basic hardware specs, such as memory bandwidth (GB/s) or processing power (GFLOPS).

Generally-speaking, if you're looking for better performance, you want the newest architectures and the highest numbers for all other specs except TDP. (But sometimes one manufacturer's technology may lag behind that of a competitor.)

Again, performance can be highly application-dependent, as you'll notice if you look the results from any comprehensive benchmark suite for two different cards (especially with different architectures). While one application may benefit from larger memory, another might perform better with greater bandwidth (memory frequency, bus type, and bus width) or processing power (core frequency and number of shaders). Depending on your application, improving certain specs may not yield any performance gain whatsoever.

As if divining the computational performance wasn't already difficult enough, most people also factor in the cost, which may include not only the initial purchase price, but also power consumption.

Solution 2:

I find GPUBoss as an excellent tool for comparing graphics cards. For example, to compare your two cards:

PowerColor-Radeon-R9-290-vs-ASUS-Geforce-GTX-660

You will see where either card excels with an in-depth comparison on the bottom. Looking at these statistics you should get a rough idea of all the different aspects involved in comparing one card to another.

There is no clean cut way of comparing graphics cards just by looking at specs, as performance may vary from card to card (with the same clock speed and memory). Of course this also depends on what your looking for as far as an upgrade goes. Whether you want overall performance, higher fps in a certain game, or the best performance per $.

Solution 3:

Considering the cards you have mentioned, I am assuming you intend to use it for gaming. Then best thing to compare in my opinion is game FPS of cards when all other specs are kept same. There are a lot of comparisons out there which have some test rigs built to compare cards. I personally prefer comparing game FPS rather than synthetic benchmarks though I could very well be wrong in doing so.

As far as specifications are concerned, graphics cards have lot more parameters to compare than say CPU or RAM. And quite often, a card will have better RAM but say lesser clock frequency than a similarly priced card. So, its quite confusing to compare two similarly spec'd cards just by looking at specifications.

PS: Unless you plan to bridge it with second card, don't go for Crossfire/SLI supporting cards because they tend to be more expensive.