Tech-Hounds.com

Graphics Performance

Now, let's talk about performance, one of the most 'interesting' aspects of any graphics card. Since many of us buy graphics card to play games, we want the highest possible performance from our graphics card (within our reasonable budget of course). Some of you may think that measuring graphics cards performance should be easy enough, just choose the graphics card that provides high enough frame rates when we play games. Unfortunately, it's not as cut and dry as that.

The Processor Connection

It's true that a faster graphics card will display (or render) images faster, but there are many other factors to a graphics card performance. For one, a graphics card will only render images of anything when it's told to, usually by your processor. That's why we need to have the fastest processor if we want to use the fastest graphics card. If the processor is too slow, your super-duper-fastest-graphics-card-on-the-planet will have to wait for instructions, doing nothing in the meanwhile. Needless to say, this is not what we want. Then of course these instructions must be sent to the graphics card itself, which is why we need a fast connection for the graphics card. In fact, this is the sole reason why we moved from ordinary PCI connectors to AGP (Accelerated Graphics Port for those who don't know). This is also the reason why graphics card have the fastest connector available in PCI Express, the x16 slot.

Inside your Graphics Card

Let's take a deeper look at your graphics card. Of course, we want our graphics card to be fast, but what does that mean? Basically, we need our graphics card to be able to process data faster. So in a way, your graphics card is like a mini PC, complete with its own processor and memory. Like we talked about before when talking about processors, pure speed isn't everything. The same things that affects processor performance also applies here such as IPC (Instruction Per Cycle) and bandwidth, the two other major factors making up the performance formula.



To be able to display those scenes in your favorite games, your graphics card uses two kinds of data: triangles / polygons and textures. Of course, one measure of a graphics card performance is how much polygons and texture can the graphics card process per second. These are usually stated as MPolys / sec (for triangles / polygons) and MPixels / sec (for textures). If a graphics card can process one polygon and one pixel per cycle, running it at 500 MHz would mean it can process up to 500 MPolys / sec and 500 MPixels / sec. Now remember, this is an overly simplistic example and graphics card are anything but simple. This number will increase if the graphics card can do more per cycle, in fact this is what modern graphics card do. They use several processing units within one chip or one card (just like a processor's pipeline - only this is basically a graphics pipeline), some dedicated to processing polygons, others for textures and then just running the entire card as fast as possible. Units dedicated for polygons are called vertex units or pipelines, while units for textures are called pixel units or pipelines.

Here comes the other part of the equation: bandwidth. Faster graphics card can process much more data, so it's only natural that we increase the amount of data. Why? More data means more detail (both polygon and texture) and that's why we are seeing more and more memory on graphics cards, the latest being 512 MB on a single graphics card. By contrast, before 3D graphics accelerator became the norm, 8 MB was enough. A graphics card with that much RAM would be able to store roughly 512 MB of polygon and texture data, but if you only have a 128 MB graphics card, the additional 384 MB of data must be placed in your PC's RAM, which is slower. So, in this case you must choose a lower detail setting that will use less data (and less detail). There's also transfer speed to consider, we want the data stored be transfered as fast possible to the graphics chip. There are two ways to do this: use the fastest memory possible (speed and timing) and use the widest channel possible (usually stated in bits), much like Intel and AMD did when they move from single to dual channel memory controller. Graphics card manufacturers opt to do both, as you can see from the memory clock speed and bits in technical specifications.

Here's a list of some internal factors that influence a graphics card's performance:

• Data processed per cycle : both in polygon and pixels. In general, you can get this number by dividing the MPolys/sec and MPixels/sec by the core's clock speed.
• Speed (in MHz) : clock speed of the core / chip and memory
• Memory capacity (in MBs)
• Memory bus width (in bits).

So, where do we start looking then? Look back at what we've discussed before: high-end, mainstream and level entry cards differ not only in price, but also performance. So you can be sure that these internal factors will vary between each series. Usually, the fastest card will have the fastest clock both for the graphics card chip itself and the memory. The chip will also be able to process more data per cycle. To feed enough data, they will use a wider bus (coupled with a high enough memory clock) to provide massive bandwidth. For mainstream card, one or several of these aspects will be lower (ie. lower bus width or lower clock or lower data processed per cycle). And of course, the entry level card will have the lowest specification (ie. lowest bus width or lowest data processed per cycle).

Looking back at the graphics card technical specification from both manufacturers, we see that both of them use the same decisions when differentiating their products. To differentiate mainstream and high-end products, they lower the data processed per cycle by lowering the number of 'pipelines' and also lowering the bus width from 256 bits to 128 bits or even 64 bits. They also offer several products in each series using different clocks for both core and memory or even the number of pipelines. So based on this specification, the products from both ATI and NVIDIA are evenly matched. We just have to find what graphics card will meet our need for speed.

It's In The Games

Now, we have some idea of what determines the performance of a graphics card. But unfortunately, there is no way of actually knowing for certain how these cards will perform while playing games except by actually measuring their performance during gameplay or a running a benchmark that simulates it. And since there are external factors affecting performance, this complicates things further. But all is not lost, many print and online publications provide benchmarks of several games and applications, either using the actual game or a specialized synthetic benchmark designed to measure performance.

Here are several things you have to consider when looking at the game's benchmark results.

• Remember this fact: we need a balanced combination between the processor and the graphics card.
• Choose the results that best suit you:
• • The monitor's resolution you're going to use for games / applications. You need a high-end graphics cards to play at resolutions such as 1280 x 960, 1280 x 1024 and 1600 x 1200. For a resolution of 1024 x 768 and below, mainstream cards is enough.
  • The level of detail - high, medium, low. High-end cards can still handle high levels of detail at resolutions above 1024 x 768. For mainstream cards you may have to use medium or low detail.
  • The graphical features activated - full scene anti aliasing, anisotropic filtering. High levels of anti aliasing and anisotropic filtering have less penalty on high-end graphics cards, so performance won't suffer much above 1024 x 768. Mainstream cards can still provide fast enough frame rate with anti-aliasing and anisotropic filtering on at 1024 x 768, but these settings are not recommended for higher resolutions.
• Does the result vary much either in resolution or detail?
• • If it doesn't vary that much from resolution to resolution and / or detail, the game or benchmark is quite possibly limited by the processor - the processor couldn't supply data fast enough. In this case, buying a faster processor would be more appropriate (if there's one or if your budget is enough).
  • If it doesn't vary much in resolution but does in detail, the game or benchmark probably needs a graphics card that process more data per cycle. Choosing a faster clock graphics card will help, but you're not going to notice that much difference.
  • If it doesn't vary much in detail but does in resolution, the game or benchmark probably needs a graphics card that has more bandwidth. Choosing a graphics card with faster memory will help, but it's better to choose one with a wider bus (higher bits).
  • For high-end cards, does it also vary much when graphical features such as anti-aliasing and anisotropic filtering is on? If it doesn't, there's a possibility that choosing a faster graphics card won't do much good. Again, that particular game or benchmark is quite possibly processor limited.
• Just how much faster is it? As a guide, a frame rate of 30 fps is still quite playable, 45 - 60 is desirable and anything above 60 fps doesn't make that much difference. Of course, a difference of 5 fps below 30 fps is very significant, while the same 5 fps difference is acceptable in the 45 - 60 range and barely noticeable above 60 fps.
• Just what is the fps? Is it the average fps or the maximum / minimum fps? While it is ideal if it is the minimum, it's usually the average fps. Why? Remember that performance in games and applications does not depend on the graphics card alone, but also the PC's processor. We need to know what causes the frame rate to drop to a minimum, is it processor related or graphics card related?
• It's also a good idea to see the difference between minimum and maximum fps if possible, this way we could see whether the game or benchmark spends more time above or below the average fps. If the average is closer to the minimum, then we're going to see lower to average fps when playing and vice versa.

[Previous Page]
[Go to top]
[Next Page]