Tips on Choosing a Video Card
Graphics processors are one of the most quickly advancing areas of computer hardware in recent years. This has resulted in very frequent updates in product lines. For example since 2002, nVidia GPUs have moved from the FX Series to the 6000, 7000 and now 8000 series. Four range updates in five years is not an overly remarkable statistic but the main point of interest is that Microsoft have only released two major versions of DirectX in this period, 9 and now 10 for Windows Vista.
Most PC users who have had experience with gaming or 3D applications may recall that, although optimised for the same DirectX platform as its successors, the FX series offered very poor performance in comparison. It would seem logical for one to update their graphics card in line with new Direct X or OpenGL versions as software required, unfortunately this decision is not that simple as "compliant" or "optimised" does not necessarily mean competent when it comes to performance.
However, there a few standard performance indicators of a card which have remained constant throughout this period and paying attention to these will make choosing the right card for your PC a lot simpler.
Firstly, memory, how much and what type? Presently, lower end cards are equipped with DDR2 memory while mid-range and higher end cards are typically sporting DDR3 at higher clock rates. With any card there's usually 3 memory specs listed by the manufacturer: bandwidth, capacity and speed. Of these three, it is bandwidth that has the greatest effect on 3D performance, not capacity as we may be led to believe by colourful packaging with a great big 256MB! or 512MB! stamped on it.
Really low end cards, which should generally be avoided for anything besides office applications, typically have a 64-Bit memory bus. Examples of cards in this range are the nvidia FX5200, 6200, 7300 and currently the 8400 (in ATI GPUs, the 9200, X1300, HD2400 are in this situation). That's a low end card in four model ranges released over five years, and they all offer pitiful performance due to a 64-Bit memory bus despite being toted as the best performing cheap video card ever seen by man.
If mediocre performance, or a good price/performance point is what you are looking for 128bit was mandatory, though now, in 2007, 256bit would be the recommended minimum for gaming. Memory is the most popular way for card manufacturers to cripple a product to make it fit into the low price end of a product line.
Capacity is the next point of confusion. More memory on a card in 99.9% of cases will not improve performance at lower resolutions. A 128bit 256MB card will still be soundly whipped by a 256bit 128MB card. The advice is, only look for higher capacity cards to run programs at higher screen resolutions, not higher frame rates.
Speed is actually straightforward, DDR3 is faster than DDR2 and so on, excactly like it reads. Will opting for a DDR2 over a DDR3 card result in a huge loss in performance? No, but it may be noticeable depending on how you use it. In this case, the bigger the number the better performance you're getting.
The GPU, that is the chip actually processing the graphics on the card e.g. 7600GS or HD2900, is much like selecting a processor from intel or AMD. The further you go up the line, the more you pay and the better it will perform. Generally the GPUs fall into a Low, Mid and High category. A low end GPU like a 7300GS will perform considerably worse than a mid range gpu like a 7600GT or an 8500GT. A mid range GPU will perform considerably worse than a high end GPU. This may seem like the obvious but this another staple of video cards. There's always the low, mid and high card. The mid card is always the minimum for gaming and the high card is the only one that will remain adequate for gaming for more than 12-18 months, if you're lucky. Therefore, if you like to play games soon after they're released you may well find yourself upgrading a mid range card every six months and a higher end card every 12months.
nVidia vs ATI
Which is better? The answer to this question is that it depends what day of the week it is. One series of GPU by nVidia may be better but the next can be worse. So it pays to do your homework. One difference I have noticed on all recent ranges from both producers is that nVidia drivers tend to be more stable than ATI (a claim also made by id software) though this gap has narrowed greatly and was typically a problem specific to the ATI 9000 series. A second point in regards to the nVidia/ATI battle is that ATI's mid range cards have historically represented better value. For example, the ATI 9600Pro was much lauded for its price/performance in its day and this continued with the X1600. nVidia's 7600GT benchmarked better than the X1600 in general, but while it was current these cards usually cost $50-$100 more and it was a tough call whether the performance increase was worth the extra cash. The best way to choose is to look at the specs and see what you're getting for your money.
Most importantly, PAY ATTENTION TO THE MEMORY! If it looks too good to be true it probably is. The biggest pitfall is the low end card masquerading as a high end card. A classic example was the nVidia FX5500, available on cards in 64-Bit and 128-Bit memory variants. One was a dog, and the other was reasonable. Both had the same thing written on the front of the box. Quite often bottom range GPU cards have outperformed mid range GPU cards simply because they were equipped with a better memory setup, and its the mid range cards that are most often crippled to attempt to grab more low price market share. An 8500 and its only $10 more than an 8400? Better go with the 8500? Not necessarily the 8400 may outperform it. Another example was the nVidia 6600GT vs the plain old 6600 at a time the 6600 was $100 cheaper than the GT and the buyers were 100% more disappointed with its performance due to a combination of differences in clock speed coupled with slower memory setups. Like most things, you get what you pay for and it pays to do your homework.
