Single page Print

A sweet 16 mid-range graphics cards compared

Making sense of the biggest sweet spot we've ever seen

We are living in perhaps the most exciting time in recent history for PC graphics. What's particularly striking about this latest flurry of excitement is that it isn't tied to a new standard in performance set by high-end "halo" cards that few enthusiasts can actually afford. Instead, it's being driven at an almost grassroots level, fueled by the class of graphics cards that most of us seem to end up actually buying—cards in the $200 range. These cards don't incite as much geek lust as those higher up the line, but they're far more attainable and often boast a better value proposition than the latest flagships.

I speak, of course, of the GeForce 8800 GT and Radeon HD 3800 series—the latest mid-range contenders battling for sweet spot supremacy. Launched not long ago and finally available with some semblance of consistency and adherence to prescribed pricing, these cards are selling like hotcakes. Given the fresh bounty of quality—and more importantly, demanding—PC titles we've been treated to over the last few months, it's no wonder gaming-worthy graphics cards are such a popular upgrade item.

If you've read our extensive coverage of the GeForce 8800 GT and Radeon HD 3800 series, you probably have an idea of which one best suits your needs and budget. That puts you one step closer to settling on the right card, but there's another important decision to consider: among the 8800 GTs and HD 3800-series cards available from various add-in board partners, which are the best? There's more variety here than one might think, since manufacturers tweak clock speeds, board layouts, cooling solutions, and even bundled games and extras in order to separate themselves from an increasingly crowded pack of contenders drawing from the same bin of chips.

Set on finding the finest mid-range graphics cards on the market, we've rounded up an appropriately sweet 16 cards from the likes of Asus, Gigabyte, HIS, MSI, Palit, PowerColor, Sapphire, VisionTek, XFX, and Zotac. Read on to see which set themselves apart from the masses.

A trio of options
We start at the core with the RV670 and G92 graphics chips powering this latest salvo of mid-range wunderkinds from AMD and Nvidia. Since this round-up will focus on the unique attributes of cards being offered by various add-in board partners, we won't dwell on the architectural intricacies that make these chips tick. For an in-depth look under the hood of each, I highly suggest reading our initial reviews of the Radeon HD 3800 series and GeForce 8800 GT.

Most of what you need to know about the Radeon HD 3800 series and GeForce 8800 GT can be summed up in a couple of handy charts. This first one covers the basics.

Graphics chip Core clock speed Shader clock speed Memory clock speed Typical memory size Memory bus width Reference cooler
GeForce 8800 GT G92 600MHz 1.5GHz 1.8GHz 512MB 256-bit Single-slot
Radeon HD 3850 RV670 670MHz NA 1.66GHz 256MB 256-bit Single-slot
Radeon HD 3870 RV670 775MHz NA 2.25GHz 512MB 256-bit Dual-slot

As you can see, there are two members of the 3800 series: the 3850 and the 3870. Both use the same RV670 graphics chip, differing on clock speeds and memory configurations rather than GPU silicon.

Before going any further, we should note that the specs above are those of reference designs from AMD and Nvidia. Add-in board partners are free to take liberties with clock speeds, memory configurations, and cooling solutions, and we have numerous examples of each in the round-up. Don't worry about the lack of shader clock speeds for the Radeons, either; they don't have independent shader clocks.

One of the most obvious differences between these competing offerings comes in the clock speed department, where the GeForce 8800 GT looks a little ill-equipped to take on the Radeons. The GT's 600MHz core clock is 70MHz lower than that of the 3850 and a good 175MHz short of the 3870. Don't read too much into core clock speeds, though. Graphics architectures differ in the amount of work they can complete in a given clock cycle, and as you'll see in a moment, the GT doesn't necessarily need higher clock speeds to be competitive with even the 3870.

Things are a little different on the memory front, where all three cards use a 256-bit bus to talk to DDR memory. Memory bandwidth isn't as dependent on GPU architecture as it is on clock speed and bus width, giving the Radeon HD 3870 a big leg up on the 8800 GT. Even the 3850's 1.66GHz memory clock is within shouting distance of the GeForce. In all fairness, however, AMD only calls for 256MB of memory to be paired with the 3850—a potentially serious handicap at higher resolutions with the latest games.

AMD's RV670 graphics chip

Nvidia's G92

Combining clock speeds with key characteristics of each graphics chip's underlying architecture allows us to calculate peak theoretical performance expectations, which we've presented below.

Peak pixel fill rate (Gpixels/s) Peak bilinear texel filtering rate (Gtexels/s) Peak bilinear FP16 texel filtering rate (Gtexels/s) Peak memory bandwidth (GB/s) Peak shader arithmetic (GFLOPS)
GeForce 8800 GT 9.6 33.6 16.8 57.6 504
Radeon HD 3850 10.7 10.7 10.7 53.1 429
Radeon HD 3870 12.4 12.4 12.4 72.0 496

Relatively slower clock speeds hinder only the GeForce 8800 GT's peak pixel fill rate. Otherwise, the GT offers higher filtering and shader capacity than the Radeon HD 3870. Peak shader arithmetic stats are relatively close between the two, although estimating relative shader throughput for different architectures isn't an exact science.

Here we can also see the Radeon HD 3870 leverage its higher clock speeds to offer substantially more memory bandwidth than the GeForce 8800 GT. A 14.4GB/s gap is a lot to make up, and things are about to get a lot more interesting as we dip into retail cards that benefit from factory overclocking.