
AMD’s decision not to compete with Nvidia’s best GPU left us puzzled, but the decision its manufacturing partners have made—to pair AMD’s second-tier GPU with a full gigabyte of GDDR4 memory—has us totally stumped.
We concluded in our August issue that PowerColor’s 512MB Radeon HD 2900XT is faster than Nvidia’s 8800 GTS, so we weren’t surprised to discover that Diamond’s card is faster too. But our benchmark testing indicates that doubling the size of the frame buffer—even if it uses fast GDDR4 memory—doesn’t improve performance by much, if anything at all. Frame rates certainly didn’t increase enough to justify the $100 price bump over 2900XT cards with 512MB frame buffers.
The GPU on Diamond’s card is clocked at the same speed as PowerColor’s entry (743MHz), but its memory is set to a cool 1GHz. With the exception of Quake 4, however, we saw an average increase of just one frame per second in our benchmark numbers compared to the aforementioned 512MB card.
We suspect the significant boost in Quake 4 performance—which jumped from 75.7fps on the PowerColor card to 86.2fps on Diamond’s—has more to do with driver improvements than frame buffer size. Performance increased along the same scale in CrossFire mode, with Quake 4’s frame rate benefiting the most (increasing from 127.8fps on the PowerColor card to 146.8fps on the Diamond).
If you use your PC for other high-end applications—such as CAD or 3D modeling—the massive frame buffer might do you some good, but those aren’t applications we focus on.
1GB frame buffer, good Quake 4 scores
Huge buffer not really necessary; better for CAD and 3D modeling than gaming.
| BENCHMARKS | |||||
| Radeon 2900XT 1GB GDDR4 | Radeon 2900XT | Radeon 2900XT 512MB GDDR3 | GeForce 8800 GTS 640 MB GDDR3 | ||
| 3DMark06 Game 1 (FPS) | 22.7 | 44.1 | 21.5 | 19.4 | |
| 3DMark06 Game 2 (FPS) | 21.0 | 46.6 | 20.6 | 17.8 | |
| Quake 4 (FPS) | 86.2 | 146.8 | 75.7 | 65.5 | |
| Fear (FPS) | 64.0 | 105.0 | 63.0 | 52.0 | |
| Supreme Commander (FPS) | 28.2 | 38.7 | 28.0 | 26.2 | |
| Best single-card scores are bolded. AMD-based cards tested with an Intel D975XBX2 motherboard; Nvidia-based cards tested with an EVGA 680i SLI motherboard. Intel 2.93GHz Core 2 Extreme X6800 CPUs and 2GB of Corsair DDR2 RAM used in both scenarios.. | |||||
Links:
[1] http://www.maximumpc.com/user/mrmick
[2] http://www.maximumpc.com/tags/2900xt
[3] http://www.maximumpc.com/tags/buildapc
[4] http://www.maximumpc.com/articles/reviews/hardware
[5] http://www.maximumpc.com/tags/radeon
[6] http://www.maximumpc.com/tags/reviews
[7] http://www.maximumpc.com/tags/videocards
[8] http://www.maximumpc.com/articles/magazine/2007
[9] http://www.maximumpc.com/articles/magazine/2007/november_2007
[10] http://www.maximumpc.com/articles/reviews
[11] http://www.maximumpc.com/articles/reviews/hardware/videocards
[12] http://www.maximumpc.com/articles/magazine