AMD made no apologies when it announced the ATI Radeon HD 2900 XT last May: Their new GPU wouldn’t compete with Nvidia’s best, and that was fine with them. “Nvidia can have the high end,” they seemed to say. “We want to build the most powerful videocard that the masses can afford.” The 2900 XT was indeed faster than the Nvidia’s 8800 GTS, but it was leagues behind the 8800 GTX and even farther behind the 8800 Ultra.
Nvidia, meanwhile, gave their engineers new marching orders: Come up with a GPU that’s faster than AMD’s best—one that we can sell for less than the 8800 GTS. The chip that emerged is the result of a die shrink and a 65nm manufacturing process (previous GPUs in 8800-series are 90nm parts), and it’s a major breakthrough in terms of price/performance ratio.
This 754-million transistor monster not only delivers better performance than the 2900 XT, it also offloads the entire HD-video decode process from the host CPU (you must step down to AMD’s 2600 XT to get that feature). It also provides HDCP decryption on both DVI links, so it’s capable of displaying Blu-ray and HD DVD movies at the native resolution of a 30-inch panel. The GPU is compliant with PCI Express 2.0, but it doesn’t offer support for DirectX 10.1 and Shader Model 4.1 (although I don’t think any of those three facts amount to a pile of dead pixels right now).
Nvidia’s 8800 GT reference design packs 112 stream processors, 512MB of GDDR3 memory, and a 256-bit memory interface into a GPU that requires a single-slot cooler. The core runs at 600MHz while the shader processors hum along at 1.5GHz; memory is clocked at 900MHz. Third-party vendors such as Asus, BFG, EVGA, and XFX quickly announced products with retail prices starting at $250. Nvidia expects third-party vendors to also offer 256MB versions cards that will sell in the $200 range.
We’ve only just started to probe the 8800 GT’s capabilities, but we’re very impressed with what we’ve seen so far. There’s no other way to say it: Nvidia’s new GPU poops all over AMD’s current best effort, delivering more performance and crucially important features at a price point that’s hundreds of dollars lower. The 8800 GT is faster than both models of the 8800 GTS, too, despite having a narrower memory interface than both it and the 2900 XT (the 2900 XT has a 512-bit interface to either 512MB or 1GB of memory; the 8800 GTS has a 320-bit interface to either 320- or 640MB of RAM).
Now the big question is can AMD’s upcoming RV670 manage to outdo the 8800 GT? I’ll be able to provide the answer in the next couple of weeks; in the meantime, I think Nvidia is going to sell a boatload of 8800 GTs.
SINGLE-CARD BENCHMARKS
| RADEON 2900 XT (512MB) | GEFORCE 8800 GT (512MB) | |
| 3DMARK06 GAME 1 (FPS) | 22.6 | 26.4 |
| 3DMARK06 GAME 2 (FPS) | 21.0 | 20.3 |
| QUAKE (FPS) | 85.6 | 83.7 |
| FEAR (FPS) | 66.0 | 71.0 |
| SUPREME COMMANDER (FPS) | 27.9 | 29.1 |
DUAL-CARD BENCHMARKS
| RADEON 2900 XT (512MB) | GEFORCE 8800 GT (512MB) | |
| 3DMARK06 GAME 1 (FPS) | 44.1 | 47.0 |
| 3DMARK06 GAME 2 (FPS) | 42.3 | 37.2 |
| QUAKE (FPS) | 145.7 | 101.3 |
| FEAR (FPS) | 113.0 | 120.0 |
| SUPREME COMMANDER (FPS) | 44.7 | 33.4 |
Best scores are bolded. AMD-based cards tested with an Intel D975BX2 motherboard; Nvidia-based cards tested with an EVGA 680i SLI motherboard. Intel 2.93GHz Core 2 Extreme X6800 CPUs and 2GB of Corsair DD2 RAM used in both scenarios.
Share to Facebook
Tidak ada komentar:
Posting Komentar