Hands-on with the First
DirectX 10 Videocard
S
o, Nvidia has won the race to produce the fi rst DirectX 10 videocard. The GPU
and the card on which it’s mounted are physically huge, there’s tons of memory,
and the list of specs and features is impressive. But the only way to tell what
Nvidia has wrought is to test the bugger, and we have—as much as is possible.
You’ll fi nd a more in-depth look at the GeForce 8800 GTX and the GeForce 8800
GTS in the Reviews section, but we couldn’t resist providing some highlights of our
time with the top-shelf GTX version here. This GPU is certainly fast with DirectX 9
games. In fact, a single 8800 GTX proved to be faster on every benchmark than a
single Radeon X1950 XTX, GeForce 7900 GTX, or GeForce 7950 GX2. It also deliv-
ered better benchmark results than two X1950 XTXs running in CrossFire and a pair
of 7900 GTXs in SLI.
But ATI’s Radeon X1950 XTX and Nvidia’s GeForce 7900 GTX are arguably fast
enough for today’s games—even at the 1920x1200 resolution we test at. And since
Vista has not yet arrived and there are no DirectX 10 games available, just what
benefi t is to be derived from even more speed? We found that the answer lies not in
sheer frame rate but in image quality. You might have to wait for games developed
for Shader Model 4.0, but you won’t have to wait to turn on or turn up all the visual
enhancements for today’s games running Shader Model 3.0.
Nvidia’s 7-series GeForce cards, for instance, can’t run high dynamic-range light-
ing and perform antialiasing at the same time. The 8800 can, and while the GPU took
a performance hit when we enabled 4x antialiasing and 16x anisotropic fi ltering in
3DMark06, it still delivered faster benchmark results than a GeForce 7950 GX2 run-
ning with zero AA and only 8x aniso.
ATI’s Radeon X1900 series doesn’t suffer from such a limitation, but ATI’s top-end
part can’t hold a candle to the 8800 GTX while performing both these tasks at once.
When we enabled soft shadows, 4x antialiasing, and 16x anisotropic fi ltering in FEAR ,
for example, the 8800 GTX delivered a benchmark result that was fi ve frames per
second faster than the X1950 XTX’s score while running without any AA at all.We won’t know how well the GeForce 8800 series performs with Direct3D 10 games until we have some Pixel Shader 4.0 games
to test it with. But, we know this card is a barn-burner when it comes to running today’s Direct3D 9 games and benchmarks; in
fact, it’s faster than two super-fast 7900 GTX cards running in SLI in most situations.In the absence of Vista, DirectX 10, Shader Model 4.0,
and games designed to run in such an environment, all
we can do to measure the prowess of Nvidia’s GeForce
8800 GTX is to run benchmarks designed for DirectX 9
and Shader Model 3.0. And based on the numbers we’re
seeing, Nvidia has much to be proud of.
The new chip is not only faster than any preceding
GPU, it’s dramatically faster. In nearly all our bench-
marks, we realized a 2x increase or came very close
to it. Game 1 and Game 2 from 3DMark06 use Shader
Model 3.0 for all its worth, so we’re accustomed to
seeing low frame rates because we test at such a high
resolution: 1920x1200. To see a performance delta just
shy of 2x in Game 2—and more than 2x in Game 1—is
nothing short of remarkable.
Game 2 depicts many thin horizontal and vertical
lines, which look extremely jagged without antialiasing.
We couldn’t run this benchmark with AA turned on with
previous generations of Nvidia GPUs, so we didn’t with
ATI’s, either. Similar limitations apply to benchmarking
FEAR and Company of Heroes. Now that the GPUs from
both camps support the simultaneous use of HDR and
AA, we’ll enable both features in every benchmark that
supports them.
We’ll also run Quake 4 in Ultra Quality mode when-
ever we test videocards with 512MB or larger frame
buffers. This mode uses higher-quality textures to deliv-
er much-improved image quality. We’ll continue to test
with 4x AA when the benchmark offers us the choice,
but we’ll boost anisotropic fi ltering to 16x.
We’re also moving our videocard testing from a
2.6GHz AMD FX-60 platform to one based on Intel’s Core
2 Extreme. Time constraints prevented us from retesting
the GeForce 7900 GTX, GeForce 7950 GX2, and Radeon
X1950 XTX cards with Conroe, but we did benchmark
the 8800 GTX with both platforms for an apples-to-
apples comparison.HOW WE TEST
BENCHMARK RESULTS
TEST BED: AMD FX-60 (2.6GHZ)
ASUS EN8800 GTX XFX 7900 GTX XXX EDITION XFX 7900 GTX XXX EDITION SLI EVGA 7950 GX2 ATI RADEON X1950 GTX
3DMARK06 HDR/SM3 #1 (FPS) 33.6 15.0 27.2 19.3 18.
3DMARK06 HDR/SM3 #2 (FPS) 38.2 20.7 38.2 27.3 20.
QUAKE 4 (FPS) 98.4 66.7 113.8 89.3 66.
COMPANY OF HEROES (FPS) 77.2 38.1 DID NOT TEST DID NOT TEST 32.
FEAR (FPS) 71 42 70 57 35
HQV SCORE 113 88 88 88 118SOFT SHADOWS ON, AA SOFT SHADOWS ON,
DISABLED, 8X ANISO 4X AA, 16X ANISO FRAME-RATE HIT
FEAR 77 43 43%Top benchmark: Best scores are bolded. Tested with an Asus A8N32-SLI motherboard with a dual-core 2.6GHz Athlon 64 FX-60 CPU and 2GB of Corsair DDR400 RAM. Bottom benchmarks: Tested with an EVGA nForce 680i SLI
motherboard with a 2.93GHz Intel Core 2 Extreme X6800 CPU and 2GB of Corsair DDR2 RAM. All benchmarks run at the default resolution of a 23-inch ViewSonic VP2330wb display: 1920x1200.32 MAXIMUMPC JANUARY 2007
HIGH QUALITY, 4X AA, ULTRA QUALITY, 4X
8X ANISO AA, 16X ANISO FRAME-RATE HIT
QUAKE 4 102.9 93.2 9%TEST BED: INTEL CORE 2 EXTREME X6800 W/ASUS EN8800 GTXNEXT-GEN GAMING