For the most part, the specs of the 7950 GT directly resemble those of the high-end 7900 GTX. Unlike the 7900 GS we reviewed two weeks ago, which has one of the pixel shader "quads" and one of the vertex shader units disabled, the 7950 GT has the full 24-pixel shader pipelines and 8 vertex units of Nvidia's high-end chip design. It's also clocked a bit higher than the 7900 GS, but a better way to think of it is as a down-clock of the 7900 GTX.
The GeForce 7900 GTX has a reference clock of 650MHz for the core and 800 MHz (1.6GHz effective) on the memory. The new 7950 GT is 100MHz slower on both fronts (individual brands may ship overclocked cards). It still comes with a full 512MB of memory. HDCP (High-Bandwidth Digital Content Protection) support is becoming increasingly more important, so it's good to note that all 7950 GT cards comes with full HDCP support standard, while it is optional on the 7900 GS. What's more, the reference design for Nvidia's new card is a rather small single-slot design with a small cooling unit, so it's a good match for small-form-factor PCs that are tight on slots.
Here's how the specs stack up against the two cards we reviewed a couple weeks ago.
The specs look good, but the numbers tell us that it probably won't outperform the 256MB version of the X1900 XT. Of course, availability is an issue: The X1900 XT is hard to find (though it is priced at the promised $279), while the 7950 GT is pretty abundant at the time of this writing.
Benchmark and Testbed Setup
Just as we compared the 256MB Radeon X1900 XT with a slightly less-expensive GeForce 7900 GS, we'll return the favor with Nvidia, as we pit their slightly more expensive 7950 GT against that same X1900 XT. We used Nvidia's own ForceWare 91.47 drivers and ATI's latest Catalyst 6.8.
Note that our card is a standard Nvidia reference card, clocked at the reference clocks of 550/700 MHz. Most cards at this clock are shipping for $299, so that will be our price reference. Note, however, that some manufacturers are shipping slightly overclocked cards, sometimes at a slight price premium. Here's our test system's components:
Our suite of 3D performance benchmarks are as follows:
- 3DMark06: Futuremark's latest synthetic 3D graphics and game performance benchmark is the only synthetic benchmark we use (that is, the only one that isn't an actual game). It features fairly complex DX9 shaders and high-dynamic-range lighting, so it pushes graphics cards pretty hard and is a fairly good indication of overall relative graphics performance. You can read much more about 3DMark06 and its features in our article on it from last year.
- Prey: Moving on from Doom 3 (it was getting a little long in the tooth), we're now testing the latest and most graphically demanding title to use the Doom 3 engine. 3D Realms' Prey doesn't have a built-in timedemo, so we recorded our own.
- Half-Life 2: The Lost Coast: Valve recently added high dynamic range rendering together with a graphics-intensive showcase level called The Lost Coast. The latest updates add a "Video Stress Test" option in the main menu. We run this and report the score.
- Call of Duty 2: We have migrated from the demo version of Infinity Ward's impressive WWII shooter to the full version, using our own custom recorded timedemo in an intense level during the African campaign of the single player game.
- F.E.A.R.: Monolith's new shooter is one of the prettiest, grittiest, and most graphically demanding games ever. We use the built in performance test to measure average frame rate, with all settings turned up to the max. There is one exception: the Soft Shadows option doesn't work properly with antialiasing. We disable it for all testing.
http://www.extremetech.com/article2/0,1558,2017362,00.asp?kc=ETRSS03039TX1K0000564
