Specifications

Before we can delve into MSI’s FX5900 TD128, we need to discuss the GeForce FX 5900 core that the card is based on.

Meet Vulcan

Think Tyson could bite his ear off?

Vulcan at work

If you recall the GeForce FX 5800 Ultra launch, you’ll remember that NVIDIA stated its CineFX architecture that the NV30 was based on would be found on all of its NV3x products. This enabled NVIDIA to provide DirectX functionality on its entire range of products; from top to bottom NVIDIA’s offering DirectX 9 support.

Being based on the NV35 core, this same feature set applies to GeForce FX 5900 as well. The key addition is a new 256-bit memory interface, twice as wide as the 128-bit interface found on GeForce FX 5800 Ultra. This allows twice as much data to be transferred each clock cycle, increasing performance in situations that involve anti-aliasing and/or anisotropic filtering, especially when combined with high screen resolution -- environments where the demands on memory bandwidth are greatest.

Last Chance Gas demo

Sunrise

Unlike the GeForce FX 5800 family, for GeForce FX 5900, NVIDIA has gone back to traditional DDR memory. With its wider memory interface, high memory clock speeds are no longer necessary to achieve high memory bandwidth figures. DDRII is also in short supply, with only one manufacturer, Samsung, supplying NVIDIA with modules. As a result, DDRII memory is more expensive than DDR. (We’ve also found that DDRII modules operate considerably hotter than traditional DDR, which is one of the reasons why the GeForce FX 5800 Ultra was such a hot graphics card.)

Other than the new memory subsystem, little is changed from GeForce FX 5800. The core is still based on TSMC’s 0.13-micron manufacturing process, while transistor count is up slightly: 125 million in GeForce FX 5800 versus 130 million in GeForce FX 5900. To address performance in next generation titles that use shadowing extensively (read: Doom 3) NVIDIA has added one new feature, UltraShadow.

Leaded gas only

Looks like truck from NV30 launch

UltraShadow is essentially designed to prevent the graphics core from rendering unnecessary areas within shadows. Previous graphics accelerators would render the entire shadow, regardless if certain areas were visible or not. With Ultra Shadow, game programmers can define the areas of the shadow that need to be calculated, the rest is discarded; this increases efficiency and allows the lighting calculations to be performed more quickly. The end result is greater performance, especially in complex scenes with multiple light sources and objects with shadows, a trait you’ve probably noticed if you’ve looked at a lot of Doom 3 screenshots. Currently, UltraShadow is only exposed through OpenGL, not Direct3D. NVIDIA has told us that they plan to address this however.

Now that we know more about the NV35 graphics core that the GeForce FX 5900 family is based on, the next question is what separates the GeForce FX 5900 from the GeForce FX 5900 Ultra that was announced last month. The key differences are clock speeds and memory size. The GeForce FX 5900 Ultra ships with a 450MHz core clock frequency and is paired with 256MB of 850MHz DDR memory. The GeForce FX 5900 core operates a little slower at 400MHz but memory clock frequency remains unchanged at 850MHz. We’ve provided a chart which summarizes the high-end segment as it stands at this point:

High-End Graphics Feature Comparison
GPU	Core Clock (MHz)	Memory Clock (MHz)	Peak fill rate (Mpixels/sec)	Peak fill rate (Mtexels/sec)	Peak Memory Bandwidth (GB/sec)	MSRP
GeForce FX 5900 Ultra 256MB	450	850	1800	3600	27.2	$499
GeForce FX 5900 128MB	400	850	1600	3200	27.2	$399
GeForce FX 5900 Value	Unknown	Unknown	Unknown	Unknown	Unknown	$299
RADEON 9800 PRO	380	680	3040	3040	21.8	$399
RADEON 9800 PRO 256MB	380	700	3040	3040	22.4	$499
RADEON 9700 PRO	325	620	2600	2600	19.8	N/A