Core architecture

While NVIDIA has integrated the same G70 graphics core that has been used for the GeForce 7800 GTX and GeForce 7800 GT in the past into the GeForce 7800 GS AGP, a number of internal changes have been made to the GPU. The following chart summarizes what’s new with NVIDIA’s latest spin on the G70 GPU:

GPU Features Comparison
GPU	Core Clock Speed (MHz)	Pixel Shaders	ROPs	Pixel Fill-rate	Texture Units	Texel Fill-rate	Vertex Shaders	Memory Speed (MHz)	Memory Bandwidth
GeForce 7800 GTX 256MB	430	24	16	6880	24	10320	8	600	38.4
GeForce 7800 GT	400	20	16	6400	20	8000	7	500	32
GeForce 7800 GS	375	16	8	3000	16	6000	6	600	38.4
GeForce 6800 GT	350	16	16	5600	16	5600	6	500	32
Radeon X1900 XT	625	48	16	10000	16	10000	8	725	46.4
Radeon X800 XT PE	520	16	16	8320	16	8320	6	560	35.8

As you can see, NVIDIA disables two pixel shading quads in the GeForce 7800 GS AGP, resulting in a 16-pixel architecture for the GeForce 7800 GS. On the GeForce 7800 GT, only one quad was disabled. The changes don’t stop there however, as NVIDIA also turns off two vertex shading units (for a grand total of six in the GeForce 7800 GS AGP) and slashes the number of functional rasterizers (ROPs) in half, to a total of 8.

In addition to trimming the number of functional units, NVIDIA also reduces the clock speeds on the GeForce 7800 GS AGP, as the graphics core runs at 375MHz, that’s 25MHz slower than the GeForce 7800 GT and GeForce 6800 Ultra, and 25MHz faster than the GeForce 6800 GT. Paired alongside the graphics core is 256MB of GDDR3 memory running at 600MHz (1.2GHz effective), this is the same speed as the GeForce 7800 GTX, and 50MHz faster than the GeForce 6800 Ultra. This should allow it to perform well at high resolutions, particularly once AA/AF is applied.

BFG 7800 GS and 7800 GTX Reference

GeForce 7800 GTX reference (top) BFG 7800 GS (middle) and 6800 Ultra reference (bottom)

The end result of all these changes is a part that looks on paper to resemble the GeForce 6800 GT/Ultra family more so than the GeForce 7800 line. For instance, the 7800 GS has just as many pixel pipes as the GeForce 6800 Ultra/GT, and is clocked right in the middle of the two GPUs at 375MHz. It also has the same number of texture address units and vertex units as the high-end GeForce 6800 cards do. NVIDIA further cripples the GeForce 7800 GS AGP by disabling half of its ROPs, although in all honesty this probably isn’t as significant as the aforementioned changes. The only area that really didn’t see any reductions is the memory subsystem, which is essentially unchanged from the GeForce 7800 GTX 256MB.

Of course, this line of thinking is all based on paper specs. By now you all should know that NVIDIA’s incorporated a number of changes into the GeForce 7 line that should improve performance on a clock-for-clock basis. The 7800’s vertex shaders have been tuned to speed up geometry processing (for instance, triangle setup is over 30% faster), while the pixel and vertex units that have been tweaked to perform better at handling MADD math operations, which are frequently used in lighting and normal map calculations (among others) in many of today’s latest titles. NVIDIA has also tweaked the GeForce 7800s to scale better at mega-high resolutions like 2048x1536.

Since the 7800 GT and 7800 GTX were so drastically different than preceding GeForce 6800 parts, we never got a chance to test these improvements – we literally couldn’t disable the necessary pixel and shading units to run true apples-to-apples comparisons. But since NVIDIA’s done all of the work for us by disabling these units in the GeForce 7800 GS AGP, we can now finally get a better idea of how the two architectures compare. Keep in mind that this still isn’t a completely accurate comparison, as the 7800 GS AGP has fewer ROPs than the 6800s, but this is the closest we’ve ever gotten to being able to run a direct comparison. We’ll start with 3DMark first.