The graphics manufacturers are usually pretty predictable in terms of the product launches; there is usually an introduction of a part in the spring/summer timeframe and a product refresh in fall/winter. Although there was the announcement of a 512MB GeForce 6800 Ultra in this February, that was not really a refresh, but the introduction of SLI can be read as such. The last few months have been filled with rumors about new cards as the spring rolled around. Rumors on the NVIDIA front had been contradictory at best with some sources suggesting that there would be a NV40 refresh while others pointed to a whole new architecture. In any event, it has been over a year since the debut of the 6800 Ultra and there has not been another single board, mainstream NVIDIA part that has superceded it until today; the 7800 GTX, codenamed G70.

Specifications

Because of the popularity of the 6600GT and the 6800GT, NVIDIA has decided to drop the Ultra moniker that has been around since the days of the TNT2; hence the new flagship 7800 GTX name. The first hint that this is not merely a product refresh is that this is no longer a GeForce 6 generation product. In almost all of NVIDIA's presentations there is an emphasis that the 7800 GTX is not merely an architectural update to the NV40 family but a new architecture built from the ground up. In fact, this will be the basis of RSX, the graphics chip which will power the Playstation 3. The four key design goals of the G70 are the following -

• Parallelism
• Shader Arithmetic Density
• High Dynamic Range (HDR)
• Power Efficiency

As the preview progresses, each of these topics will be touched on in more detail as the 7800 GTX is examined more closely.

The chart above is a quick summary look at how the 7800 GTX compares to some cards of the previous generation but do keep in mind that there are a number of changes in hardware that goes beyond what is outlined. The first thing to the increase in the number of pipelines from 16 to 24 and along with that, the massive increase in the number of transistors from 220M to 304M. To put things into perspective, AMD and Intel's dual core chips weigh in at just over 230M transistors each and a large part of the die space is cache whereas all of the 304M transistors on the 7800GTX is logic. Graphics are inherently parallel, meaning that the workload can be divided up between functional units easily, which is why there is always an emphasis on increasing the number of pipelines. Something not shown in the specifications chart is the number of geometry units which has increased from 6 on the 6800 series (4 on the 6600) to 8 on the 7800 GTX. Game developers are encouraged to write more complex shaders that tax arithmetic abilities of modern graphics cards more so than geometry, hence, proportionally, the number of vertex shaders is not increasing at the same rate as the number of pixel pipelines. The amount of onboard memory stays put at 256MB; while we saw some benefits in moving up to 512MB of memory while testing ATI's X800 XL 512MB, for the majority of cases, the justfication for extra memory was not there. When questioned, ATI themselves admitted that their next generation part may not necessarily ship with 512MB of memory either.

Clockspeed and memory bandwidth however are two areas that have not grown as much in comparison; there is merely a 7.5% increase in core speed and 9% increase in memory speed. NVIDIA feels that adding pipes is, for the foreeable future, the method which will net the greatest scaling in performance. The same for which cannot be said about clockspeed something the current state of affairs on the processor side seems to confirm. Although there is an ever growing disparity between processing power and memory bandwidth, NVIDIA is stressing greater efficiency with the 7800 GTX - by going wider and deeper arithemetically, fewer passes are needed to render a scene and as a result there is less traffic on the memory bus. This will be explored in a bit more detail in upcoming sections, but first a look at the card itself before we take a peek under the hood as we and explore the updates to the graphics pipeline.