Newcomers of This Season: Unrivaled NVIDIA G70
Unlike in the same period of the last year when two new graphical architectures were brought to market, only NVIDIA made the same in the first half of 2005. On June 22, the company introduced the G70 chip and the G70-based GeForce 7800 GTX graphics card to the public.
NVIDIA changed the nomenclature of its GPUs and the chip that had earlier been referred to as NV47 got the new name of G70 which indicated its belonging to the GeForce 7 series. The NV40 chip had earlier set a record as consisting of as many as 222 million transistors, but the G70 turned to be a much more complex die – 302 million transistors! That was an astounding level of complexity for consumer graphics hardware, yet quite explicable: the number of pixel pipelines had been increased from 16 to 24 by G70, and of vertex processors from 6 to 8. The number of raster operators remained, however, the same (16 ROPs), as the developers must have thought it unreasonable to increase it to 24.
Then, each pixel pipeline got two additional mini-ALUs besides the two existing ALUs and that addition increased the complexity of the chip even more. The die area of the G70 got larger in comparison with the NV40/45, even despite the use of a thinner 0.11-micron tech process. The power consumption and heat dissipation remained almost on the same level, however.
The frequency of the new graphics processor was a little higher than that of the previous model and equaled 430MHz, but only for the pixel pipelines. As we wrote in our earlier review, the G70 chip was the first solution in the consumer 3D graphics market to feature independent clocking of its subunits. That is, the pixel pipelines of the GeForce 7800 GTX work at 430MHz, but the ROPs and vertex processors of that card are clocked at 470MHz. This independent clocking feature gives more flexibility in controlling the GPU than usual and also helps in making a well-balanced product.
Generally speaking, the G70 was not a completely new solution, despite the fact that NVIDIA claimed otherwise. From the architectural standpoint this chip is rather a greatly improved NV40/45. The functionality of the new chip remained the same (i.e. the widest available on the market), but its performance improved considerably (you can refer to our special report for more details about the G70 architecture).
The new family of graphics cards from NVIDIA, based around the G70 GPU, was quite logically named GeForce 7800. Right now this family includes two graphics card models with the following characteristics:
- GeForce 7800 GTX (24 pixel pipelines, 16 ROPs, 8 vertex processors, 256MB GDDR3, 430/470/470/1200MHz, 2 DVI-I outputs, PCI Express x16, $599)
- GeForce 7800 GT (20 pixel pipelines, 16 ROPs, 7 vertex processors, 256MB GDDR3, 400/440/440/1000MHz, 2 DVI-I outputs, PCI Express x16, $449)
We will surely see new products in this family in near future, but so far the rest of NVIDIA product line-up is represented with graphics cards on NV40/45 and NV41/42 chips. The latter have 12 pixel pipelines and seem to differ in tech process alone (the NV42 is a 0.11-micron chip).