Articles: Graphics

Bookmark and Share

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 ]

GeForce 8800 GTS Specification

To see what the two GeForce 8800 GTS models can do we should examine the characteristics of the GeForce 8800 series at large.

All the three graphics card models in this series are based on the 681-million-transistor G80 graphics core and utilize an additional NVIO chip which incorporates TMDS transmitters, RAMDACs, etc. Using such a complex chip in multiple graphics card models belonging to different price categories is not optimal in terms of the end product’s self cost, yet it is not an altogether bad solution: Nvidia can sell defective samples of GeForce 8800 GTX (which can’t work at the required frequency and/or have a few defective subunits) while the self cost of graphics cards selling at $250 and higher is not critical. Nvidia and ATI have both used this policy before. You can recall the G71 chip that could be seen on the massive inexpensive GeForce 7900 GS as well as on the top-end dual-chip monster GeForce 7950 GX2.

So, this is how the GeForce 8800 GTS was built. As the table shows, this graphics card differs greatly from its elder brother when it comes to technical characteristics. It has lower frequencies and a few disabled streamed processors. It also has less graphics memory, a narrower memory bus, and a few inactive TMUs and ROPs.

The GeForce 8800 GTS features 6 groups of streamed processors, 16 ALUs in each, for a total of 96 ALUs. This card’s main market opponent AMD Radeon X1950 XTX has 48 pixel processors each of which consists of 2 vector and 2 scalar ALUs, a total of 192 ALUs.

The GeForce 8800 GTS seems to be weaker than the Radeon X1950 XTX in pure computing power, but there are some factors to be accounted for. First of all, the GeForce 8800 GTS’ streamed processors, like ALUs in the Intel NetBurst micro-architecture, are clocked at a much higher frequency than the other subunits of the graphics core, 1200MHz as opposed to 500MHz. This means a considerable performance boost. Another factor comes from the peculiarities of the R580 architecture. Theoretically, each of its 48 pixel shader processors is capable of executing 4 instructions per clock cycle, not counting a branch instruction. But only two of those instructions can be of the type ADD/MUL/MADD. The other two are always ADD instructions with modifiers. As a result, the R580’s pixel processors cannot deliver their maximum efficiency all the time. As opposed to that, the G80’s streamed processors have a fully scalar architecture and each of them can execute two scalar operations, e.g. MAD+MUL, per clock cycle. Although we still don’t have precise data on the architecture of Nvidia’s streamed processors, we’ll check out in this article how the new unified GeForce 8800 architecture compares with the Radeon X1900 architecture in games.

As for performance of texturing and rasterization subunits, the GeForce 8800 GTS has more such units (24 TMUs and 20 ROPs against the Radeon X1950 XTX’s 16 TMUs and 16 ROPs), but clocks them at a lower frequency (500MHz against 650MHz). So, neither party is handicapped in today’s tests, and the cards’ relative performance in games will depend exclusively on differences in their micro-architectures rather than on a superiority in numbers.

A curious coincidence, the GeForce 8800 GTS has the same memory bandwidth of 64GB/s as the Radeon X1950 XTX, but uses a 320-bit memory bus to access its GDDR3 memory locked at 1600MHz. The Radeon X1950 XTX makes use of 2GHz GDDR3 memory with 256-bit access. Considering that AMD declares an advanced memory controller with a ring-bus topology, it’s interesting to see if the Radeon will have any advantage over its opponent at high resolutions with enabled FSAA, just as it had over the GeForce 7.

The less expensive GeForce 8800 GTS 320MB was announced on February 12, 2007. It is expected to replace the GeForce 7950 GT in the performance-mainstream class and differs from the regular GeForce 8800 GTS in the amount of graphics memory only. In fact, Nvidia created this card by simply replacing the latter’s 512Mb memory chips with 256Mb ones. This solution helped Nvidia establish its technical superiority in the popular $299 category. We’ll see shortly how the reduced memory amount affects the card’s performance and if it makes sense to pay extra $150 for the 640MB model.

The GeForce 8800 GTS 640MB model is represented in this review by the MSI NX8800GTS-T2D640E-HD-OC graphics card. Let’s learn more about this product.

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 ]


Comments currently: 21
Discussion started: 04/12/07 05:41:18 PM
Latest comment: 04/10/08 02:07:35 PM

View comments

Add your Comment