Zotac 9800GTX and SLI review
Author: Luka Rakamaric
Date: 01 Apr 2008

Today we are taking a first look at the new NVIDIA series 9 graphics card, the 9800GTX. So you might ask yourself what does the new GTX bring us, and how much is it going to cost us? We will try to answer that in the following article.The ‘new’ 9 series is, from what we saw in its previous incarnations, just a new marketing name for the G92 chips that were also used in 8800GT and GTS 512. So, there is nothing revolutionary here, the chip is virtually the same, with increased clocks. It wouldn’t be too far from the truth to say that 9800GTX is just an overclocked 8800GTS 512. Perhaps the word ‘just’ is not appropriate, since if you overclock a great card, you will get an even better one, but we are puzzled why didn’t NVIDIA then use the 9 series prefix on all G92 chips, like it did with G80 – series 8, or G71 – series 7.

Design

At first glance, after the experience we had with the recent ‘X2’ cards, the GTX looks quite normal. The cooler is different from the GTS, because the card itself is a little longer. It is based on the same G92 chip that has been around for 6 months already, but with the highest clocks to date. The core clock on the GTX is 675 MHz. Granted, there have been many GTS OC versions that breached the 700MHz margin by a fair amount, but this is the highest reference clocked G92. It has the same 512 MB of GDDR3 memory, which is hooked up to the GPU using a 256 bit wide memory interface. Just to make things clear, the ‘old’ GTX had 384 bits, so this is the downgrade of the GTX name, which has become immensely popular in the past year and a half. Check out our table at the bottom of the page...

So, the 9800GTX is technically not really an improvement over the 8800GTX except for the higher clocks. Everything else has been cut down to reduce pricing. The production process is the only of these things that is actually positive from the performance part as it increases the available clocks. The rest are compromises made to nearly half the price of the new GTX when compared to the old one. This is actually something that we don't consider as being something really negative - NVIDIA is competing against AMD here and it could do very well. So, we have a cooler, higher clocked chip that tries to compensate for the loss of ROPs and memory bandwidth. It still doesn’t manage that convincingly as soon as resolutions go to or more than 1920x1200, because that is where great memory throughput comes into play.
The cooler on the card is a standard fan that pumps air through the ribs and exhausts it out of the case. It is similar to the GTS and old GTX cooler, and is similarly quiet.

The card uses two 6-pin power connectors, although that's a little bit surprising. But it’s better to have them than not if you want to do some serious overclocking. The card also has two SLI connectors, first time after the 8800Ultra. This means that TriSLI can finally slash its price by almost half, at least the graphics part.
Since it is basically a standard card with increased clocks, we are not going to go wide into the architecture of NVIDIA’s DX10 compatible chips. For that you can check our previous articles in the last year and a half.

To benchmark the cards we used our testbed:
- Intel Core 2 Quad QX9560
- ASUS P5N-T Deluxe
- OCZ PC-2 8000 Platinum XTC 2x1GB
- PC Power & Cooling Turbo-Cool 1KW-SR
- Western Digital RaptorX 150 GB

We were actually sent three different cards for the purpose of this review - from Xpertvision and Zotac, both standard clocked so we had a chance to do SLI round of testing as well.

We ran both single card setups and one SLI setup of stock clocked cards on two resolutions, 1920x1200 and 2560x1600. Anything lower than that and you start running into severe CPU limitations, because the vast majority of games don’t support Quad Core CPUs.

So, let’s see that table we discussed earlier: