: News
: Reviews
: Editorials

: Archives
: About
: Contact
: Advertising
: Privacy
: Links
: Forums




NVIDIA GeForce 7950 GX2 Review


More than the Sum of its Parts?


by Josh Walrath


            When NVIDIA re-introduced SLI in its new form with the GeForce 6x00 series, most people regarded it as more marketing than anything particularly useful.  When SLI was first released to the public, it had its warts but in some situations it worked remarkably well.  At that time those positive situations were few and far between, but as the months went along NVIDIA continued to improve the software support, and was able to sign developers on to support native SLI functionality in their applications.  Enthusiasts of all stripes soon started to look at SLI as a viable alternative to a higher end single graphics card system.  The amount of users that achieved greater than 6800 Ultra performance with 2 x 6600 GT’s that cost less overall became significant, and these people were often quite vocal about their accomplishments in gaming and benchmarks.

The reference 7950 GX2 is about as long as the 7900 GTX.  With two PCB's a lot of the components can be stretched out and given a lot of room, so it looks rather sparce from this angle.

            We are now at the two year anniversary of the introduction of SLI, and in those intervening years we have seen SLI go from being a curiosity to a technological and marketing juggernaut.  While NVIDIA did not bet the entire farm on SLI, they did put some tremendous resources behind it.  When NVIDIA decided to pursue dedicated multi-GPU setups, they designed the communication link internally, so no external dongles or chips would be necessary to get it to run.  And unlike the old 3dfx Voodoo 2’s, the signals between the cards would be pure digital vs. analog.  In fact, the new SLI connection has more in common with the chip to chip communication featured in the Voodoo 5 (though the Voodoo 5 series used “digital scanlines” and did not use modes such as Alternate Frame Rendering).  This is not terribly surprising since NVIDIA bought up much of 3dfx’s IP as well as hiring around 100 engineers from that now defunct company.

            The amount of software work that has gone into making SLI a reality is unprecedented.  The first level of this is ensuring that the hardware works together as necessary.  This low level driver work is time consuming, especially when considering that when the GeForce 6 series were introduced, the driver folks had to start from scratch with new hardware to make sure their rendering schemes will work as advertised.  Throughout the years the driver folks have continued to optimize the hardware connection, as well as utilize the PCI-E bus to the fullest extent.  The hardware has progressed as well.  The most evident is when NVIDIA started designing their chips as a native PCI-E device, and used the HSI chip to go from PCI-E based chips to the AGP bus instead of vice-versa.  In the latest 90 nm GeForce 7 chips, NVIDIA apparently pared down the SLI connector circuitry and made it more efficient (plus saving on transistors).

            NVIDIA also takes a lot of time finding the best way to render individual games, and then create “database” entries in their Forceware drivers which will help scale that particular application to the highest degree possible.  So far NVIDIA, if memory serves correct, has put in around 50,000 man hours towards this endeavor.  At this time NVIDIA natively supports 300+ games with the latest 91.47 WHQL drivers.  NVIDIA in fact lists the games that are “SLI Ready” at www.slizone.com.  If a particular game is not supported there, NVIDIA allows the end user to set up individual SLI profiles for un-supported games.  In some cases the games may not take kindly to SLI, and show visual artifacts or decreased performance.  Most times though the user can select AFR or AFR2 and get solid scaling without artifacts.

            SLI has also not sat still in terms of features.  AFR2 was introduced to solve some specific problems that the initial rendering option caused in certain applications, and it is now by far the most used and useful method of accelerating games via SLI.  SLI-AA was also introduced well after the initial introduction, and it allows users to increase the visual quality of many applications which are more CPU bound than graphics bound.  We have also seen performance of this rendering mode increase dramatically from its own introduction, and now the SLI-AA modes are very useful in a wide variety of situations as it does not have nearly the performance hit that it initially did.

            So for the past two years NVIDIA has been diligently working on its SLI technology and to push it past the “enthusiast only” label.  This Spring NVIDIA felt that it had advanced SLI far enough to officially introduce a dual GPU product.  While dual GPU cards are nothing new, the level of performance and integration often left much to be desired in such products, not to mention motherboard compatibility issues.


Next: The 7950 GX2


If you have found this article interesting or a great help, please donate to this site.


Copyright 1999-2006 PenStar Systems, LLC.