   |
|||
|
|||
|
:
News
: Archives
|
|
The Definitive Multi-GPU Roundup |
|
|
Introduction Summer 2006 |
|||
|
|
|||
|
by Josh Walrath |
|||
|
Multi-GPU Overview The now ancient history of multiple GPU rendering goes back to the days of 3dfx, and perhaps because NVIDIA bought most of the tech from 3dfx they were the first to pursue a modern variant of this technology. SLI (Scalable Link Interface) was introduced with the GeForce 6 series of products, and it has been further refined with the latest GeForce 7 series. To enable SLI a user has to have a SLI capable motherboard with 2 x PEG slots (PCI-Express Graphics) and two nearly identical cards. As long as the cards are within the same general family (7600’s could only be used with other 7600’s, same with 7800’s and 7900’s). Even if the cards are from different manufacturers and have different clockspeeds, SLI will work with these cards (though it will normalize the clockspeeds to match the slower card). SLI can be enabled and disabled without rebooting as well. NVIDIA does everything internally and uses both an “Over-The-Top” connector and the PCI-E bus the shuffle data back and forth between the cards. NVIDIA designed the compositing hardware into the GPU’s themselves, and there is no need for an external compositing engine. Quite a bit can be done by using both the OTT connector as well as the PCI-E bus, and we can see that throughout its lifetime SLI has improved not only in performance and compatibility, but NVIDIA has also brought out new features not initially in the release (namely SLI-AA). NVIDIA designed SLI into their products from the ground up, so there is a bit more of a polish to the product than most of the competitors have. Also, the speeds of even the budget cards has required the use of the OTT connector, so all cards from the 7600 GS up will feature this connectivity. NVIDIA has spent a lot of time working on software and driver compatibility, so it has over 300+ games that SLI natively supports with profiles. After the initial launch of SLI, NVIDIA exposed the functionality to enable user created profiles for games that are not officially supported by NVIDIA. Those folks hoping to run System Shock 2 with SLI-AA enabled can certainly do so at their discretion by creating a custom SLI profile for that game. A full list of supported applications can be found at www.slizone.com. There are several modes of SLI that NVIDIA utilizes. While 3dfx used Scan-Line Interleaving either in analog (Voodoo 1/2) or digitally (Voodoo 5), NVIDIA utilizes several modes that can clean up a scene, push the pixels, or scale geometry. AFR is the most common form used, and it stands for Alternate Frame Rendering. Each card is given a frame to render, and it alternates the frames being output from the primary and secondary card. This maximizes both pixel pushing power and geometry scaling, but is not maximal for memory due to the scene being replicated in full by each card. SFR stands for split frame rendering. Each card takes a portion of each frame and renders it. Depending on the complexity of the scene, each card can dynamically allocate a certain percentage of the frame to each card. For instance, if we see a scene with sky taking up 50% of the frame, and complex shading is done on the low half, one card can take upwards of 70% of the frame, while leaving the more shading intensive 30% to the other card. This maximizes pixel performance and memory space, but geometry scaling is not as effective as each card has to handle the geometry for that frame in full. The final rendering feature of SLI is SLI-Anti-Aliasing. This is a very performance intensive operation which combines the anti-aliased frames to improve AA quality and coverage. This should only be used in applications which are more CPU bound, but the AA quality is much improved by allowing a mixture of supersampling and multisampling AA at high levels. SLI still has room to grow. This Spring NVIDIA announced that they will be offloading certain physics effects onto their GPU’s through the use of HavokFX. This means that portions of the GPU, or a GPU by itself, can be utilized to do physics effects work.
Next: ATI's Multi-GPU Technology If you have found this article interesting or a great help, please donate to this site.
Copyright 1999-2006 PenStar Systems, LLC. |
|||
