Overall Thoughts
It's taken me until the end of this third article in our SLI series to get to any kind of large conclusion on SLI as a whole. My overall impression of just the technology is one of excitement, and also admiration that NVIDIA have it working to such a good degree with the majority of applications. Multi-GPU 3D acceleration is not a trivial problem to solve, especially in the load-balanced SFR case, with it only really working across the board if you throw the AFR mode at everything. And even that has its problems.So a large amount of respect for NVIDIA for seeing it through to completion. SFR, the most publicised SLI mode, regardless of whether it's the recommended way of doing things any more, works well in a range of popular games. That it's an application agnostic technology (for the most part), requiring little developer work providing they follow good GPU programming practices, is another strong point. Having NVIDIA and their large resources bear the brunt of the work in having SLI execute properly is a good thing. The harder it is for developers to make use of it, the harder it would be to gain large market acceptance.
It also allows NVIDIA some headroom in terms of mindshare that comes from having the fastest 3D solutions on the market. In some cases SLI will buy you nothing, and a pair of SLI-able 6800 Ultras may very well be slower than a single ATI Radeon X800 XT Platinum Edition (or similar), but that's not true in the majority of cases. In absolute terms, NVIDIA can rightly claim to have the fastest consumer 3D solution in the world, and by a large margin.
However, as with all initial technology releases, exciting and speedy and spectacular or not, it's for nothing if the consumer fails to adopt it and make good use of it. In some respects, SLI's success really depends on the next round of SLI-able graphics processors, giving more choice in the GPU sense and time from now for NVIDIA's mainboard partners to create affordable designs. Adopting it right now almost seems folly for everyone but the early adopter, but there's plenty of those people out there lusting for a slice of dual GPU action.
I thoroughly recommend it as an enthusiast, excited about the technology and what it gives you. I've spent time fawning over Half-Life 2, Doom3 and others recently, on FX-55 with dual Ultras, and as a geek and enthusiast, I think it's fantastic. There wasn't time to me to cover it for these three articles, but I'd urge other SLI reviewers to check out NVIDIA's 8X mixed-mode anti-aliasing if they've been ignoring it or have forgotten it until now. Also, playing at resolutions of 2000 pixel width or higher in recent titles becomes an option for pretty much the first time, and I even played some Doom3 with 16X antialiasing (where the AA is done in the fragment shader) at 1024x768 and it was decently playable. Quad-head is another feature that's worth exploring (although SLI needs to be disabled for that).
So there's a lot extra that SLI does that I haven't been able to cover for these articles, but that we will go over in the future.
To sum up, nobody was ever under any illusions that it'd be cheap or readily available in terms of parts, soon after release, and the upgrade path options need very careful consideration. But it works, and it works well. So, if you can stomach the caveats needed for successful implementation and you understand its limitations, and most importantly you have, or will have, the cash, go forth and seek it out. Truly awesome framerates await.
As I leave you dribbling over the thought of dual graphics cards in your own machine, the final part of my consumer SLI article series will focus on a pair of SLI workstations from an SI partner of ours, before I take a look at Quadro SLI for the professional user in coming weeks. Look out for those articles in due course.
