SLI: Before Its Time?

We are starting to get numbers that give an accurate view of what SLI can and cannot do. You can see some here and here

If you look at the numbers and look past the hype; it’s very hard to say SLI helps a whole lot at anything less than 1600X1200 with all the eyecandy turned on. Certainly nothing that would fit into even a vague notion of bang for the buck.

SLI isn’t just a matter of buying two video cards. It’s also a matter of buying a (more expensive) SLI-capable motherboard, and in many cases, a new, bigger monitor.

When you look at the commentary around the benchmarks, you keep seeing the term “CPU-limited,” “CPU-limited.” When you realize that the limiting CPU is an $850 processor and that no single CPU coming out in the foreseeable future will be seriously faster than that, this is not too encouraging.

In short, for the most part, in the single CPU desktop of today, SLI is a mostly stunted technology. Not bad, just premature.

Well Then, What About Dual-Core?

Well yes, what about dual-cores? Potentially, this is the light at the end of the tunnel for SLI. If a CPU has to handle half a screen output rather than all of it, that ought to push back any CPU limitations quite a bit, even given the likelihood that dual-core processors will run rather slower than the single CPUs of today.

However, there aren’t going to be any terribly affordable dual-core processors any time soon; realistically, late 2006/early 2007. While the initial 90nm dual cores may well run on some of the motherboards you’ll be able to buy shortly, it would probably be unwise to bet that this will be true by the time 65nm dualies show up. For one thing, DDR-2 (if not three) will probably be the universal memory standard by then, and that will certain require a mobo replacement.

Of course, a game has to be written to split up the workload that way, but that shouldn’t be too big a deal given that the video workload is already being split up with SLI. Nonetheless, it has to be done, and odds are, in most cases, it’s not going to get done until we’re well into the dually era.

Yet another factor which comes into play is how fast video cards are likely to get the next few years. Will it be business as usual, or will they run into their own heat/power wall, too?

On the one hand, since the circuits themselves run much slower than CPU circuits, GPU manufacturers shouldn’t run into the new physics problems the CPU folks face. On the other hand, video cards are already voracious heat-generators, and in the end, it doesn’t matter whether you generate several hundred watts of heat with a few fast or a lot of slow circuits, you still are faced with a heat wall.

Best guess is that the video card folks have a couple generations to go there, so two 2004-era video cards may not look too appealing in 2007.


Those with money to burn and/or egos to boost will pay no mind to any of this. They’ll be willing mice to the video Pied Piper and helpers.

For the rest of you, for those who cringe at the thought and cost of this kind of upgrade, for those who don’t want to spend multiple times what they’re used to spending on a video card, or perhaps an amount close to what they’re used to spend on a whole box, you shouldn’t feel like you have to take the leap yet.

You can buy a single video card now (you can buy an SLI-capable one as insurance), and so long as you don’t play at 1600X1200 with maximum eye candy, you’ll be fine for a few years, and won’t pay a lot more for not all that improvement.

By 2007, when dual-cores are common place, and games are written for it, and SLI truly begins to shine, then will likely come the optimal time to shift over.


Be the first to comment

Leave a Reply