Intel in a dilemma - to SLI or not to SLI

OUR THEO and Charlie discovered some upcoming workstation boards based on the Seaburg dual-FSB1600 chipset for Harpertown quad-core 45 nm Penryns - SLI was mentioned as one of the features there.

It made perfect sense, since those boards have much higher PCI capacity - dual PCI-E x16 v2 slots for graphics, plus a few more shorter ones for other I/O. And, as our team reported earlier, some kind of Intel-Nvidia licensing deal already happened - if Nvidia got FSB1600, Xeon dual-FSB and Nehalem CSI licences, surely it makes sense that at least Nvidia passed the SLI license for the Intel chippery?

Well no. We all had an interesting exchange with Derek Perez* from Nvidia over the weekend, with him vehemently denying the SLI licence - it would either mean that Intel acted as a Santa Claus to Nvidia, which would never happen, or that Nvidia has some truly good platform stuff coming up for Intel, worthy enough that Intel doesn't bother pushing them to take SLI for its own chipsets.

My take. It's Case 2. While X38 chipset does overtake Nforce 680i in some things, like PCI-E v2, cooler operation and DDR3 memory, Nvidia is rushing out a 65 nm successor to Nforce 680i, supposedly planned for around October, which will incorporate all of these plus maybe more - how would you like a hardware XOR engine for RAID 5, for instance?

Coupled with, I suspect, again a superb memory controller, good hardware-accelerated dual Ethernet and so on, it would be of some benefit to Intel to have such a good chipset out. But then, having SLI on Intel's top end X38 chipset may kill off the business case for Nvidia's new Nforce in that segment. So, X38 may be for the Crossfire fans then, while the SLI ones go for the Nvidia platform altogether.

Then, neither Nvidia nor Intel are negating the old story about an ultra high end dual-FSB dual-socket Nforce platform for Xeons. This was talked about almost a year ago, and now, with the Xeon dual FSB license, it became legally doable.

The games are now increasingly multithreaded, and even 3DMark 06 CPU can get an extra 40% moving from 4 cores to 8 cores, in my V8 tests previously. That's a good sign for a piece of code which probably never had 8 cores of dual Clovertowns (or soon Harpertowns) in mind. On top of it, we got a steady workstation market with healthy growth, and even HPC visual servers for parallel 3-D or GPGPU use.

So, since even gaming rigs these days can cost 6 grand in the States, or 4 kiloquid in Londinium, what's the cost to add an extra 4-core socket? AMD already promoted it in the "few units around only" QuadFX platform, and now is the time for Intel to do the same across its high end.

Nvidia now can go ahead officially with a nice, overclockable, well tuned, low latency high bandwidth dual-FSB Intel chipset if they want to - with or without SLI advantage, it should be ahead of Seaburg on performance and features - at least, no FB-DIMMs there. Of course, such platform would help Intel fight AMD at the workstation and multithreaded gaming level, where AMD's architecture still has the platform-level advantage - especially on memory bandwidth and latency.

So, I feel Intel has given Nvidia some leeway on this to get the job done - and that might include not demanding the SLI freebie. After all, it's not as everyone rushes after SLI or CrossFire - in quite a few cases, whether it is dual 8800 Ultras or some workstation stuff, SLI brings more problems than benefits, for now. I've heard enough feedback about SLI's questionable use from vendors and users alike. µ

However, knowing the (lack of) trust between the two, I strongly believe Intel has an alternative high-performance dual-socket high-end PC chipset coming up soon too, just in case Nvidia plays unusual games or simply turns fickle. Skulltrail is just one of the codenames circulating around, besides San Clemente and few others. After all, between now and late 2008, when Nehalem and its CSI is supposed to arrive, there is well over a year to capture more market share in the profitable segments, isn't it? At the end, that may be the first year where Intel will again have the complete CPU, chipset and discrete graphics line coming out... so, to them, SLI might not even matter by then. µ

* NVIDIA is actually talking to the INQUIRER? Shorely shome mishtake. Ed.