Digitimes: NVIDIA-VIA merger talks failed but may resume. [Analysis]
Wednesday 19th March 2008, 02:03:00 PM, written by Arun
Multiple options were considered including only selling the x86 processor business, which obviously VIA wouldn't agree to given that most of its other divisions are dependent on it. Interestingly, a stategic alliance was also discussed according to Digitimes, which might have been an effective way to handle the x86 license issue, which is NOT transferable but should be renegotiable especially given that NVIDIA is based in the USA and so is VIA's x86 design team. Intel has no legal right to keep their business monopolistic or duopolistic, they can only ask for fair compensation in exchange for that IP.
I have been a public proponent of a NVIDIA-VIA merger for nearly two years, and given how much sense it would make for the right price (~$500M net of cash we'd wager) this is about what I was expecting. On first sight, it may seem arrogant for them to think they don't need a x86 core unless it's very cheap - but the reality is likely much more complex than that although it can be summarized in a single word: Montalvo.
From a significant amount of Googling and analysis, our conclusion is that Montalvo's heterogenous architecture has two kinds of very different cores: likely one that is VIA C7 or Silverthorne-like, and one more VIA Isaiah or Intel Penryn/AMD K10-like. The former should at the very least be viable for MIDs and ultra-high-end smartphones, while the latter should be viable for desktop and laptops. The combination of the two in a single chip improves perf/watt and perf/mm² for multi-threaded workloads as smaller cores are typically much more efficient.
What does that mean for NVIDIA? Well, Montalvo is a processor company, not a southbridge company. Intel, AMD and VIA are unlikely to make those compatible with Montalvo's processor, so that leaves only NVIDIA and SiS as potential partners. And obviously the former is a much more likely choice here (although SiS is far from impossible I'll admit), especially given that the chipset needs to be an IGP. So this and the fact NVIDIA is a likely acquirer either way makes seem probable that the company has an inside track into Montalvo and will be amongst the very first to know about real-world performance for the recently taped-out chips.
If Montalvo's design delivers, they'll just buy them. If it doesn't, they'll take another hard look at VIA. If neither option pleases them, they'll either wait, make a hard decision and buy one of the two, or bite the bullet and buy both to combine the design teams. Montalvo should be cheaper than VIA (maximum $250M?) but they also don't have any southbridge, graphics or analogue expertise. All of those are hardly worth nothing, even though they would either be semi-redundant to NVIDIA or not match their strategy very well.
We don't know if the above analysis is perfectly correct - but everything does seem to point in that direction. If so, it's a fairly attractive risk management strategy and we certainly couldn't blame Jen-Hsun for not yet making his move. It doesn't mean their strategic position is optimal though (in the low-end, it clearly is not) or that they have an uncontested history of good and effective roadmaps (MCPs, anyone?) - but it certainly looks like competition is going to heat up on many fronts in the next few years.
Tagging
Related nvidia News
NVIDIA Fermi GPU and Architecture Analysis
NVIDIA's Parallel Nsight finally released
NVIDIA GeForce GTX 460 - GF104 breaks cover
PhysX87, ancient tragedy in 5 acts by RWT
So long, Chris, and thanks for all the fish
NVIDIA GF100 graphics architecture details
NVIDIA Fermi: new GPU architecture, starting with GF100
NVIDIA release OpenCL GPU drivers for Linux and Windows
NVIDIA GeForce GTX 275 at $250 to fight HD 4890
