A long time ago, NVIDIA (NVDA +0.08%) CEO Jen-Hsun Huang seemed to be very interested in having Intel (INTC +0.64%) build its Tegra chips (and indeed, chips for all of the mobile companies). While Intel would most likely avoid such business as it would threaten its Core and Atom processors, NVIDIA's discrete graphics business would certainly be a fantastic candidate for Intel's Custom Foundry.
GPUs: A near-perfect fit
If you look at NVIDIA's highest-end GK110 part (known colloquially as "big Kepler"), you'll notice that it is a 7.1 billion transistor beast with a die size that weighs in at 550 square millimeters. These chips also sell for hundreds of dollars and with better performance, the company can command a premium. With Intel's transistor performance lead, its density lead, and its reputation for having some of the best manufacturing yields (that is, how many chips built actually work on a given wafer) in the industry, this could be a beautiful partnership.
From a strategic perspective, if NVIDIA were able to build these graphics chips at Intel, this would completely tip the scales further in what has already been a pretty lopsided duopoly in NVIDIA's favor. NVIDIA already has the marketing/brand down, extremely good hardware designs, and some of the world's best graphics software teams in-house. Augment that with a manufacturing advantage, and it's very difficult to see how rival Advanced Micro Devices (AMD +1.15%) could compete on performance, power, or -- frankly -- cost.
But what's the downside?
While Intel does not today build discrete graphics cards (and probably has no interest in doing so), it is competing head-to-head with NVIDIA's GPUs (branded Tesla) in the high-performance computing space with its own Xeon Phi. By giving NVIDIA access to its latest and greatest manufacturing technology, Intel neutralizes what is probably its biggest competitive edge against Tesla. However, if we step back for a moment, we realize the following:
- NVIDIA's Tesla business is on the order of a few hundred million dollars per year and capturing it entirely via Xeon Phi would be more of a philosophical victory rather than a financial one.
- Capturing the entirety of NVIDIA's high-end discrete business would more or less deprive TSMC (TSM 0.80%) -- Intel's real long-term enemy -- of all of those leading-edge wafer starts, which is a much more critical strategic objective than winning fairly niche high-performance computing accelerator card sockets.
The next potential point of contention would be NVIDIA's Tegra business. It would be awfully weird for NVIDIA to have to build its Tegra processors over at TSMC while its big discrete GPUs get preferential treatment. The problem here is that if Intel were to do a Rockchip-like deal with NVIDIA, the latter company has enough graphics IP (among others) chops to really deal some damage to Intel's higher-end Atom line and even parts of the Core line with an X86-based Tegra.
Foolish bottom line
While these thoughts are, admittedly, highly speculative, they are also not impossible. Intel's recent willingness to open up its factories and given how well suited big, performance-sensitive graphics processors are to Intel's manufacturing strengths make this an interesting consideration, but as is always the case in business, the devil's in the details.
