All the way back in 2011, graphics chip specialist NVIDIA (NVDA 2.77%) announced that it was putting together its very own high-performance ARM (ARMH +0.00%)-compatible CPU core known as "Project Denver." The goal with Denver, at least as outlined in a blog post by NVIDIA Chief Scientist, Bill Dally, was to power chips spanning "personal computers to servers and supercomputers."
The first chip powered by the Denver core was the 64-bit flavor of the company's Tegra K1 processor. The chip found its way into the Nexus 9 tablet, but as far as I can tell, the chip didn't see widespread adoption; the 32-bit Tegra K1 based on off-the-shelf ARM Cortex A15 cores seems to have been more popular (and it's the only version of the K1 currently listed on NVIDIA's website).
The follow-on chip to the Tegra K1, known as the Tegra X1, didn't use the Project Denver core but instead used ARM's Cortex A57 and Cortex A53 cores in a "big.LITTLE" configuration.
And, frankly, it would seem that NVIDIA is doing quite well with these ARM-based solutions in the automotive space, particularly given that the company has reported multiple consecutive quarters of impressive revenue growth in its automotive business.
At this point, I believe that it's not necessary for NVIDIA to continue to invest in developing custom CPU cores. Here's why.
Low return on investment
The main market for NVIDIA's Tegra chips seems to be automotive; the company has exited the smartphone business, and I honestly can't remember the last time a new Tegra-based tablet was announced.
Developing a (good) custom, high-performance CPU core is both difficult and expensive, meaning that a company needs to have a good reason to do so. From my perspective, a "good reason" would be that the investment in a custom core has the potential to translate into a large enough competitive edge that the incremental revenue/gross profit dollars more than offsets the required research-and-development costs.
If NVIDIA were still trying to go extremely broad with Tegra -- that is, if it were aiming to grab significant microprocessor share in the smartphone, tablet, and even PC markets -- then continuing to develop custom cores in a bid to differentiate would make sense.
However, in the automotive infotainment market, I get the impression that the true points of differentiation lie outside of the CPU core.
Points of differentiation
Although I don't wish to diminish the value of fast CPU cores for in-vehicle infotainment systems, I believe that ARM's processor core road map is strong and should deliver good performance for what NVIDIA intends to achieve with its automotive-focused processors.
Where NVIDIA seems to be investing heavily vis-a-vis automotive is in the platforms that it provides to car manufacturers in terms of both hardware and, perhaps more importantly, software. Indeed, NVIDIA CEO Jen-Hsun Huang has indicated that one way that the company can grow the margins in its Tegra/automotive business is to capture additional software content.
Additionally, for automotive, the other capabilities built into the chip such as the graphics engine as well as the image signal processor seem to be more important for many critical tasks than the CPU.
No need for a custom CPU core
For a company in NVIDIA's position, I believe that it makes more sense to license cores from ARM and make sure they are well implemented rather than try to reinvent the wheel with custom CPU cores. The money that would be spent trying to roll a custom CPU core could very well be used to bolster other portions of the company's automotive-related portfolio.
