2018 promises to be an exciting year for tech, and could be the year that automotive tech takes a big step forward.
Buried in the CES avalanche of sketchy new hardware and flashy televisions are cars. In years past, automakers were keen to show their own self-developed technology for automation or networked vehicles. The issue was always the same: how would my Ford talk to your Nissan if everyone was building their own tech stack? It hardly seemed safe, or smart.
This year shows auto makers may have found their hero in NVIDIA. With increasing emphasis on automobiles over the past few years, CES seems custom-made for NVIDIA, which has quickly become a smart-car platform after branding itself as a gaming-hardware company.
NVIDIA’s focus is the DRIVE IX platform (yes, the company loves branding in all-caps). The idea is to create a co-pilot for drivers that can help them stay safe and engaged. It will be able to alert drivers to sudden conditional changes, including other cars in your blind spots and potential dangers around the next curve in the road. It also tracks head movement and gaze to know if you’re bored or sleepy, and can hold a conversation via a bit of natural language and artificial intelligence.
On the hardware side, NVIDIA’s DRIVE PX Xavier boards power everything. Its seven billion transistors can deliver 30 TOPS (trillion operations per second) while consuming 30 watts of power. Its A.I. counterpart, the DRIVE PX Pegasus, swings for the fences with 320 TOPS. NVIDIA also has two boards for autopilot and smarter cruise control.
NVIDIA’s stack is attempting to corner the smart car market. What seemed a questionable bet a few years ago could pay off as automakers grow more interested in autonomous software and “smart” dashboards.
Where Does This Leave Auto Developers?
At square one, in many ways.
NVIDIA’s developer efforts are centered on CUDA, its parallel platform and API model. In a nutshell, CUDA offers access to the CPU and GPU with little lag or confusion, which is great when you’re trying to cycle through trillions of operations every second.
CUDA was designed to work with C, C++ and Fortran. There’s also F#, which is available for CUDA’s GPU modeling. If you’re wondering whether or not R would serve as a plausible replacement for Fortran, that’s a tough call. R is a lot more concise, but perhaps not as snappy as Fortran, which can also port to C++ (more natural for CUDA). Indeed, the NVIDIA CUDA documentation expresses C/C++ support.
That’s about as far as NVIDIA has taken it, mostly because that’s the end of its responsibility. It can solve for ‘x’ (in this case, data modeling), but any end-user interfacing will have to come via the auto manufacturer’s platform. We’ve yet to see what VW may offer on that front, but GM’s SDK may serve as a precursor for what all car manufacturers will offer to developers.
GM’s Dev Client lets developers monitor driver performance, supply directional assistance, and access a ‘smart grid’ for charging electric vehicles. There’s good reason to question the utility of these services if the auto manufacturer maintains its own stack, especially since many users already have mapping apps on their phone (Google Maps, Waze and Apple Maps on mine!). One could also see a bespoke (and very niche) ‘get me to a Starbucks!’ app for your next car.
We should also keep in mind that, while NVIDIA is capturing the hearts and minds of various auto manufacturers and rideshare providers, it’s not the only game in town. GM, Ford and others have their own tech silos, while Google and Apple are busy fleshing out proprietary in-car interfaces. We’ve also yet to see how these platforms and services may interact, underscoring how early we are in the smart-car process.