It should have come as no surprise that Nvidia held a well-attended press event prior to the Consumer Electronics Show last night. It is, after all, one of the leading suppliers for GPUs for many consumer devices.
However, what may have come as a surprise to some is that only 10 minutes of the 90 minute presentation was given over to consumer-oriented details, with the announcement (to distinctly non-automotive whoops, cheers and applause) of the company’s Tegra X1 Mobile “Superchip” based on its latest Maxwell Architecture.
The remaining 80 minutes were used for detailed presentations of how the device can be used in next-generation cockpit HMI and advanced driver assist/automated driving systems. This included a guest-slot presentation from Audi’s Ricky Hudi, Executive VP of electronics development.
Two new products based on the X1 were announced:
- The Drive CX Digital Cockpit Computer. This features a single X1 and is capable of driving up to 16.6 million pixels worth of displays – equating to 8 full HD panels.
- The Drive PC Auto Pilot Computer, featuring dual X1 processors. This offers 2.3 Teraflops of computing power and can handle 12 camera inputs. Along with this hardware, Nvidia also revealed its Deep Neural Network Computer Vision software that is claimed to allow rapid training of the hardware to recognise a host of features, such as roadsigns, pedestrians and vehicles, and even segment these vehicles into SUVs, sedans, emergency vehicles etc.
Nvidia is thus positioning itself at the head of the race to offer the necessary compute platforms for highly automated driving, and its hardware-to-the-max approach differs markedly from many other vendors.
The current leader in automotive computer vision is undoubtedly Mobileye, which has gained its success (and somewhat astounding market capitalization) based upon the strength of its software algorithms. These are run on what even Mobileye itself would admit is fairly ordinary hardware. Mobileye is essentially a software company that has produced optimized hardware to run its intellectual property on, and has to date managed to command a premium price for its expertise.
Other vendors current roadmaps also offer significantly less hardware performance (and power consumption) than the ~10 Watt Nvidia’s X1 devices, with sub-watt class devices optimised for one or two tasks now becoming commonplace.
Nvidia’s advantage over traditional automotive vendors is thus that it offers performance way beyond what its currently being offered. However, the flip-side of this is that its devices operate at power levels and command price points significantly ahead of where the current automotive sweet-spot lies. The number of cars requiring 12 camera inputs thus side of 2020 will be small – but for those that do, Nvidia is pretty much the only game in town.
Nvidia has demonstrated that it is in automotive for the long haul. When it first started turning up at automotive events a lot of the feedback we received both from its competitors and potential customers was a mixture of disbelief and condescension. That is not the case now. It’s analysis of automotive future needs has been largely correct – but these needs, outside of some typically luxury OEMs also still largely in the future.
The challenge for Nvidia will be maintaining its focus, investment and engagement as the industry essentially catches up to having large-scale needs for the technology that it offers. It will also need to continue its evangelism for high-performance, centralized graphical computing into an industry which currently sees wide-scale use of more distributed, low-power and lower-performance devices. The adoption of high-speed bus interfaces such as Gbps-class Ethernet will be likely required to network and move all this camera data around the car – and is pretty much outside of Nvidia’s control.
However, the key challenge for Nvidia will be to develop an ever-deepening degree of software and algorithm support for its customers. It has started on this journey, but there is always more that can be done.
The challenge for everyone else is to ensure that their current roadmaps have the capability of producing something competitive to Nvidia’s high-end devices. The high-end needs will emerge, and these companies will need to ensure that that have something to meet these needs when they go mainstream.