NVIDIA's water-cooled self-driving car computer. Image source: NVIDIA.
In March 2015, graphics specialist NVIDIA (NVDA +2.59%) unveiled a development platform known as DRIVE PX, which it said would become available a few months later. DRIVE PX is a hardware and software development platform that handles all of the complex calculations to allow a car to drive itself. The initial DRIVE PX was powered by a pair of the company's Tegra X1 chips.
This year, at a pre-CES press conference, the company unveiled its next-generation autonomous driving platform, which it calls, unsurprisingly, DRIVE PX2. It's quite a huge step forward for the graphics specialist. Let's take a closer look at it.
It's far more than a pair of Tegra chips this time around
The prior-generation DRIVE PX had "only" two Tegra processors, each with 192 Maxwell graphics cores, and eight CPU cores (four Cortex A57s and four Cortex A53s). This delivered a reasonable amount of performance, but given how much performance self-driving cars will ultimately need, NVIDIA (rightly) felt the need to step it up with Drive PX2.
And so it did.
The Drive PX2 comes with two of the company's next-generation Tegra processors. Each chip features two next-generation Project Denver CPU cores (NVIDIA CEO Jen-Hsun Huang confirmed to me that these are updated CPU core designs and not just a "shrink" of the first-generation Denver cores onto the new 16-nanometer process) that first showed up in the 64-bit variant of the Tegra K1 chip. In addition, each chip comes with four Cortex A57 processors.
So, in terms of CPU performance, Drive PX2 should be substantially faster than the original Drive PX. The Tegra chips also include NVIDIA's next-generation Pascal graphics processors, which should also help to boost performance.
Where things get really interesting, though, is that the company also included two of its next-generation Pascal stand-alone graphics chips on the board. Given that NVIDIA has long promoted the capabilities of its graphics processing units in the kinds of deep learning tasks that an autonomous driving solution requires, this is quite a logical step.
What does it all add up to?
NVIDIA says that the Drive PX2 should be able to perform 24 trillion deep learning operations per second -- a 10-fold increase from the first iteration of the DRIVE PX development platform. Per the company, it delivered the original DRIVE PX to more than 50 automakers since launch. I suspect that with such a significant increase in computational power over the prior-generation model, DRIVE PX2 has the potential to see even more interest from automakers. Perhaps more interestingly, though, it seems that DRIVE PX2 should be able to command much higher average selling prices compared to the prior-generation DRIVE PX due to the inclusion of two full-fledged GPUs.
Now, in a discussion with another attendee of the press event, we noted that NVIDIA is claiming that the entire DRIVE PX2 solution consumes about 250 watts. If we assume that each of the Tegra processors is rated at around 10 watts, then this would leave a power budget of around 230 watts for the two graphics processors.
This would imply a power rating for the two stand-alone graphics processors of around 115 watts. This strongly suggests that the Pascal-based graphics chips built into DRIVE PX2 won't be NVIDIA's flagship gaming graphics chips (since such graphics processors tend to have higher power budgets). So we're probably not looking at a couple of $400-$500 graphics processors here, but it could be more along the lines of $150-$200 per graphics processor.
At any rate, with such an increase in high-value chip content, I fully expect DRIVE PX2 to be significantly more expensive than DRIVE PX, which should help boost the company's revenue (and potentially margins) in this segment. That being said, it's worth noting that this is still a development platform, so I wouldn't expect to see a massive revenue contribution from DRIVE PX2 once it becomes available in the fourth quarter of 2016.
However, NVIDIA does seem to be laying the hardware and software groundwork to be a leading vendor of self-driving car computing solutions for the years ahead by getting car vendors on board with these development platforms.
And, for the Foolish investor, that's the sort of long-term thinking that counts.
