Although CES may have started as the Consumer Electronics Show, the global event has expanded over the years to include everything from business technologies to cars. As a result, the show was not only a regular key event for things like CPU and GPU announcements, but also became home to big announcements about automotive technology. And for Intel’s autonomous driving subsidiary, Mobileye, the two are merging directly this year as the group announces its new generation of Level 4 autonomous driving SoC, EyeQ Ultra, this morning.
Intended for release in 2025, the EyeQ Ultra is Mobiley’s most ambitious SoC to date, and not just for performance. In fact, as a site that admittedly rarely covers car-related announcements, it’s relative lack performances that make today’s announcement so interesting. But here we go ahead; so let’s start with the basics.
EyeQ Ultra is Mobileye’s seventh-generation car SoC and is designed to enable Level 4 autonomous driving – otherwise known as “high automation” level driving. While not exactly the holy grail of Level 5 (full automation), the L4 is a more immediate target for car companies working on self-driving cars, as it is a degree of automation that allows cars to start and, if necessary, safely stop themselves. In practice, Level 4 systems are likely to be the basis of robo-taxis and other fixed-location vehicles, where such self-driving cars will only have to operate over a known and well-defined area under limited weather conditions.
Mobileye already has L4 automation hardware today, however that hardware consists of six to eight EyeQ chips working together. For the R&D phase, that’s more than enough – after all, just making everything work is a pretty big deal – but as the L4 is now within Mobileye’s reach, the company is working on the next step in technology production: it’s cheap enough to build and compact for mass market vehicles. And here comes the EyeQ Ultra.
At a high level, the EyeQ Ultra is intended to be Mobileye’s first autonomous driving system in the mass market. To achieve this, Mobileye is designing a single-chip L4 drive system – that is, all the necessary processing hardware is contained in one high-end SoC. So, when connected to the right cameras and sensors, the EyeQ Ultra – and the Ultra itself – could drive the car to L4 standards.
But perhaps the most interesting thing about the EyeQ Ultra is that Mobileye intends to drive the L4 with a chip that isn’t particularly powerful on paper. The official performance figure for the chip is 176 TOPS, which is certainly a lot of performance today (2022). But that’s just a fraction of the performance planned for high-end SoCs in the 2025 timeframe that Mobileye is targeting. In short, Mobileye not only believes they can make the L4, but that they can do so with a fraction of the performance and energy consumption of their competitors.
Finally, the argument with which Mobileye will come to market is that autonomous driving has matured enough as a field that not everything has to be done in software or in very flexible accelerators. Instead, it’s time to start building real ASICs, with highly specialized fixed-function components (or otherwise limited flexibility) that do one thing, and do it well. All in all, it’s a natural way forward for most task-specific processors, and Mobileye believes self-driving car systems are finally ready to move in that direction.
By diving into speeds and feeds, the EyeQ Ultra will have several different types of cores, each for different tasks involved in driving a self-driving L4 vehicle. These are:
- 12 RISC-V CPU colors
- Arm GPU
- DSP arm
- SIMD colors
- VLIW colors
- Coarse-grained reconfigurable array core (CGRA).
- The core of deep learning
The last 4 groups of cores are Mobileye’s “proprietary accelerators” and most of the work is done on them. According to Mobileye, there are 64 such accelerator cores, although the company does not analyze how many cores are in each particular group. The entire chip, in turn, will be built on a 5nm process (probably TSMC).
Finally, Mobiley’s investment in so much limited flexibility hardware means that the chipmaker has relatively little hardware available for high-end general computing – and, as their thinking suggests, relatively little need for it. With sufficient gas flow, as much as 176 TOPS should be enough for an L4 autonomous vehicle.
In Mobiley’s eyes, focusing so much on task-specific hardware will bring several benefits. The first and most important thing is that it reduces the total amount of silicon required. This not only allows them to switch everything to one chip, but also reduces the cost of connected systems. Another advantage is power: the fewer transistors you have to turn on, the less energy is consumed. And although it is relatively easy to get energy in the car – especially EV – more chips can be added. So Mobileye wants to make it the main distinguishing feature with the EyeQ Ultra: if the lower price doesn’t appeal to carmakers, hopefully lower power and cooling requirements.
Meanwhile, although Mobileye does not name any specific competitors, overall their press release / sale looks like a thinly disguised opportunity at NVIDIA, whose Atlan SoC is set to appear in the same 2025 time period. Indeed, even the timing of today’s announcement (8 a.m. Pacific time) is set relative to the start of NVIDIA’s CES 2022 introductory remarks, instead of timing it to go with Intel’s main presentation at 10 p.m. Mobileye seems to have a very clear idea of who their main rival is.
This is a significant rivalry not only because of the companies involved (essentially Intel versus NVIDIA), but also because of NVIDIA’s promotion of their automotive SOCs focused on TOPS. For reference, NVIDIA is promoting Atlan which offers over 1000 TOPS bandwidth for SoC alone, and that would be even higher bandwidth if used in a high-end multi-chip Drive PX setup, as NVIDIA likes to do.
Consequently, Mobileye is very careful to end up in a war of specifications with NVIDIA, as the latter’s high throughput deep learning certainly looks attractive compared to Mobileye’s approach based on a lower bandwidth accelerator. Basically, the difference in TOPS bandwidth only reflects the difference in design philosophy between the two groups – NVIDIA’s software-defined approach to more task-specific Mobileye accelerators – but the company is rightly concerned about how large the large TOPS number will look.
But perhaps more important – albeit in the long run – is which approach will work better. Mobileye’s approach essentially locks in their current technology attack and related self-management algorithms, while NVIDIA leaves the door open to more flexible approaches. But then what is the benefit of flexibility if it will cost an arm and a leg, especially at a time when the industry as a whole is trying to cut costs to introduce self-driving technology in more vehicles?
To round things off, it is this contrast that makes Mobileye’s approach with EyeQ Ultra so interesting. Although NVIDIA uses an equivalent approach to brute force, there is little reason to doubt that it will work. On the other hand, Mobiley’s greater investment in hardware for specific tasks carries more risk, but if they can deliver on what they promise, then they would do so with a fraction of silicon and a fraction of the energy consumed, all of which would be a big advantage over NVIDIA.
For better or worse, there are still early days in the autonomous vehicle industry. So far there has been one big series of experiments, and even when 2025 turns around and EyeQ Ultra comes out the door, it will still be just the beginning of a much bigger shift. So there is still plenty of time for companies to fight for a position in the car market – but the real commercialization of the mass market is not far off. So for Mobileye and other car SoC manufacturers, it’s time to start playing for good.
Friendly communicator. Music maven. Explorer. Pop culture trailblazer. Social media practitioner.