Brian Krzanich, Intel's chief executive, recently delivered a keynote address at AutoMobility LA 2017 in Los Angeles to show how autopilot technology can change every aspect of the future - from cabin design to entertainment and life-saving Safety system, a detailed forecast.
As a leading auto-driving chip company, Intel also seized the opportunity to promote the EyeQ 5 chip efficiency record developed by Mobileye (now an Intel company) and compare with Nvidia's Drive PX Xavier SoC designed for autopilot .
In a speech, Krzanich mentioned that he recently completed the acquisition of Mobileye and emphasized that Intel "can provide deeper learning efficiency than its competitors (Nvidia in this case) by twice as much."
As Nvidai continues to promote its Drive PX platform and is positioning itself as a leader in AI-based autonomous driving, Intel seems to be working hard to make the most of this brewing technology The battle of specifications is higher.
Misinterpreted?
Jack Weast, Intel chief engineer and chief architect of autonomous driving solutions, said in an interview with EE Times that Intel has always been more conservative in promoting its chip performance, however, Weast said that when the talk is fiercer, "We are tired of always seeing us mistakenly quoted and interpreted."
Weast complains that competitors and the media often mistakenly compare Nvidia's Drive PX to Intel's desktop PC chips, saying that if you really want to make an evenly matched comparison with Apple-to-Apple , Mobileye's fifth-generation vision sensor fusion chip must be compared to Nvidia's Xavier SoC. The EyeQ5 delivers 24 trillion operations per second (TOPS) at 10 watts. In contrast, the DRIVE PX Despite the fact that Xavier delivers 30 TOPS, it consumes 30W, "Weast said." We are 2.4 times more energy-efficient. "
Of course, Nvidia is now rolling out its latest Pegasus SoC, which is scheduled to launch in 2018 with the AI Compute chip designed to perform 320 TOPS performance (up to 10x better performance than its predecessor) and consuming 500W , Which is said to be used for fully autonomous driving at L5 level, said Weast: "Although Pegasus is a new product, its efficiency performance has not significantly improved."
Nvidia's Pegasus is powered by two of the company's latest Xavier SoCs, featuring a new generation of discrete GPUs with hardware acceleration.
However, one of the unanswered secrets is how Intel intends to combine Mobileye's "Eye" (EyeQ vision chip) with Intel's microprocessor "brain" in a highly automated vehicle, even though some experts may not be aware.
In fact, Intel itself may and may have to take some of the blame for this misunderstanding and confusion in the market, but with the exception of the Mobileye EyeQ5, the CPU giant remains silent about what kind of autonomous driving SoC it is developing.
Autonomous driving multi-chip platform coming soon
According to Weast, Intel plans to launch a "multi-chip platform for autopilot" at the International Consumer Electronics Show (CES) in early 2018. He explained that the solution will incorporate the EyeQ 5 SoC, Intel's low-power Atom SoCs, as well as hardware including I / O and Ethernet connectivity.
Earlier this year, when Intel introduced the GO autopilot development platform, Intel described the Atom processor C3000 as a "chip that offers high performance-power ratios that can handle large amounts of computation in a low-power design."
So how does the Atom SoC share processing tasks with EyeQ 5? Weast said, "Let's allocate and partition the computational load among multiple chips based on the overall workload we need to drive the car."
As for whether or not FPGAs are also used in this multi-chip solution, Weast's case was negative. "Weast said:" Some customers may seek to meet certain applications with FPGAs such as custom I / O or security but This will not show up in the new multi-chip platform we are going to release. "
How to divide labor?
Mobileye released EyeQ 5 before it was acquired by Intel, and the Israeli company touted the new SoC as the "brain" of the autonomous vehicle at the time and was able to implement "Sensor Fusion Vision for a fully autonomous vehicle (Level 5) Central Computer "mission.
If you really like Mobileye planning, then Intel Atom SoC again in this multi-chip program where?
Weast explains that autopilot requires different levels of sensor fusion, and some sensor fusion in deep learning acceleration applications requires the chip to handle a large number of highly parallel, multithreaded code for it. "Therefore, the EyeQ 5 is ideal." Meanwhile, We need a higher level of environmental sensor convergence that looks at motion trajectories and data validation. "Weast explained:" At this point, CPUs are better suited for such missions. "
It seems to Intel that it is highly automated and "does not need to have all its tasks plugged into a single SoC like the EyeQ 5." Weast also explained that Intel has "a lot to clarify about the system's idle periods Where and how to divide workload and workload. "
Weast said that following the completion of Mobileye's acquisition in August last year, each member of the team "immediately participated in the development of a multi-chip platform plan."
Weast promised that Intel will soon release details of this multi-chip platform designed for autopilot. "At that point, we should be able to take a platform level between the Nvidia Drive PX and Intel's solutions Comparison".
Compile: Susan Hong
(Reference text: Intel Touts Auto AI Chip 's Efficiency, by Junko Yoshida) eettaiwan