Why two tegra CPUs were used in DRIVE PX 2?

What is the difference between Tegra A and Tegra B?
I think they are two seperated CPUs connected by ethernet, so what is the design purpose?
And the peak performance 8 TFLOPS seems be the sum of the two CPUs with iGPU and dGPU working together, so how can we use the two CPUs together to reach the 8 TFOPS performance?

Dear liuhb16,

What is the difference between Tegra A and Tegra B?
I think they are two seperated CPUs connected by ethernet, so what is the design purpose?
-> It is for redundancy.

Thank you for your response, what about the last question? Is the peak performance 8 TFLOPS the sum of the two CPUs with iGPU and dGPU working together?

Dear liuhb16,
you can offload some of the work to CPU and enable vectorization in the code to extract peak performance from ARM cores.

Thanks
Take Tegra A as example, I can access Tegra B with ssh command, but how can I assign work to Tegra B in the codes? I cannot find any documentation about that.

Dear liuhb16,
Tegra-A and Tegra-B are inter-connected with Ethernet over the Marvell Gb AVB Switch. They can communicate with each another through this network connection.
You can run two different algorithms on two Tegra simultaneously or you can use socket programming for communication between two Tegra processes.

Dear SivaRamaKrishna,
I appreciate it for your response, thank you!

In continuation with the above question, If it is Drive PX2 (AutoChauffeur) provided Tegra A and Tegra B to use it as a dual redundant system. Example, A DriveNet application executes on-board Drive PX2 both Tegra A and Tegra B at the same time. If DriveNet application fails/hangs on Tegra A then it automatically taken over the same DriveNet application on Tegra B. If it possible this way? could you please provide links or available sample with SDK DriveWorks.

Dear n.kemsaram,
You can run two different DNN algorithms on each Tegra. The DriveNet sample uses only one Tegra. It does not get scheduled on both Tegras when you run it. If you launch DriveNet sample on both Tegra at a time, by default the automatic taking over does not happen, you need to take care of this in your application by checking the outputs from both Tegra in every interval

Hi,

Can we install difference DriveOS on TegraA and TegraB?