You are not going to play video games on mission computers. Its an embedded device that runs single pre-determined application and not a general purpose computer.
Aircraft's mission computers in current configuration don't need of AI processing power.
AI is a soft computing field while aircraft's mission computers are based on hard computing. Any induced error due to soft computing may cause crash. We have not yet received such an matured state in AI, that it (AI) can control planes.
Why you want to put 2 snapdragons on a card and stack 2-4 such cards with a common block of RAM? Their is no application need for it. The entire mission computer application is optimized for use on e5500 core. Its not always about hardware, look at apple even though their hardware dwarfs in comparison to Android counterparts but as their software is optimized the give awesome experience compared to Android. Just look at their iPhones or MacBook series.
Do you realize how complex the architecture would be for the configuration you are suggesting not to mention how fragile the system will be.
Why you want a video card on mission computers? You are not going to play games and nor going to do insane number crunching.
Why you want DL on aircraft, the field is not matured enough to put in aircraft. More importantly what good will come from having DL rig on aircraft.
Just because one thinks AI, ML and DL are fancy hot field that doesn't mean we use them without proper appropriations of advantages and disadvantages.
I am not talking about the mission computer. Read my post again carefully. A mission computer is almost always a rules based system.
A pocket DL rig lets you process and extract real time useful intelligence from images and video lightning fast. Image and video tracks from your DAS, irst, targeting pods etc can be processed for information on the aircraft itself and then the resulting data shared with the necessary receivers. Such a device can be used in applications for target identification, interpreting SAR data realtime, etc.
The other application of such a compute block is to do heavy number crunching for digital signal processing for your EW systems, jammers, etc.
"Why you want to put 2 snapdragons on a card and stack 2-4 such cards with a common block of RAM?" ==> RISC processor with supreme number crunching ability and low power and thermal footprint. + commercially available en masse and cheap. The stacking and resource sharing architecture is decided as per our requirements.
"look at apple even though their hardware dwarfs in comparison to Android counterparts but as their software is optimized the give awesome experience compared to Android. Just look at their iPhones or MacBook series." ==> Apple CPUs have been kicking Qualcomm ass for a while now, and they just butt banged Intel with their M1 series. Also Android uses the JVM to run, which itself is a memory hogger. Which is why resource footprint of JVM based language apps are higher than those on C/C++.
"Do you realize how complex the architecture would be for the configuration you are suggesting not to mention how fragile the system will be." ==> These are common card designs. You need to design the stacking and connection architecture mostly. The system's fragility physically is determined by how it has been constructed, fitted, and cooled. Its software fragility depends on the quality of the software and support provided for the same. And upgradeability and integration with our systems as needed.
"Why you want a video card on mission computers? You are not going to play games and nor going to do insane number crunching." ==> Not on mission computers. I mentioned this earlier.
"Why you want DL on aircraft, the field is not matured enough to put in aircraft. More importantly what good will come from having DL rig on aircraft." ==> We are approaching a state where targeting pods will be able to zoom in more than ever before. FLIR imagery will become even more clear. Passive sensors will become even sharper. DL will certainly be needed to interpret this data for the pilot. Else he will have an information overload. Even if you ignore the DL part, upcoming comms will become even more secure and convoluted in its implementation. So the hardware must be good enough to crunch numbers fast.