Discussion Main / CranialProcessingUnit

I'm pretty sure that in order for it to shave anything close to a microsecond the mecha would have to be truly immense.

Korbi, the mecha would have to be wired with fiber optics or something of the sort for what you said to be true.

You should look up reaction times of humans. Pay attention to how long it takes for individual signals to be sent throughout the body. Sure, a human nervous system isn't going to be like what might be used for a robot's electronics, but it's still using electric pulses.

On modern CP Us, we have different levels of cache (L1, L2, etc). The lower the number, the smaller the cache, closer to the CPU, and faster the access time. The reason for this is that being even marginally farther away reduces access speeds, and being larger necessitates portions of the memory being further away physically and thus take more time for electrical impulses to travel. At the 3 to 4 gigahertz (0.33 to 0.25 nanoseconds per clock cycle) timescales we're dealing with, these micrometers are significant and mean that retrieving from L2 cache and beyond takes more than a clock cycle simply due to limits in speed of light in silicon.

Worse are issues with parallel transmission. Differences in the length of the path, even those introduced by **bending the wire** can cause issues with synchronicity of data on different lines in the same wire bundle (one big reason we moved from Parallel ATA (IDE) to Serial ATA (SATA) hard drives was to alleviate this issue).

As an Electronics and Software Engineer (designs computers), I see many engineering reasons to keep at least essential processing and cache close to the sensors. However, like modern computers, it would also be wise to keep long term storage someplace further away to be paged into the cache as needed.