[subwayclub]

It's the race for core counts that we've been expecting since the mid 2000s - just a lot later than anyone expected since we detoured into GPU and mobile computing.

Our software still isn't really ready for it, but I guess that's to be expected.

[ahartmetz]

In addition to software not being ready:

It detoured into a company having no need to improve its products too much because they were already the best. If the technology is somewhat stagnating, it is best not to release all the improvements at once. After core counts have increased, what else will still make people buy new CPUs? Maybe I lack some imagination here - and so do companies making phone CPUs, which can't think of anything better than upping the core count.

Also, high core count CPUs are slightly more similar to GPUs, and Intel seems to be trying to keep the GPU threat at bay. There is no reason to believe that Intel couldn't design a competitive high end GPU, but that is not where its entrenched advantage is. (This part is not my own analysis, I read it somewhere else and consider it highly plausible.)

[jorvi]

I wonder if we'll also move to a console-like architecture where there's only 1 pool of memory (in the case of the PS4 GDDR5) which is shared by both parts of the APU.. it would be quite nice because it allows the system to repurpose memory to whichever task is critical at that moment, you'd never have 'wasted' memory sitting around doing nothing.

[naasking]

Probably still won't happen because of the software limitations. Like the GPU detour, a different ASIC or FPGA detour will probably happen instead.

[godzillabrennus]

The tail will wag the dog with this one.