Maybe this is just my own brain, but it's interesting to me how the potential for special-purpose coprocessors/cores has never decreased, but seems to be much less visible than it once was.
I remember way back in the 68k days, for "normal use" benchmarks--which in the day was largely integer-based--an '030 and and '040 might rate similarly, but for floating point operations the FPU in the '040 made it exponentially faster. Without an FPU, decoding an MP3 took literally days; with, it could be played in real time.
Later, vector processing units made certain things vastly faster on chips that had them versus chips that didn't.
These days we still benefit immensely from things like hardware video codecs, but they're common enough and invisible enough that we just don't think much about it anymore. Of course your phone has an HEVC encoder, almost everything does now, it's just sort of there doing its thing invisibly now.
Which is why the incredibly low friction between Apple's chip team and their software group could make AS Macs more interesting than just Geekbench scores. Already we are (I assume) seeing the benefits of this with some of the flashy software tricks that an iPhone camera can do because of specialized processing capabilities of A-series chips, so it wouldn't be a stretch at all to think that a group at Apple developing MacOS or Mac software might say "If we had a core that could do [thing] 100x faster than it can be done by a general-purpose core, we could implement this awesome new feature." and have that result in an actual core that does [thing] within a couple of chip generations.
It's mostly hypothetical at this point, but I have to keep reminding myself that even though Apple Silicon turns in truly impressive Geekbench scores that doesn't actually capture a lot of the real potential of a completely vertical development stack like Apple is moving toward.
I remember way back in the 68k days, for "normal use" benchmarks--which in the day was largely integer-based--an '030 and and '040 might rate similarly, but for floating point operations the FPU in the '040 made it exponentially faster. Without an FPU, decoding an MP3 took literally days; with, it could be played in real time.
Later, vector processing units made certain things vastly faster on chips that had them versus chips that didn't.
These days we still benefit immensely from things like hardware video codecs, but they're common enough and invisible enough that we just don't think much about it anymore. Of course your phone has an HEVC encoder, almost everything does now, it's just sort of there doing its thing invisibly now.
Which is why the incredibly low friction between Apple's chip team and their software group could make AS Macs more interesting than just Geekbench scores. Already we are (I assume) seeing the benefits of this with some of the flashy software tricks that an iPhone camera can do because of specialized processing capabilities of A-series chips, so it wouldn't be a stretch at all to think that a group at Apple developing MacOS or Mac software might say "If we had a core that could do [thing] 100x faster than it can be done by a general-purpose core, we could implement this awesome new feature." and have that result in an actual core that does [thing] within a couple of chip generations.
It's mostly hypothetical at this point, but I have to keep reminding myself that even though Apple Silicon turns in truly impressive Geekbench scores that doesn't actually capture a lot of the real potential of a completely vertical development stack like Apple is moving toward.