Will compute at least keep getting cheaper, perhaps through economies of scale?
Is Kurzweil's magical next information technology, to carry on the exponential, anywhere in sight?
Old gaming consoles had cartridges (with memory); I can imagine a future in which complex software is transported in the same manner, except cartridges contain specialized asics. Or perhaps a step forward - a chip making device in every home, an equivalent of sorts of burning music to cd.
In that case, rather than a Javascript engine, wouldn't you have an ASIC for the script itself?
Wait. We have most of that already. :)
Wiki: https://en.wikipedia.org/wiki/List_of_Super_NES_enhancement_...
Custom asics would be expensive now, but perhaps... exponential... cost decrease with sufficient demand. A related approach is programmable asics. I've heard this was researched decades ago, but (I presume), silicon was so cheap, so to speak, it wasn't worth it then.
Today, with "peak silicon", perhaps all these discarded tailings will be picked over. To mix mining metaphors.
>A related approach is programmable asics. I've heard this was researched decades ago
What do you mean, if not fpgas?
[0] https://en.bitcoin.it/wiki/Mining_hardware_comparison https://en.bitcoin.it/wiki/Non-specialized_hardware_comparis...
wait, we're looking at 18,000,000 Mhash/s vs 2568 Mhash/s Like... 4 orders of magnitude?
And, 11,000 Mhash/J vs 3 Mhash/J, over 3 ordera of mag, just like you said.
yah, fpgas
I think if that scenario happens, the user won't notice because those ASICs will be deployed in "the cloud".
See also: how ridiculously cheap microcontrollers have gotten, and the current messy DRAM pricing (high-capacity chips used in phones sometimes ending up cheaper than low-performance/capacity chips).
I find it interesting that they are crossing the levels proposed for human equivalence around now depending on how you figure it - Moravec figured about 100 terraflops and you can now get "118.5 teraflops tensor-based deep learning performance" from "The Quadro GV100 is available now on nvidia.com for $8,900" https://www.hpcwire.com/2018/03/27/nvidia-brings-the-power-o... https://www.jetpress.org/volume1/moravec.htm
Just need a bit of software.
They're doing better than CPUs but they're still falling behind Moore's law.
because they aren't intended to run non-neural-net workloads. you don't judge a dishwasher by it's ability to wash your clothes, do you?
It's cool that they made some neural net things go hugely faster. But they could make the same fundamental chip on 6-year-old foundries, only slower by a factor of 2 or 3. That really impressive number has nothing to do with Moore's law.