@HerraBRE @enkiv2 @awilfox Devs are expensive and hard to hire. Users will regularly buy the newer faster gear required to run the latest pointless abstractions because capitalism anyways. The web is largely optimised to scale so it works and will only be optimised if optimising is cheaper than buying more AWS instances. The financial incentives dictate everything.
Conversation
Notices
-
toby (tobypinder@mastodon.social)'s status on Saturday, 29-Apr-2017 00:43:53 UTC toby - Hallå Kitteh repeated this.
-
Ꮢ๐ϲoᴄo Ⅿoԁem Ᏼasіlisk (enkiv2@mastodon.social)'s status on Saturday, 29-Apr-2017 02:03:18 UTC Ꮢ๐ϲoᴄo Ⅿoԁem Ᏼasіlisk @tobypinder @awilfox @HerraBRE
Devs are expensive *because* they are hard to hire, *because* of poor hiring practices.
Users buy newer faster gear when they can afford to, but moore's law ended a decade ago, so newer gear won't remain faster gear for much longer, and fast turnaround was never sustainable anyhow.
The web wasn't optimized for anything, & never scaled well.
Tech industry finances are based on the whims of VCs, & thus are a fantasy.
Hallå Kitteh repeated this. -
Hallå Kitteh (clacke@social.heldscal.la)'s status on Saturday, 29-Apr-2017 02:12:11 UTC Hallå Kitteh @enkiv2 @tobypinder @herrabre @awilfox You make many good points, but new gear is still faster, because Moore's Law is still in charge. We're just applying it differently so that increasingly, new gear is faster in such a way that we can move down the scale, stay on equivalent performance, and lower the cost instead. We're not moving into 12 GHz clocks, but that's not necessarily what Moore is about. -
Hallå Kitteh (clacke@social.heldscal.la)'s status on Saturday, 29-Apr-2017 03:17:20 UTC Hallå Kitteh @enkiv2 @tobypinder @herrabre @awilfox
> Transistor count stopped doubling at the 18 month rate in 2005.
Yes, true. Eight Moore cycles later the latest "Pentium IV" would have gone from ~200 Mt to ~50 Gt and the Xeon on enwp.org/Transistor_count only has ~7 Gt. So over the last 12 years the cycle has been around 28 months. But Moore said "2 years" already in 1975. It wasn't doubling every 18 months 1995-2005 either. The 2-year cycle held up until 3-6 years ago, and now it's more like a 3-year cycle.
The median consumer CPU fell off the strict cycle a long time ago, I agree with that. I don't have an analysis of cycles per watt or dollar and whether it's exponential, but it's still growing.
I think we will naturally add less "pointless abstractions" as the curve tapers off, and go back to optimizing. It's just like Peak Oil in that sense. There won't be a sudden calamity, just a gradual reassessment of priorities. -
Hallå Kitteh (clacke@social.heldscal.la)'s status on Saturday, 29-Apr-2017 03:22:28 UTC Hallå Kitteh @tobypinder @enkiv2 @herrabre @awilfox TL;DR: I take back that Moore's Law is still in charge, but it ended in 2011-2014 rather than in 2005. Although the curve is now sub-exponential it's still closer to exponential than to any other simple formula. -
Hallå Kitteh (clacke@social.heldscal.la)'s status on Saturday, 29-Apr-2017 03:27:05 UTC Hallå Kitteh @lnxw48a1 @jeffcliff @enkiv2 Adding cores may be considered more Kurzweil than Moore, and we may argue about what "fundamentally faster" means, but I believe @tobypinder's point was that hipster devs still have breathing room to just add stuff because things still keep getting faster.
My pocket computer being faster than one of my decade-old laptops or desktops, I can't disagree with this.
I do agree that it won't go on forever. Half a century from now, the incentives will look different. -
Hallå Kitteh (clacke@social.heldscal.la)'s status on Saturday, 29-Apr-2017 15:02:33 UTC Hallå Kitteh @tobypinder We probaly disagree when I say "as it should be". I don't see the virtue of tighter programming as long as you stay within your (end customer's) desired performance budget.