“Moore’s law is the observation that, over the history of computing hardware, the number of transistors on integrated circuits doubles approximately every two years”
from wikipedia
in 1971 the intel 4004 processor was released with a cpu die area of 12mm square and 2300 transistors
lets assume for our calculation that the die area remains the same size just the transistor count doubles every 2 years ? due to a halving of fab process once every four years.
2013 – 1971 = 42 years double very 2 would mean 21 doublings of transistors since the intel 4004 in 1971
this would mean
2300 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2 * 2
= 4,823,449,600 transistors or 4.8 billion in the same cpu die area 12mm square as the original 4004 ? actually i think I got it wrong 21 doublings is double this figure so 4.8 billion is being generous.
which is 13 times smaller than the CPU die of the average quad core today ?
meaning if you were to use the same cpu die area of an average quad core today 160mm squared
the number of transistors should be in the region of 64,151,879,680 transistors in todays processors by moores laws prediction ?
which means were missing about 62,751,879,680 transistors from moores law projections ?
a current quad core 4670k has 1,400,000,000 transistors ie 1.4 billion transistors sadly in a cpu die area of 160mm square,
if we were using 300mm wafers to produce 4004 processors using the original 10um transistor scale you could get 5282 cpus per wafer at 10um, if you were using the 22nm scale of todays process you could get 228,869,565 ie 222.8 million 4004 processors from one 300mm wafer, thats if you could theoretically cut them out with zero width cutting of the wafer ;)
literal scientifically accurate moores law has not occurred.
This can be explained easily because fab nm process halving changes I presume must have not occurred every 4 years, were missing about 5 and a bit doublings which is about 98% off from the number of transistors, easily done with square area transistor reduction doubling.
looking through this page I calculate the number of doublings achieved by halving of fab process scale that has actually occured
it took intel from 1971 till 1976 to third the fab process from 10um to 3um in the 8085 cpu
then 1976 to 1982 to get to 1.5um 286
then from 1982 to 1993 to halve the fab process and get to 0.8um as was used in the pentium etc
this was practically 11 years for not even a full a halving of the fab process size ? and thereby not even a full doubling of the transistor count. if fab transistor scale was halved every four years we would now be at 5nm to 2.5nm features in lithography, moores law is a marketing tool used by intel, not a guarantee-able law. Nor could it ever be, since its dependant on scientific breakthroughs that are not a given, Intel have done an amazing job, but moores law is not a scientific law.
as you can see the halving of process nm scale is not occuring every 4 years which would quadruple the transistor count with the square area, leading to moores law and the doubling of transistor counts every 2 years, if once every four years fab processes did half this would stand, but since were still behind on that projection we can see that moores law does not stand true, because of fab process not halving every four years, the other element to take into account is wafer size increase and cpu die size increase which goes some way to explaining why transistor count has not been much worse in relation to moores law.
Intel have deliberately dumbed down their new “package specifications” on their website, so as to no longer highlight die size or transistor count compared to spec listings for older processors
all so as to avoid the concept theyre not keeping pace with moores law nor have matched it over the years.
mobile parts is what theyre bent towards now in haswell ? well whose left making parts for desktop then !? not every maker can focus on mobile, dont follow amd, the cpu transistor count has not significantly gone up, i had my i7 920 at 4.2ghz, i5 2500k could easily get 4.8ghz, haswell slightly worse due to the IHS issues unless its delidded core temp variation is too high, so the desktop CPU is getting worse every year, this is not moores law, this is not progress, haswell is a pathetic increase in power over Ivy, no excuses, quad core cpus have stagnated, your argument doesnt stack for the desktop user.
I pay for a new CPU and motherboard when there is a significant transistor and speed increase on offer, theyre seeing desktop cpu sales dropping because we dont have significant enough power increases to justify it, and theyre bending they’re whole platform toward mobile to compete with AMD and more power saving states and giving us smaller CPU die size chips from the wafer, with integrated graphics we didnt ask for, at the same speed and price as last years chip, there is no point in paying for this ?
8 cores next chip or I’ll wait for an affordable 8 core amd jaguar setup once the consoles have had their fill, hyper threading is weaksauce compared to higher clock frequency. and how about a little big core scenario so It can be my always on low power home server until I decide to crank it up to full bore mode.
there are now phones with more cores than your average i5 desktop chip, samsungs octacore arm 4 big 4 little seems intriguing and innovative at least in comparison.
