Losses are proportional to the square of current, so having the current over a long transmission line not have to handle current peaks allows some combination of better efficiency and lower transmission line cost.
As an example, compare transmission of 1 ampere versus the transmission of (1 + sin(t)) amperes, through 1 ohm. The average dissipation of 1 ampere is 1 watt; the average dissipation of (1 + sin(t)) amperes is 1.5 watt.
And this is why learning integral calculus is a good thing.
Full static CMOS is almost never used in fast microprocessors. The signal path is NMOS, possibly with PMOS devices used to precharge dynamic nodes and as a part of latches.
Irrelevant. The attempted and achieved government-driven changes to healthcare have the purpose of increasing government power and the effect of ruining lives. The government has been forcing changes in education for 50 years ("New Math") and the results have been disastrous. These have nothing to do with defending a status quo and everything to do with preventing malicious loonies from collapsing civilization.
Intel is hardly even trying to get heat off the die. An aftermarket has developed that de-lids Intel CPUs and replaces Intel's crap thermal interface compound with better stuff. For example, https://siliconlottery.com/products/delid
Consumer software that doesn't run fast enough because processors aren't fast enough, doesn't get into the hands of many consumers and may not even be developed (because software writers know that it won't be fast enough.) The claim that consumers don't need more speed is based on not knowing what more speed could provide.
Here's an example of an application that would require hundreds of times more speed than what is currently available: a program that could create a new piece of fiction and render it in real time, in 4k video. Don't think that couldn't be a best-seller, an individualized story creator?
That's nonsense. One critical factor is lithography. Optics, resists, etching processes, more precise steppers, etc. all need to be developed and in production before ICs get to a new process node. None of that requires state-of-the-art processors to design.
Things like Cadence's design software can run adequately on hardware several generations back. The slowness of old hardware is a damned nuisance, but it's not prohibitive.
huge breakthroughs do not happen in mature technologies.
That's in the range of question begging to tautological. If there's a huge breakthrough in what is thought to be a mature technology, then obviously it wasn't a mature technology.
Conceivably, you could save the machine's status in flash each time there's been some sort of meaningful change, and reload that at turnon. You wouldn't have checked hardware for malfunctions, you wouldn't be connected to the internet, and your hard disks wouldn't be ready. Nonetheless, if someone designed a system properly, that system could at least pretend to be ready for use in a second or so.
How long did your C64's monitor take to produce a display? Unless the filament in the CRT was kept warm all the time, it wasn't ready in 0.1 seconds.
Advances in semiconductor manufacturing technology are extremely difficult. If you, as a semiconductor manufacturer at the 50 nm node, decided to make 5 nm your next process node, here's what would happen: you'd go broke before you ever developed the new tech, as your competition reached the 25 nm node and you couldn't sell your uncompetitive products any more.
Even assuming you somehow managed to remain profitable, you'd probably get your 5 nm process to market only 18 months ahead of your competition. It's that difficult to overcome the many obstacle in the way of getting to 5 nm production.
That wikipedia article is based on assumptions that are almost irrelevant to the construction of computers made of realistic components, like, say, atoms. The article is based on things like black holes and should not be given serious consideration here.
If I were IBM and created a processor 4 times faster than Intel's at the same manufacturing cost as Intel's, I'd be delighted to drive Intel out of the market.
So computers have hit the wall, where computing performance in the consumer market only serves to play games, the only element of the consumer market that needs the power
Try waiting 2 minutes for your computer to sharpen a 4000 x 5000 image, only to find that you've chosen the wrong parameters and have to try again... and again... and again... and fianally discover that the edges of your photo need more sharpening than the center. There's a whole world of image enhancement out there that isn't even tried in the consumer photo market because it takes too long (blind deconvolution on a single 4000 x 5000 image can take several hours, again, only to find that the parameters are wrong.) A 1000 X speedup would be very useful, and the typical consumer isn't even aware of the potential for improvements.
With modern small geometries, wire capacitance can dominate gate capacitance, particularly for long connections such as might exist between ALU and cache.
Exception: stages whose output is boolean or a small number of possibilities. The next stage can be run speculatively.
A takes an hour to run, the result is TRUE or FALSE.
Two instances of B are started, one with an input of TRUE, the other with an input of FALSE. In each case, B takes an hour to run.
An hour after the program was started, knowing the output of A allows us to choose the correct output of B.
the vast majority of the increases in computer power have been architectural
"More transistors" makes architectural improvements possible. On-chip caches started being used when there were enough transistors available to make on-chip caches. SSE, MMX, etc. were introduced when there were enough transistors available. Same for floating point units, on-chip memory controllers and multiple CPUs per die.
20 years ago was when dot matrix impact printers were mostly gone from the market because print quality was too poor, and color ink-jet printers had become affordable.
Slashdot Mirror
Whoosh!
I think you should learn some German history.
Losses are proportional to the square of current, so having the current over a long transmission line not have to handle current peaks allows some combination of better efficiency and lower transmission line cost.
As an example, compare transmission of 1 ampere versus the transmission of (1 + sin(t)) amperes, through 1 ohm. The average dissipation of 1 ampere is 1 watt; the average dissipation of (1 + sin(t)) amperes is 1.5 watt.
And this is why learning integral calculus is a good thing.
TFA says they're assuming an efficiency of 80% to 85% for a charge-discharge cycle. That sounds too optimistic to me, but I'm only guessing.
Full static CMOS is almost never used in fast microprocessors. The signal path is NMOS, possibly with PMOS devices used to precharge dynamic nodes and as a part of latches.
For the limited set of applications for which it was useful, the 8X300 was at least 8 times faster than its competition, circa 1980.
Irrelevant. The attempted and achieved government-driven changes to healthcare have the purpose of increasing government power and the effect of ruining lives. The government has been forcing changes in education for 50 years ("New Math") and the results have been disastrous. These have nothing to do with defending a status quo and everything to do with preventing malicious loonies from collapsing civilization.
Intel is hardly even trying to get heat off the die. An aftermarket has developed that de-lids Intel CPUs and replaces Intel's crap thermal interface compound with better stuff. For example, https://siliconlottery.com/products/delid
Consumer software that doesn't run fast enough because processors aren't fast enough, doesn't get into the hands of many consumers and may not even be developed (because software writers know that it won't be fast enough.) The claim that consumers don't need more speed is based on not knowing what more speed could provide.
Here's an example of an application that would require hundreds of times more speed than what is currently available: a program that could create a new piece of fiction and render it in real time, in 4k video. Don't think that couldn't be a best-seller, an individualized story creator?
https://www.researchgate.net/figure/238594798_fig1_Fig-1-The-number-of-transistors-per-microprocessor-chip-versus and many similar graphs show that Moore's observation of transistor count has been maintained at least through 2011.
That's nonsense. One critical factor is lithography. Optics, resists, etching processes, more precise steppers, etc. all need to be developed and in production before ICs get to a new process node. None of that requires state-of-the-art processors to design.
Things like Cadence's design software can run adequately on hardware several generations back. The slowness of old hardware is a damned nuisance, but it's not prohibitive.
That's in the range of question begging to tautological. If there's a huge breakthrough in what is thought to be a mature technology, then obviously it wasn't a mature technology.
As an interesting side note, Moore's observation could remain true for many more years by just making chips larger.
Conceivably, you could save the machine's status in flash each time there's been some sort of meaningful change, and reload that at turnon. You wouldn't have checked hardware for malfunctions, you wouldn't be connected to the internet, and your hard disks wouldn't be ready. Nonetheless, if someone designed a system properly, that system could at least pretend to be ready for use in a second or so.
How long did your C64's monitor take to produce a display? Unless the filament in the CRT was kept warm all the time, it wasn't ready in 0.1 seconds.
Advances in semiconductor manufacturing technology are extremely difficult. If you, as a semiconductor manufacturer at the 50 nm node, decided to make 5 nm your next process node, here's what would happen: you'd go broke before you ever developed the new tech, as your competition reached the 25 nm node and you couldn't sell your uncompetitive products any more.
Even assuming you somehow managed to remain profitable, you'd probably get your 5 nm process to market only 18 months ahead of your competition. It's that difficult to overcome the many obstacle in the way of getting to 5 nm production.
That wikipedia article is based on assumptions that are almost irrelevant to the construction of computers made of realistic components, like, say, atoms. The article is based on things like black holes and should not be given serious consideration here.
Silicon replaced germanium fairly quickly.
If I were IBM and created a processor 4 times faster than Intel's at the same manufacturing cost as Intel's, I'd be delighted to drive Intel out of the market.
Try waiting 2 minutes for your computer to sharpen a 4000 x 5000 image, only to find that you've chosen the wrong parameters and have to try again ... and again ... and again ... and fianally discover that the edges of your photo need more sharpening than the center. There's a whole world of image enhancement out there that isn't even tried in the consumer photo market because it takes too long (blind deconvolution on a single 4000 x 5000 image can take several hours, again, only to find that the parameters are wrong.) A 1000 X speedup would be very useful, and the typical consumer isn't even aware of the potential for improvements.
With modern small geometries, wire capacitance can dominate gate capacitance, particularly for long connections such as might exist between ALU and cache.
Exception: stages whose output is boolean or a small number of possibilities. The next stage can be run speculatively.
A takes an hour to run, the result is TRUE or FALSE.
Two instances of B are started, one with an input of TRUE, the other with an input of FALSE. In each case, B takes an hour to run.
An hour after the program was started, knowing the output of A allows us to choose the correct output of B.
"More transistors" makes architectural improvements possible. On-chip caches started being used when there were enough transistors available to make on-chip caches. SSE, MMX, etc. were introduced when there were enough transistors available. Same for floating point units, on-chip memory controllers and multiple CPUs per die.
20 years ago was when dot matrix impact printers were mostly gone from the market because print quality was too poor, and color ink-jet printers had become affordable.
A battery charger that doesn't try to keep lithium batteries fully charged, thereby killing them.
Ask Thomas Crapper.