>linear/exponential extrapolation doesn't work after some point
Imagine my surprise.
>Why did we never get 10ghz processors?
Because you have to compete against moore's law itself. Silicon isn't that good at high freq, the gate voltage is too high, the noise is terrible and you can't switch to exotic non-silicon semiconductors without loosing decades of silicon VLSI research. Similar to the itanium fiasco.
heat density
in order to bring heat dissipation down to levels we have today, transistors need to reach switching speeds close to terrahertz, which is almost impossible with silicon
>exotic
compound semicon products (both digital and analog) are among us since the invention of cell phones
because to get to that point you need to start using light signals instead of throwing much slower electrons and voltage potentials around. nobody (publicly) has that technology so the speeds we're at now are probably close to the theoretical limit of silicon
news flash, faraday, photons are electromagnetic waves
i thought electrons already move at the speed of light
>compound semicon products (both digital and analog) are among us since the invention of cell phones
Yes, SiGe, but not for VLSI at <100 nm and the power target for +100 million transistor of the P4 era.
all they need for development is a similar level of funding every new node is receiving nowadays. then again, you don't need to scale iii-v devices down to 20nm to reach the same performance of modern cmos.2pxyy
2 weeks ago
Anonymous
>all they need
No shit! it seems so simple! like competing with Cell against x86.
The point of Itanium was to drive development off non-Intel architectures, and in that it succeded. AMD also indirectly helped them with AMD64. Itanium actually managed to perform well, but by then Opteron/Athlon 64 was already eating Intel's lunch
while the peak performance may be the same, there are (many) tasks where a 10GHz single core would out-perform a 5GHz dual core
plenty of tasks can't be (perfectly) parallelised
No shit, given equal architecture and everything else, a 10ghz cpu will always beat a 5ghz cpu in single core performance anytime because its twice as fast.
and 8 times as hot
it doesn't work like that
it's like, imagine comparing a car doing 100km/h, versus 6 cars across 6 lanes doing 100km/h. if you only have one person to transport, adding more cars won't get them to their destination any faster, all having more cars does it allow you to transport more people in the same amount of time
2 weeks ago
Anonymous
except computers don't work like that
single threaded workload or not, data has to go back and forth a lot. maybe each car can only move one person at 100km/h, but that means i can move six people 100km/h, one at a time when they are ready, meaning the next six people don't have to wait a full hour (not counting the return trip because thats not how a CPU works) to be moved 100km
so a single core CPU at 10GHz will become bottlenecked and drastically slow down as it has a much longer I/O delay, while the 6 core, 5GHz procesdor, even being half as fast, can do six times as many operations at once, drastically reducing the I/O delay. why does this matter in single-threaded workloads? because the OS is never just performing that one operation, and if the storage or memory for example have a small pause, it can be amplified as during that delay the CPU drives a different person 100KMs, and when it's ready to continue the original operation it might not be on time, causing a knock-on slowdown effect
i'd take a slower CPU with more cores (to a point) over a faster CPU with fewer cores, because there will be less delays in processing leading to an overall faster system, even for single-core workloads due to reduced I/O delay
2 weeks ago
Anonymous
for practical use with an OS involved, yes, a slower dual core has advantages over a faster single core even with a single threaded process, for the simple fact that you rarely are ever actually only running one thread when you include the os, i/o, drivers, and such. having another core means that your process can have a core entirely to itself, with os tasks not interrupting it
2 weeks ago
Anonymous
all valid points but there's no point in discussing the necessity of multiple threads. that's already settled long ago, they are a must. point is, you'd take 5ghz 6 cores over 4ghz 8 cores all day and that is still a big issue.
>linear/exponential extrapolation doesn't work after some point
Imagine my surprise.
>Why did we never get 10ghz processors?
Because you have to compete against moore's law itself. Silicon isn't that good at high freq, the gate voltage is too high, the noise is terrible and you can't switch to exotic non-silicon semiconductors without loosing decades of silicon VLSI research. Similar to the itanium fiasco.
https://i.imgur.com/Tj7ZlIP.png
Why did we never get 10ghz processors?
10ghz is possible with dual core, just not with quad core (its a cooling issue)
Intel Tejas specifically failed because it had the "heat density of a small star". There was just no way to cool it even in theory.
Shortly after it failed and Intel limped along for a while with garbage like the Pentium D someone there got their shit together and glued a couple Pentium M mobile CPUs together which turned out to be a way better solution.
>and doesn't need cooling to -200
physically not possible, heat is qyite literally speed, the faster something moves the hotter it is to anything relatively slower. you cannot have the transistors move that fast and not get that hot, doesn't matter what you make it out of. that's simply not how physics works.
because to get to that point you need to start using light signals instead of throwing much slower electrons and voltage potentials around. nobody (publicly) has that technology so the speeds we're at now are probably close to the theoretical limit of silicon
heat density
in order to bring heat dissipation down to levels we have today, transistors need to reach switching speeds close to terrahertz, which is almost impossible with silicon
>exotic
compound semicon products (both digital and analog) are among us since the invention of cell phones
[...]
news flash, faraday, photons are electromagnetic waves
[...]
they do, he doesn't know what he's talking about
>photons are electromagnetic waves
Correct. Electrons are not. Guess which one carries electric currents
em waves are composed of electric and magnetic fields, moron. voltage waves are propagated by electric field hence its unit, volts per meter while current waves are propagated by magnetic waves hence its unit, ampere per meter. by the way you don't carry current, you carry charge.
Ackchyually both photons and electrons are wave-like point-like particles with no volume. But photons are massless whereas electros have a tiny bit of mass.
Electric Currents are carried by EM fields.
The EM fields do, in fact, propagate at the speed of light.
because to get to that point you need to start using light signals instead of throwing much slower electrons and voltage potentials around. nobody (publicly) has that technology so the speeds we're at now are probably close to the theoretical limit of silicon
No, it doesn't really change much. Signals propagate through copper at like 0.6c.
Fiber does get much closer to the speed of light (still loses a bit in total internal reflection).
But no, going from 0.6c to 0.999c isn't going to matter much.
The bigger benefit of fiber optics is bandwidth and interference - which is why it is preferred in networking.
>i thought electrons already move at the speed of light
No. Electric fields on the other hand do. When you "push" an electron into a conductor, a different electron "falls" out the other end. But that transmission is what happens at nearly the speed of light.
it's more like pulling on a string, that is, when pull on one side, the other side almost immediately gets pulled as well, you needn't pull the length of the string to affect the other side
same thing with electricity in wires, put put an electron on one side, and the "pressure wave" that causes moves at near light speed, but the electrons themselves move quite slowly
Its not a transistor problem. Individual transistors can operate up to 200 GHz.
The problem is with energy consumption, heat, and architectural limitations. IPC is more important than overall speed nowadays.
Personally I think the biggest issue is latency.
Sure, you could get your CPU running at ridiculous speeds, but it still has to work with external RAM. No matter how fast the CPU runs, unless you fix the latency issues it's only going to be wasting even more clock cycles waiting for data to travel to and from memory. It's why shit like speculative execution is so popular, it lets the CPU chew on the data it has while it waits for the RAM to feed it more.
More cores were more important and research/development had to be focused on those for competitive reasons
You can make your processors better at a relatively constant speed, but you can't give them high core counts and high clocks.
A 12ghz dual core would be useless, however.
>Why did we never get 10ghz processors?
this pajeet made a career out of it and now a bong with a fake phd is managing the site.
anandtech was always a joke and always will be a joke.
>linear/exponential extrapolation doesn't work after some point
Imagine my surprise.
>Why did we never get 10ghz processors?
Because you have to compete against moore's law itself. Silicon isn't that good at high freq, the gate voltage is too high, the noise is terrible and you can't switch to exotic non-silicon semiconductors without loosing decades of silicon VLSI research. Similar to the itanium fiasco.
heat density
in order to bring heat dissipation down to levels we have today, transistors need to reach switching speeds close to terrahertz, which is almost impossible with silicon
>exotic
compound semicon products (both digital and analog) are among us since the invention of cell phones
news flash, faraday, photons are electromagnetic waves
they do, he doesn't know what he's talking about
>compound semicon products (both digital and analog) are among us since the invention of cell phones
Yes, SiGe, but not for VLSI at <100 nm and the power target for +100 million transistor of the P4 era.
all they need for development is a similar level of funding every new node is receiving nowadays. then again, you don't need to scale iii-v devices down to 20nm to reach the same performance of modern cmos.2pxyy
>all they need
No shit! it seems so simple! like competing with Cell against x86.
sus
What is the titanium fiasco?
https://duckduckgo.com/?t=lm&q=itanium+fiasco&ia=web
tl;dr failed project of intel for their "future arch" (to kill the competency)
The point of Itanium was to drive development off non-Intel architectures, and in that it succeded. AMD also indirectly helped them with AMD64. Itanium actually managed to perform well, but by then Opteron/Athlon 64 was already eating Intel's lunch
I thought "itanium fiasco" was something that anon literally made up as they were writing their comment, just so you know.
>never get 10ghz processors?
we did, dual core 5 GHz effectively are
while the peak performance may be the same, there are (many) tasks where a 10GHz single core would out-perform a 5GHz dual core
plenty of tasks can't be (perfectly) parallelised
No shit, given equal architecture and everything else, a 10ghz cpu will always beat a 5ghz cpu in single core performance anytime because its twice as fast.
and 8 times as hot
Meh. Stop using C++ and most of the CPU bound stuff is just graphics and neural nets.
ok but what about my 5ghz processor with 6 cores, even with losses that's still gotta be like at least twice as good right
it doesn't work like that
it's like, imagine comparing a car doing 100km/h, versus 6 cars across 6 lanes doing 100km/h. if you only have one person to transport, adding more cars won't get them to their destination any faster, all having more cars does it allow you to transport more people in the same amount of time
except computers don't work like that
single threaded workload or not, data has to go back and forth a lot. maybe each car can only move one person at 100km/h, but that means i can move six people 100km/h, one at a time when they are ready, meaning the next six people don't have to wait a full hour (not counting the return trip because thats not how a CPU works) to be moved 100km
so a single core CPU at 10GHz will become bottlenecked and drastically slow down as it has a much longer I/O delay, while the 6 core, 5GHz procesdor, even being half as fast, can do six times as many operations at once, drastically reducing the I/O delay. why does this matter in single-threaded workloads? because the OS is never just performing that one operation, and if the storage or memory for example have a small pause, it can be amplified as during that delay the CPU drives a different person 100KMs, and when it's ready to continue the original operation it might not be on time, causing a knock-on slowdown effect
i'd take a slower CPU with more cores (to a point) over a faster CPU with fewer cores, because there will be less delays in processing leading to an overall faster system, even for single-core workloads due to reduced I/O delay
for practical use with an OS involved, yes, a slower dual core has advantages over a faster single core even with a single threaded process, for the simple fact that you rarely are ever actually only running one thread when you include the os, i/o, drivers, and such. having another core means that your process can have a core entirely to itself, with os tasks not interrupting it
all valid points but there's no point in discussing the necessity of multiple threads. that's already settled long ago, they are a must. point is, you'd take 5ghz 6 cores over 4ghz 8 cores all day and that is still a big issue.
10ghz is possible with dual core, just not with quad core (its a cooling issue)
I wish they made one
Intel Tejas specifically failed because it had the "heat density of a small star". There was just no way to cool it even in theory.
Shortly after it failed and Intel limped along for a while with garbage like the Pentium D someone there got their shit together and glued a couple Pentium M mobile CPUs together which turned out to be a way better solution.
We're almost there
Oh this is actually legitimate. It's not that meaningful but it almost makes the dream of OP's headline come true, just 20 years late lol
Tell me when we manage it without it crashing after a second and doesn't need cooling to -200
>and doesn't need cooling to -200
physically not possible, heat is qyite literally speed, the faster something moves the hotter it is to anything relatively slower. you cannot have the transistors move that fast and not get that hot, doesn't matter what you make it out of. that's simply not how physics works.
Supposedly, there's a store that sells 10 GHz (or 15) laptops. One might have to go there in person and ask. It's in a mall in Indonesia.
>One might have to go there in person
>in Indonesia
no thanks anon, I'm not getting malaria just to get scammed by a shitskin
the resonant frequency of silicon won't let us get that high
because to get to that point you need to start using light signals instead of throwing much slower electrons and voltage potentials around. nobody (publicly) has that technology so the speeds we're at now are probably close to the theoretical limit of silicon
i thought electrons already move at the speed of light
Electrons have mass.
They do not
>photons are electromagnetic waves
Correct. Electrons are not. Guess which one carries electric currents
em waves are composed of electric and magnetic fields, moron. voltage waves are propagated by electric field hence its unit, volts per meter while current waves are propagated by magnetic waves hence its unit, ampere per meter. by the way you don't carry current, you carry charge.
Ackchyually both photons and electrons are wave-like point-like particles with no volume. But photons are massless whereas electros have a tiny bit of mass.
moron
Electric Currents are carried by EM fields.
The EM fields do, in fact, propagate at the speed of light.
No, it doesn't really change much. Signals propagate through copper at like 0.6c.
Fiber does get much closer to the speed of light (still loses a bit in total internal reflection).
But no, going from 0.6c to 0.999c isn't going to matter much.
The bigger benefit of fiber optics is bandwidth and interference - which is why it is preferred in networking.
>i thought electrons already move at the speed of light
No. Electric fields on the other hand do. When you "push" an electron into a conductor, a different electron "falls" out the other end. But that transmission is what happens at nearly the speed of light.
it's more like pulling on a string, that is, when pull on one side, the other side almost immediately gets pulled as well, you needn't pull the length of the string to affect the other side
same thing with electricity in wires, put put an electron on one side, and the "pressure wave" that causes moves at near light speed, but the electrons themselves move quite slowly
no software actually needs more than 3Ghz single core (that's 3 BILLION cycles a second). It's all devs being moronic and bloatmaxxing
tell that to my home server that virtualizes a bunch of different systems. fricking moron.
>virtualization
read
>It's all devs being moronic and bloatmaxxing
This is really the answer.
You're talking to a moronic first year student, at best.
>your eyes can't see more than 30 fps, mooron
but we kinda did? speculative execution and running multiple instructions per cycle, the cpu is doing way more than it used to
Its not a transistor problem. Individual transistors can operate up to 200 GHz.
The problem is with energy consumption, heat, and architectural limitations. IPC is more important than overall speed nowadays.
it is a transistor problem. clock signals are nowhere near sine waves. their harmonic content spectrum spreads well into three digit ghz.
>Anand Lal Shimpi
Back to the poo with you loo
Anand founded tech journalism you dumb underage homosexual
Funny that you mention it, the Intel chief engineer at the time was a jeet, and still is.
We do. We use them to cook food
>Running Microsoft Word can only take so much processing power, regardless of how complex your documents may be
>Microsoft: Hold my beer
but we did
>4 cores
>2.8 Ghz
>4 x 2.8 = 11.2 Ghz
Core speed is not multiplied. You cannot add it. What are you doing?!
already did, in 2000s
Because you refuse to hook up a fridge sized air conditioner to your machine. Blame yourself.
because you were moronic enough to listen to a pajeet blogger
Personally I think the biggest issue is latency.
Sure, you could get your CPU running at ridiculous speeds, but it still has to work with external RAM. No matter how fast the CPU runs, unless you fix the latency issues it's only going to be wasting even more clock cycles waiting for data to travel to and from memory. It's why shit like speculative execution is so popular, it lets the CPU chew on the data it has while it waits for the RAM to feed it more.
>10Ghz by 2005 running at < 1 volt
Lol even
Was this a typo?
no. 0.07 micrometer is 70nm
Wow. So far off.
of what? yonah in 2006 was 65nm
More cores were more important and research/development had to be focused on those for competitive reasons
You can make your processors better at a relatively constant speed, but you can't give them high core counts and high clocks.
A 12ghz dual core would be useless, however.
>Why did we never get 10ghz processors?
this pajeet made a career out of it and now a bong with a fake phd is managing the site.
anandtech was always a joke and always will be a joke.
Because they would be frickhuge and use 1MW of power. Multicore+multithreading is optimal.
There were never going to be processors that fast. It was a marketing stunt because AMD was lower frequency but faster.
but we hav 60ghz wifi