I've always wondered about this, theres got to be a limit to clock speed right? The processor obviously uses electricity to transfer data, and electricity travels slightly slower than light?? So, wouldnt there be a Ghz rating for the speed of light? Basically my question is what is the theoretical limit to the speed of a processor?
-
AMD deprioritizing flagship gaming GPUs: Jack Huynh talks new strategy against Nvidia in gaming market.
Is there a limit to CPU clock speed?
- Thread starter Krb686
- Start date
- Status
jitpublisher
Splendid
- May 16, 2006
- 8,747
- 2
- 31,960
Using the current technology, they are already hitting the ceiling.
Google is your friend to learn about how CPU's work and why.
Google is your friend to learn about how CPU's work and why.
Murissokah
Distinguished
- Aug 12, 2007
- 1,467
- 105
- 20,040
I see what you mean, but that is WAY theoretical. Before considering the speed of the electron, there are numerous other factors, such as the heat dissipation caused by their movement. Also, Hertz means 1/sec (or per sec), as in cycles per second. The number of cycles per seconds depend on what you call a cycle. There is a big difference between going from 0V to 5V then back (common logical cycle) and going from 0 to 1.3V (common CPU clock cycle).
As for the limitation, maybe the most important factor, is the switching time of the transistor. When you are working on very high frequencies (GHz), parasite capacitance will affect your system's performance (every electronic component stores a little bit of electric energy in form of electric field, acting like a capacitor). That stored energy needs to be discharged for the transistor to go back to its ideal LOW state, and that takes a little bit of time. If you set it too fast, it won't be able to go back at all, causing the system to fail.
Also, the electric rails on a CPU are tiny (i mean nanometricaly tiny), and cannot transfer a huge amount of charge at once. Too much current would cause them damage.
Intel and AMD engineers have to consider all this and a whole lot more when building a system, setting an ideal environment for the unit to work in (electrical parameters, heat condition, frequency), just so we can screw everything later.
As for the limitation, maybe the most important factor, is the switching time of the transistor. When you are working on very high frequencies (GHz), parasite capacitance will affect your system's performance (every electronic component stores a little bit of electric energy in form of electric field, acting like a capacitor). That stored energy needs to be discharged for the transistor to go back to its ideal LOW state, and that takes a little bit of time. If you set it too fast, it won't be able to go back at all, causing the system to fail.
Also, the electric rails on a CPU are tiny (i mean nanometricaly tiny), and cannot transfer a huge amount of charge at once. Too much current would cause them damage.
Intel and AMD engineers have to consider all this and a whole lot more when building a system, setting an ideal environment for the unit to work in (electrical parameters, heat condition, frequency), just so we can screw everything later.
aznguy0028
Distinguished
- Dec 14, 2007
- 887
- 0
- 18,990
i say we all wait for quantum computer technology and celebrate a beer to it the day it comes out...all is well x)
MU_Engineer
Splendid
- Feb 18, 2006
- 3,868
- 4
- 23,165
Theoretically, your upper limit would be due to the propagation delay in sending electrons from one point to another, which would be the speed of electricity. Electricity travels close to the speed of light and light travels a foot in a femtosecond (10E-15 s) so the top speed is somewhere in the 10E22-10E23 Hz range. However, this limit will never be reached as this assumes no capacitance and no resistance in the wire. Semiconductors have a significant amount of resistance and capacitance in their circuitry. The capacitance in transistors (such as the hundreds of millions in your average CPU) is a direct function of the transistor size (actually, volume, but you get my point.) Murissokah explains why capacitance delays switching pretty well. Capacitance used to be the limiting factor for the shipping clock speed on CPUs up to about the 180 nm generation. Now it is thermal dissipation as we could get the transistors to switch faster than they are rated for at the present but we have to deal with a ton of heat in the process as he also explained.
I think the question you REALLY want answered is how fast do you think CPUs will actually get before they are replaced with something else. I highly doubt that silicon-based CPUs will exceed 10 GHz as transistor gate length cannot be shrunk too much more, lest the gates get thin enough that electrons can leak from the emitter to the collector without any voltage at the base. In fact, I doubt that CPUs will be clocked much above 5-6 GHz before we switch to another substrate or use a different technology. However, a 5-6 GHz CPU from 10-15 years from now will be significantly more powerful than a 5-6 GHz CPU of today (one that somebody uses liquid nitrogen to do a few-minute "suicide run" overclock stunt with.)
Read carefully, he said that current reacts faster than light, not that it travels faster than light. With light you have a source and light travels from this source to destination (through fiber optics for example) with pretty constant and defined speed. With current it's a bit different - a wire always has some electrons on board and to create current you need to supply a potential difference: electron on one side and empty holes on other. This causes efect that electrons moving into a a wire are pushing electrons ahead of them in form of blast wave and holes on the other are sucking out electrons from wire way before "original" electron that entered a wire have reached end of it.
Another interesting thing about cpu's is actual RF problematic of ac current.
- While you switch with higher frequency current starts to flow on the outside of conductor, which limits your conductance (there been some theories about using nano tubes to overcome this issue, but there is no technology at the moment that would allow feasibly produce billion of tracks inside of cpu, Actually cooling your cpu with liquid hydrogen cancels out this effect as with lower temperatures electrons start to flow more inside of conductor)
- While you switch with higher frequency you can expect more and more cross talk - that is also contributing to extra heat production which is caused by fact that you need to drive your other lines harder to eliminate this phenomenon.
- If you switch at higher frequency it becomes harder to "sync" cpu. In great simplicity: imagine that your switch signal propagates at some speed - it needs to travel X tracking from point A -> B. Other signal is travelling at same speed but need to go from C -> D which is far shorter. Both signal will arrive at different times. Here is the question: will you time between "clock cycles" be higher than arrival time difference ? This is a reason why cpu's are organized in segments with own clock domains which sometimes can run asynchronously. Signals between this blocks can get synchronized by some trick but only by introducing delay which will delay early signals more than late signals. This is all doable but higher you get with frequency your design becomes more comlex. At the moment both intel and amd are doing fine job with that so you can even do a nitrogen suicide runs but ~10 years ago you coul only imagine about that.
More power related problems are:
- with higher temperature your switching characteristic will usually go tits up.
- with higher temperature your resistance of tracking will increase resulting in more power required.
- with more temperature variation you will end up with dice shrinking and expanding that is often hard to compensate. Your tracking will have exatly same back lash.
Anyway why you would like more Hz ? I've spent 3 years in enviroment dealing with 30GHz + at making equipment was a true art. Literally moving equipment in room could affect your performance - a true RF feng shui. I prefer to have 10 more cores than x2 frequency.
And finally if you disagree don't flame, check first and honour that somebody might be wrong and provide some evidence to that.
Another interesting thing about cpu's is actual RF problematic of ac current.
- While you switch with higher frequency current starts to flow on the outside of conductor, which limits your conductance (there been some theories about using nano tubes to overcome this issue, but there is no technology at the moment that would allow feasibly produce billion of tracks inside of cpu, Actually cooling your cpu with liquid hydrogen cancels out this effect as with lower temperatures electrons start to flow more inside of conductor)
- While you switch with higher frequency you can expect more and more cross talk - that is also contributing to extra heat production which is caused by fact that you need to drive your other lines harder to eliminate this phenomenon.
- If you switch at higher frequency it becomes harder to "sync" cpu. In great simplicity: imagine that your switch signal propagates at some speed - it needs to travel X tracking from point A -> B. Other signal is travelling at same speed but need to go from C -> D which is far shorter. Both signal will arrive at different times. Here is the question: will you time between "clock cycles" be higher than arrival time difference ? This is a reason why cpu's are organized in segments with own clock domains which sometimes can run asynchronously. Signals between this blocks can get synchronized by some trick but only by introducing delay which will delay early signals more than late signals. This is all doable but higher you get with frequency your design becomes more comlex. At the moment both intel and amd are doing fine job with that so you can even do a nitrogen suicide runs but ~10 years ago you coul only imagine about that.
More power related problems are:
- with higher temperature your switching characteristic will usually go tits up.
- with higher temperature your resistance of tracking will increase resulting in more power required.
- with more temperature variation you will end up with dice shrinking and expanding that is often hard to compensate. Your tracking will have exatly same back lash.
Anyway why you would like more Hz ? I've spent 3 years in enviroment dealing with 30GHz + at making equipment was a true art. Literally moving equipment in room could affect your performance - a true RF feng shui. I prefer to have 10 more cores than x2 frequency.
And finally if you disagree don't flame, check first and honour that somebody might be wrong and provide some evidence to that.
Speed is something but as you are already seeing it,
Speed is not everything, the number of cores is the base for exceeding limitations to equal or exceed speed limitations of a single core.
Ans I fully agree with MU_Engineer... he gets my vote on this one, and Jaguar gets my vote for his philosophy.
Speed is not everything, the number of cores is the base for exceeding limitations to equal or exceed speed limitations of a single core.
Ans I fully agree with MU_Engineer... he gets my vote on this one, and Jaguar gets my vote for his philosophy.
Humm I forgot to mention....
I watched a documentary on Science a few days ago and an engineer is creating a computer based on quantum mechanics by stimulating "I think if I remember correctly" atom or electrons "I cannot remember correctly" to produce 1 and 0 through some type of resonance
I watched a documentary on Science a few days ago and an engineer is creating a computer based on quantum mechanics by stimulating "I think if I remember correctly" atom or electrons "I cannot remember correctly" to produce 1 and 0 through some type of resonance
- Status
TRENDING THREADS
-
-
Question NVME M.2 drives - outside of capacity are they all the same ?
- Started by eco_bach
- Replies: 5
-
-
Facebook
Twitter
Instagram