If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. |
|
|
Thread Tools | Display Modes |
#1
|
|||
|
|||
Are we approaching a physical barrier of the CPU speed?
The electromagnetic waves in conductor have a limited speed of about
v=2*10^8 m/s. The area of the common chip is about 1 cm^2, so we can assume that electromagnetic wave has to pass about s=10^-2 m inside chip. I will not take relativistic mechanic, because electromagnetic waves have fixed speed. From t=s/v we have time required that signal comes from one side of the chip to the another side is t=0,5*10^-10 s Taking f=1/t we get that at 20 GHz signal can not even travel from one side of the chip to another one before arrival of the new clock signal. If we take average size of one motherboard, here we have a road of about 10 cm. This means that motherboard clock can not go beyond 2 GHz. Intel promissed 30 GHz CPU in 2017, with smaller transistors, but bigger number of them (as usual). Therefore it is not to be expected that overal size of the CPU chip will be much reduced. I could conclude that future CPU will work with several clocks. For example at 30 GHz will work only ALU and some registers, where L1 RAM cache will drop to 5 GHz and RAM to less than 2 GHz. This means a lot of wait states, and a very weak effect of the clock raising. |
#2
|
|||
|
|||
not really, the only problem with the 90nm process is the prescotts design.
Dothans (based on P3 design) can hit 2ghz with ease with the 90nm proccess, and only produce 24watts of heat output. Hardly the ceiling of cooling. These CPU's do more work per mhz than the P4 so infact BEAT it in some tests, and also beat the A64 in some tests too. So the future of CPU's??? MORE work per mhz, not less like the P4 design, thats where intel went wrong, and a few of us knew that from the start, but was called amd fan boyz......... fact is if you find out that sooner rather than later your going to run out of clockspeed head room, you DONT design the CPU to do less work per mhz now do you? Well thats what they did, instead of fixing the P3's problems, they designed a new chip that did very little work per mhz, and everyone thought it was great it could do 1.4-1.8ghz, shame a 1.1ghz P3 utterly thrashed it and the AMD 1.4ghz made the P4 look like the new celeron... I for one cant wait for the "old" P3 design intel chips to surface in the desktop market. then maybe AMD could drop the "rating" of there chips. -- From Adam Webb, Overlag --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.768 / Virus Database: 515 - Release Date: 22/09/2004 |
#3
|
|||
|
|||
snip
I could conclude that future CPU will work with several clocks. For example at 30 GHz will work only ALU and some registers, where L1 RAM cache will drop to 5 GHz and RAM to less than 2 GHz. This means a lot of wait states, and a very weak effect of the clock raising. Optical processors are in development (search on bbc tech news) and should provide somewhat higher speeds. Further on, we should be seeing organic / biological processors... they you really could kill your pc! hamman |
#4
|
|||
|
|||
The limit is perhaps one of rising R&D per performance gain, but that
has long been with us. The issue will be whether we can maintain the requisite rising revenue to fund it. Intel along with many IT companies are as much low single-digit growth vehicles, not the old 30-40%/yr. Yes the P-M architecture is good - based on the P3s quiet development for industrial & other applications re low thermal power, good CPU power. Ramping clock to outperform the disadvantages of longer pipelines was the story of the P4 - ie, achieving less per clock but ramping clock faster. The future seems to be a mix of also achieving more per clock a la P3. It was easy in the past to get a 10x increase in performance by the benchmarks on an upgrade - so at least perceptible in real world :-) That is getting more difficult: o CPUs have very high bandwidth, as does memory o HDs have much quicker too - but are still electromechanical ---- density has greatly increased, rotational latency less so The real problem with HDs based-I/O is in the realised throughput: o As years have gone by we moved away from few-app file & 1 data-file o Today we have MS-IE & such like with vast numbers of tiny files o Lots of seeking & multiple accesses stress the mechanical speed issue The mechanical speed issue is still a major barrier - not just CPU speed. Ability to re-order commands (TCQ/NCQ) will help, but as yet whilst the market has HD drives offering it - cards & O/S up the chain do not yet. -- Dorothy Bradbury www.stores.ebay.co.uk/panaflofan for quiet Panaflo fans & other items www.dorothybradbury.co.uk (free delivery) |
#5
|
|||
|
|||
"Samir Ribic" wrote in message m... The electromagnetic waves in conductor have a limited speed of about v=2*10^8 m/s. The area of the common chip is about 1 cm^2, so we can assume that electromagnetic wave has to pass about s=10^-2 m inside chip. I will not take relativistic mechanic, because electromagnetic waves have fixed speed. From t=s/v we have time required that signal comes from one side of the chip to the another side is t=0,5*10^-10 s Taking f=1/t we get that at 20 GHz signal can not even travel from one side of the chip to another one before arrival of the new clock signal. If we take average size of one motherboard, here we have a road of about 10 cm. This means that motherboard clock can not go beyond 2 GHz. Intel promissed 30 GHz CPU in 2017, with smaller transistors, but bigger number of them (as usual). Therefore it is not to be expected that overal size of the CPU chip will be much reduced. Again an assumption that has two potential flaws. It ignores the possibility that getting smaller wont go through a quantum jump beyond adding more transistors. Then there's the possibility that I've been reading about on some scifi sites about extra-dimensional processing...well in the mean time maybe we'll just settle for the 3rd. I could conclude that future CPU will work with several clocks. For example at 30 GHz will work only ALU and some registers, where L1 RAM cache will drop to 5 GHz and RAM to less than 2 GHz. This means a lot of wait states, and a very weak effect of the clock raising. All that assumes that the entire chip in synchronously clocked. Obviously an dubious assumption. Currently the clear physical barrier seems to be KW per acre and that translates directly into horsepower per peck in the future. Maybe the watercooled OCers aren't such nuts after all. |
#6
|
|||
|
|||
"Adam Webb" wrote in message ... not really, the only problem with the 90nm process is the prescotts design. Dothans (based on P3 design) can hit 2ghz with ease with the 90nm proccess, and only produce 24watts of heat output. Hardly the ceiling of cooling. These CPU's do more work per mhz than the P4 so infact BEAT it in some tests, and also beat the A64 in some tests too. So the future of CPU's??? MORE work per mhz, not less like the P4 design, thats where intel went wrong, and a few of us knew that from the start, but was called amd fan boyz......... fact is if you find out that sooner rather than later your going to run out of clockspeed head room, you DONT design the CPU to do less work per mhz now do you? Did ya come up will multi core too? |
#7
|
|||
|
|||
Did ya come up will multi core too? yup...multi core comes under "more work per mhz" in my mind ;-) -- From Adam Webb, Overlag --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.768 / Virus Database: 515 - Release Date: 22/09/2004 |
#8
|
|||
|
|||
I'd put my money on silicon carbide devices as being in real world
applications before optical. Organic and bio-processors are still only at the stage of being really great ways to part funding organisations from their money in order to fund academic research. Pete "Hamman" wrote in message ... snip I could conclude that future CPU will work with several clocks. For example at 30 GHz will work only ALU and some registers, where L1 RAM cache will drop to 5 GHz and RAM to less than 2 GHz. This means a lot of wait states, and a very weak effect of the clock raising. Optical processors are in development (search on bbc tech news) and should provide somewhat higher speeds. Further on, we should be seeing organic / biological processors... they you really could kill your pc! hamman |
#9
|
|||
|
|||
Do you really think it matters?
The 'transistors" of the most powerful computer, the human brain, switch in milliseconds. Even when clock speeds hit a wall, more transistors will give us massively parallel microprocessors that will continue to do more work in the same amount of time than their predecessors. "Samir Ribic" wrote in message m... The electromagnetic waves in conductor have a limited speed of about v=2*10^8 m/s. The area of the common chip is about 1 cm^2, so we can assume that electromagnetic wave has to pass about s=10^-2 m inside chip. I will not take relativistic mechanic, because electromagnetic waves have fixed speed. From t=s/v we have time required that signal comes from one side of the chip to the another side is t=0,5*10^-10 s Taking f=1/t we get that at 20 GHz signal can not even travel from one side of the chip to another one before arrival of the new clock signal. If we take average size of one motherboard, here we have a road of about 10 cm. This means that motherboard clock can not go beyond 2 GHz. Intel promissed 30 GHz CPU in 2017, with smaller transistors, but bigger number of them (as usual). Therefore it is not to be expected that overal size of the CPU chip will be much reduced. I could conclude that future CPU will work with several clocks. For example at 30 GHz will work only ALU and some registers, where L1 RAM cache will drop to 5 GHz and RAM to less than 2 GHz. This means a lot of wait states, and a very weak effect of the clock raising. |
#10
|
|||
|
|||
"Immuno" wrote in message ... I'd put my money on silicon carbide devices What the hell ever happened to diamond? as being in real world applications before optical. Organic and bio-processors are still only at the stage of being really great ways to part funding organisations from their money in order to fund academic research. |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
37 Gig Barrier Question | Jim | General | 4 | January 19th 05 12:34 AM |
Physical and software structures of a HD | Susma Rio Sep | General | 0 | January 21st 04 11:55 PM |
Physical sector location | Nuno Magalhaes | General | 3 | November 6th 03 02:48 AM |
Physical repair for IDE harddrive | WDsux | General | 36 | August 29th 03 04:24 AM |
Dumping Physical Memory?? | Wheat Muncher | General | 0 | July 25th 03 05:58 AM |