CPU Technology is running on the same spot (1 Viewer)

[AntiHyper]

Did you remember the days when CPU speed increased by 200 MHz about every month (this was around years 2000-2003)?

How come we're not going "faster" anymore? Have we reached the limit of information processing speed?

I was curious enough to google it, may read/view these:
PcMech (The first pages are like a blast from the past, i wasn't even born when those guys came about)
Toms Hardware (This page actually shows a chart and surprisingly cpu speeds have DROPPED from 2004)

 

[sunny]

Its not possible to just increase the core speed every release, there are limits because of the manufacturing process. So instead, you look at more innovative ways to increase performance: number of cores, cache technologies, optimisations in the pipeline, etc.

Core speed is no longer the sole factor - AMD has always done this with the way they have named their processors (ie, not by core speed), and Intel has only just done this recently by restructuring the Pentium 4 nomenclature.

 

[gordo]

they say with transistor technology, the standard cpu will never surpass 5 ghz

 

[AntiHyper]

Well the world record of clock speed was 7.1GHz although it was achieved with vast amounts of cooling, it's not impossible.

From X-bit labs.

A Japanese overclocker has managed to overclock Intel Pentium 4 670 microprocessor to 7.132GHz and even run certain benchmarks on the system that was cooled down by liquid nitrogen.

EDIT: ooh rite, he's asian
____________________________________________

I know that the processing bandwidth has been increased to 64-bit but even with those CPUs, there hasn't been any newer releases.

 

[sunny]

they say with transistor technology, the standard cpu will never surpass 5 ghz

It is more of a problem due to the fabrication process and how small the process can make the transistors, rather than the transistors themselves at that sort of speed.

 

[insert-username]

It basically got to the point where increasing the clock speed only increased performance by a fraction, and produced half again as much heat and used half again as much power. It simply wasn't efficient to keep ramping speeds up when it wasn't actually helping performance that much. Case in point: AMD's Athlon line has never been clocked as quickly as Intel's Pentium 4 line, yet the Athlon dual cores have been the performances kings for the past few years. Why? Dual core, vastly more efficient architectures, integrated memory controllers. Same deal with Intel's laptop chip, the Pentium-M. Much slower clockspeed than the Pentium 4, yet it could perform almost equal to it in most benchmarks thanks to its efficient architecture. Shorter pipelines and dual-core are the way of the future.


I_F

 

[gordo]

yeh i am saying that the type of transistors used arn't gonna get much smaller

 

[SashatheMan]

yeh i remember the cpu speeds going crazy. every two weeks in a cateloge i saw a faster processor. Now they are going into duel core and other stuff, which is good aswell.

 

[Templar]

Pipeline length is generally inversely proportional to clockspeed though. There has to be a compromise somewhere (hopefully not the 30+ of Prescott).

 

[Collin]

Well, you can't keep decreasing transistor sizes indefinitely, like people have mentioned.

Heat is also a major issue, and also related to transistor count. There's a healthy equilibrium of performance vs. heat going on, and that seemed to be at around the 90nm process. I.e, up until the 90nm process, generally as you reduced the process size, less power was needed and we got a cooler operation. However as transistors cram closer together past the 90nm threshold, the problem of leakage current (which causes electrons to jump from one transistor to another) becomes a major detrimental factor. Plus we're trying to push the limits of transistor switching times (i.e clock speed ramping).. so overall we have more transistors producing heat, with worse leakage and each of those transistors switching even faster than before.. that means alot of heat. Since the chip size doesn't seem to change much, that's even less surface area to dissipate the heat too.

History shows that 90nm was pretty painful for Intel, not so much so for AMD (SOI helped a TON with the leakage issue, and the more efficient design of the A64 helped them keep performance up with lower clock speeds.) Supposedly 65nm has a few breakthroughs that will help with the leakage issue. 45nm comes after that (figure late 07 - mid 08 release date).

 

[SashatheMan]

yeh i heard 65nm processors are coming on the market soon. i hear they produce less heat and costs cheapter for intel to make