10ghz promised

W

Wesley1357

Limp Gawd
Joined
Jul 4, 2005
Messages
374
cpu companies once said that we would one day get to 10ghz. now we have stopped at 3.8ghz. what is the technical reason why we can't get more clock speed? :confused: :)
 
Willseyvilleny said:
think of it this way, my cpu is 5 ghx amd x2 2.5x2
Click to expand...
no. its not.

its two 2.5ghz cpu's. nothing more, nothing less.

heat.. heat.. and oh yeah, heat
Click to expand...
yup. until they can fix transistor leakage, we wont hit insane clockspeeds with any conventinal CPU design. netburst has the potential, but current technologies in transistors are holding it back.
 
Didn't we just have this same exact discussion a couple days ago?
 
Wesley1357 said:
hey lithium726. "transistor leakage" is that heat or?
Click to expand...
That's electrical current jumping outside of a predetermined pathway. It creates heat and- more importantly- electrical interference.
 
i was jsut stating in total, i know its not true, but if you think it is also. I also agree with the thermal problems, also there has been a big push away from mhz=power had amd not been so sucess ful we would have 8.4 ghz p4's that did very little work, all hype
 
is that the reason having to do with all heat related problems, because of "electrical current jumping outside of a predetermined pathway", is this the explanation?
 
Wesley1357 said:
is that the reason having to do with all heat related problems, because of "electrical current jumping outside of a predetermined pathway", is this the explanation?
Click to expand...
yep.
 
Wesley1357 said:
cpu companies once said that we would one day get to 10ghz. now we have stopped at 3.8ghz. what is the technical reason why we can't get more clock speed? :confused: :)
Click to expand...

That's not exactly a simple question. :)

As mentioned the leakage current from transistors contributes to the heat problem. (Electrical energy becomes heat energy as the electrons move from ground to v+). "Leakage" current (which encompases a number of different things) is extra heat that's created without contributing to the operation of the CPU.
In particular faster low voltage threshold (that intel made extensive use of in prescott) transistors have massive leakage current (about 10 times that of more power efficient, slower, transistors).

Wire delays - global wires in particular that don't tend to get physically smaller from process to process - are becoming more troublesome in scaling (P4 already had 2 stages dedicated to allowing signals to propogate along long wires).

Traditionally large jumps in clock speeds have come as a result of longer pipelines, P4, 686, and to a lesser extent 586. After the issues encountered with P4, Intel is moving back to a much more modest pipeline length.
 
Wesley1357 said:
if making it smaller results in more heat! why not making the cpu bigger?
Click to expand...
they consume more power as the transistors are further apart, which ramps up the heat too when you get to high clock speeds - transistors that are closer together can run faster than those that are further apart.
 
lithium726 said:
they consume more power as the transistors are further apart, which ramps up the heat too when you get to high clock speeds - transistors that are closer together can run faster than those that are further apart.
Click to expand...
Also because having the transistors further apart allows more distance over which the traces can interfere with one another. Longer signal equals higher voltage, higher latencies, and higher noise or interference.
 
Asian Dub Foundation said:
like 2 7800GTXs SLI dosent make 48 pipes
Click to expand...

So what does it make? 34.6 Pipelines?.... The Pipelines are there.... 24x2 = 48.... :p Might not make 100% 48pipelines in the bechmarks... but they are still there :)
 
Look at your hair - a single strand - that is about 10 microns wide - the current usable internal gold wiring is .13micron - I used to work for a chip manufacturing company called Saw Tech (the have a web site) problem is simple they can actually make them smaller, the heat can be 'handled', but the 'crossover' of information is an issue - with 300+ pipelines of electrical travel within a chip the crossover is already an issue, at .07 micron this would give the ability to have over 600+ pipelines with no control over 'crossover' ... once they have a way to control this 'crossover' the next generation of processor will be ready for market.

This way Both AMD and Intel have gone with the Dual Core - which is simply 'stacking' of current technology...
 
Asian Dub Foundation said:
like 2 7800GTXs SLI dosent make 48 pipes
Click to expand...

Well, two of them won't ACT like there are 48 pipes there, but, math says 24 + 24 = 48, so there are indeed 48 of them... ;)
 
Wesley1357 said:
if making it smaller results in more heat! why not making the cpu bigger?
Click to expand...

Making them smaller isn't the problem.
If you were to make transistors very very large again (say .35 microns), you'd still have massive heat issues because large transistors have large capacitances, so you need a lot more electrical engergy to drive a state change.
And if you try to make large transistors anywhere near the same speed of very small ones you'd end up with massive leakage current again.

Smaller transistors need less power and are faster. That's a fundemental rule that's driven chip design over the last 30 years.

The empahsis right now is on slower more power efficient transistors. Using (comparatively) high voltage thresholds to reduce subthreshold leakage. And physically thicker insulators without sinifigantly increasing gate capaciatnce to reduce electrons tunneling through the insulating layer around the gate / channel.
 
I'd bet they have pallets of 10 ghz's sitting around somewhere along with 9's and 8's and so on. The wonderful world of marketing however is going to make sure they get every single cent from us for every new 100mhz increment until they get to 10ghz. ;)
 
Puterguru said:
I'd bet they have pallets of 10 ghz's sitting around somewhere along with 9's and 8's and so on. The wonderful world of marketing however is going to make sure they get every single cent from us for every new 100mhz increment until they get to 10ghz. ;)
Click to expand...
i'd bet not.

the WR for a prescott is a little over 7ghz on some insane triple (or more) cascade cooling system - and you think intel has 10ghz chips just hangin around that can run on air cooling? i think not.
 
lithium726 said:
i'd bet not.

the WR for a prescott is a little over 7ghz on some insane triple (or more) cascade cooling system - and you think intel has 10ghz chips just hangin around that can run on air cooling? i think not.
Click to expand...
I was being a little sarcastic. My point is that if they could mass produce a 10ghz tomorrow that was affordable for everyone we would not see it for years. The big wigs that make all the money will milk us for every mhz/ghz increment they can get until we hit 10ghz.
 
Heat is the most pertinant issue. Rumor has it, once the process of creating synthetic diamonds is perfected and costs less, processors will mostly be made of it, because it does not conduct the electricity, making heat a non-issue. It's been a long time since I've had that explained, so if I screwed something up, bear with me.
 
Puterguru said:
I was being a little sarcastic. My point is that if they could mass produce a 10ghz tomorrow that was affordable for everyone we would not see it for years. The big wigs that make all the money will milk us for every mhz/ghz increment they can get until we hit 10ghz.
Click to expand...

If a company did manage to put out a 10GHz chip they wouldn't hold on to it, they would just price it insanely high knowing companies would still buy it. Some software licenses are per cpu (Oracle is an example), so it's far cheaper for them to buy fewer more expensive cpus then have many more cheaper cpus.
 
Don't judge chips by ghz, instead by MFLOPS.

Reason? - Team Japan (and a few others) have hit over 7ghz with prescott and have realised that it is only slightly better than a dothan at 3.5ghz. The reason for this is because the CPU processes more in a single clock (someone help me out on the word for it, tired atm, begins with S). The future isn't how many ghz you can clock, its how many MFLOPs you can process.
 
Zxcs said:
Don't judge chips by ghz, instead by MFLOPS.

Reason? - Team Japan (and a few others) have hit over 7ghz with prescott and have realised that it is only slightly better than a dothan at 3.5ghz. The reason for this is because the CPU processes more in a single clock (someone help me out on the word for it, tired atm, begins with S). The future isn't how many ghz you can clock, its how many MFLOPs you can process.
Click to expand...

Looking for.. IPC?

(Instructions per clock-cycle)
 
You must log in or register to reply here.
Back
Top