Processor question.

SeanW

There used to be a general rule that processor power available on the market doubles every 24 months (later fell to 18)

Now, my computer is a dinosaur with a P4 1.7 Ghz, but reading recent computer ads, and indeed discussion on this board - indicates that processor speeds haven't really jumped up that much in the past four years, I just read on an "I'm getting a PC" thread someone was talking about an AMD 2.2 Ghz :eek: Indeed, that's all that seems to be reasonably priced for AMD, although Intel, seems to do processors in the 3Ghz range now.

Clearly I'm missing something.

What determines a processor's worth (if anything) besides pure Hz values? Or is just a case of there having been little need to improve processing power since 2002?

28064212

AMD Athlon 2.2GHz ≠ P4 2.2 GHz

djmarkus

Indeed, an amd processor can do more calculations per clock cycle than a p4,


The only true way you can compare intel vs amd is benchmarking.

Stephen

clockspeed is not the be-all and end-all it used to be.

tba

Amd have a mumber with a plus after the title

eg AMD 3800+

the 3800 is a close guess at intel clock speeds ie 3.8ghz

SouperComputer

There used to be a general rule that processor power available on the market doubles every 24 months (later fell to 18)

If you are referring to Moore's Law then im afraid that that's wrong. He stated that the complexity of an IC doubles with respect to cost.

Processing power has to do with much more than clock speed, unless of course one is comparing two identical cores. The amount of "pipelines" it has is one of the very first areas to look at. As well as error correction, code optimisation, cache memory\type speed etc etc.

Good example is the Xbox 360. Marketing goop could lead you to believe that it is a powerhouse by saying its got 3x 3.2 Ghz cores.

What they dont mention is that each core has only two "pipelines" or hardware threads. Last time I looked, AMD's Athlon had in the region of 11 and the P4 9. That said, im not taking the 360's RISC architecture into account, but you get where im coming from.

astrofool

SouperComputer wrote:

If you are referring to Moore's Law then im afraid that that's wrong. He stated that the complexity of an IC doubles with respect to cost.

Processing power has to do with much more than clock speed, unless of course one is comparing two identical cores. The amount of "pipelines" it has is one of the very first areas to look at. As well as error correction, code optimisation, cache memory\type speed etc etc.

Good example is the Xbox 360. Marketing goop could lead you to believe that it is a powerhouse by saying its got 3x 3.2 Ghz cores.

What they dont mention is that each core has only two "pipelines" or hardware threads. Last time I looked, AMD's Athlon had in the region of 11 and the P4 9. That said, im not taking the 360's RISC architecture into account, but you get where im coming from.

Thats a bit wrong, the 360 seems to use a form of Hyperthreading like function and actually handles two threads per core, in the same way p4 might have.

They saved the die space complexity by making it an in-order cpu rather than out-of-order, which basically makes it a specialist cpu not useful for general purpose computing (i.e. a PC). It seems to be about half way between a PC processor compared to the PS3's Cell, which is all about lots of small fast specialised processors.

SouperComputer

@Astrofool, I agree they are very different and as such difficult to compare, its still portrayed as a "9Ghz Uber Computer" though and that's what im getting at.

zilog_jones

Stephen wrote:

clockspeed is not the be-all and end-all it used to be.

Clock speeds were never really a good way of comparing processors. It kinda worked back in the days of earlier x86 clones because there wasn't too much of a difference betwen Intel, AMD, Cyrix and whatever else there used to be. It was only really when L2 cache (and especially on-die cache) came into place that big differences were seen.

Trying to compare, say a i386 to a 68000 by just clock speeds is almost equally pointless to comparing today's Intel x86s and AMD Athlons as they've changed so much.

Wertz

We're also reaching the ceiling of what current chip fabrication methods can offer...read something someplace recently that intel were having trouble breaking the 4ghz mark on a single core cpu, hence the move toward multicore and 64bit...

Capt'n Midnight

Wertz wrote:

We're also reaching the ceiling of what current chip fabrication methods can offer...read something someplace recently that intel were having trouble breaking the 4ghz mark on a single core cpu, hence the move toward multicore and 64bit...

Intel thought they could get to 10GHz with the P4 but hit heating problems. LNB's on old analog sky dishes will do 11GHz. In the UK there is an amatuer band at about 250GHz. So faster cores are possible for a while yet. If anyone solves the problem with gallium arsnide or other exotic of getting it to the same ease of fabrication as silicon then speeds will also jump.

I'm supprised Intel haven't gone down the road of building heat pipes into the CPU's to improve thermal characteristics close to the silicon.

Intel are making noises about multiple cores. Others have built them Ultrasparc T1 has 8 cores each of which can run up to 4 threads. So 32 threads per chip.

riptide

Also number of transistors on the chip, is something to be considered. Take Northwood vs Extreme Edidtion. So fastest northwood is 3.4 Ghz. Extreme Edidition is 3.73. But Extreme edition has circa 150 million transistors I think, Northwood has something like 55 million. So you can see the performance jump isn't just about proportion of speed... 3.4->3.73 = 9% jump. But I'm sure EE has a bigger jump in instructions per second, by virtue of more transistors.... and that is before you consider differnences in cache and pipelines. I thought Intel was hoping to get just the 5GHz out of prescotts...

riptide

SouperComputer wrote:

If you are referring to Moore's Law then im afraid that that's wrong. He stated that the complexity of an IC doubles with respect to cost.

Processing power has to do with much more than clock speed, unless of course one is comparing two identical cores. The amount of "pipelines" it has is one of the very first areas to look at. As well as error correction, code optimisation, cache memory\type speed etc etc.

Good example is the Xbox 360. Marketing goop could lead you to believe that it is a powerhouse by saying its got 3x 3.2 Ghz cores.

What they dont mention is that each core has only two "pipelines" or hardware threads. Last time I looked, AMD's Athlon had in the region of 11 and the P4 9. That said, im not taking the 360's RISC architecture into account, but you get where im coming from.

http://arstechnica.com/articles/paedia/cpu/moore.ars/1

Yes... like myself... Moore i think was very interested in the cost and this is really what he was pushing at when he gave that interview... It just so happens that the interpretation about transistors and size also hold true....

To best explain it... my late grandfather used to say... If the costs for cars changed like the way costs of chips did... a top drawer merc today would cost you 50cent!!!!!

Capt'n Midnight

riptide wrote:

If the costs for cars changed like the way costs of chips did... a top drawer merc today would cost you 50cent!!!!!

Yeah but the boot would be a tad smaller.

SouperComputer

Capt'n Midnight wrote:

Yeah but the boot would be a tad smaller.

and would need a copper base and a funny aluminum fin on top.

The boy racers would love it!

AT OP, generally, AMD athlon64 is the way to go at the moment, especially for gaming.

stevenmu

riptide wrote:

Also number of transistors on the chip, is something to be considered. Take Northwood vs Extreme Edidtion. So fastest northwood is 3.4 Ghz. Extreme Edidition is 3.73. But Extreme edition has circa 150 million transistors I think, Northwood has something like 55 million. So you can see the performance jump isn't just about proportion of speed... 3.4->3.73 = 9% jump. But I'm sure EE has a bigger jump in instructions per second, by virtue of more transistors.... and that is before you consider differnences in cache and pipelines. I thought Intel was hoping to get just the 5GHz out of prescotts...

Afaik, the increase in transistor count for the EE was mainly to do with the extra cache, not any extra processing logic.



Currently Intels, and to a lesser degree AMDs, have reached a speed limit due to heat and power requirements. To combat this Intel are currently moving to a 65nm production process (AMD next year), this means smaller transistors which means less power/heat, fitting more transistors in per chip and reducing production costs. They're also planning on moving down to a 45nm process in 2007, 32nm in 2009 and I think to go beyond that they need a radicall new process due after 2010.

riptide

stevenmu wrote:

Afaik, the increase in transistor count for the EE was mainly to do with the extra cache, not any extra processing logic.



Currently Intels, and to a lesser degree AMDs, have reached a speed limit due to heat and power requirements. To combat this Intel are currently moving to a 65nm production process (AMD next year), this means smaller transistors which means less power/heat, fitting more transistors in per chip and reducing production costs. They're also planning on moving down to a 45nm process in 2007, 32nm in 2009 and I think to go beyond that they need a radicall new process due after 2010.

Beyond that it'll be mad quantum computing with mad physics been involoved... the lithographic process is entering its twilight years...

rsynnott

astrofool wrote:

They saved the die space complexity by making it an in-order cpu rather than out-of-order, which basically makes it a specialist cpu not useful for general purpose computing (i.e. a PC).

I'd question that. If you can build a faster processor for the same money by going in-order, that is a good thing.

Sun Ultrasparc T1 is another example of same.

Stephen wrote:

clockspeed is not the be-all and end-all it used to be.

Er, was it ever a be-all and end-all?

Capt'n Midnight wrote:

Intel thought they could get to 10GHz with the P4 but hit heating problems. LNB's on old analog sky dishes will do 11GHz. In the UK there is an amatuer band at about 250GHz. So faster cores are possible for a while yet.

There's no problem getting something to oscillate at high frequencies. There is a serious problem, though, with building general purpose processors (or indeed any processors) which run at those frequencies.

Tom Dunne

rsynnott wrote:

Er, was it ever a be-all and end-all?

Non-techies always had the misconception that faster is better. Why do you think there was such a race to reach 1Ghz?

Of course, us techies knew better, didn't we?:)

rsynnott

tom dunne wrote:

Non-techies always had the misconception that faster is better.

Faster IS better. Higher frequency isn't necessarily faster

Stephen

What I meant was that processors in years gone by were marketed pretty much exclusively on their clock speed, and to a lesser extent L2 cache. Therefore, the average joe consumer gets conditioned into the belief that higher clockspeed = better CPU, period.

Tom Dunne

rsynnott wrote:

Faster IS better. Higher frequency isn't necessarily faster

Touché

rsynnott

Stephen wrote:

What I meant was that processors in years gone by were marketed pretty much exclusively on their clock speed, and to a lesser extent L2 cache. Therefore, the average joe consumer gets conditioned into the belief that higher clockspeed = better CPU, period.

On a consumer level, assuming you ignore Apple, more or less, yes.