Why couldn't they make todays best cpu ten years ago?

Radiation · 6 Sep 2011 at 17:56

Just a thought i had recently while looking into processing power and the progress of technology, to me it feels like things have slowed, moores law says the number of transistors doubles about every two years but that doesn't necessarily mean we get anywhere near double the performance in general use, so i wondered what made it physically impossible to get the equivalent of todays cpu just ten years ago?

I know architectures etc change and improve but don't we know enough now that we could make big jumps in processing power, i mean we have the knowledge and tools, is it about money as always, do the chip makers just want a steady product line for the next 50 years?

Philz0r · 6 Sep 2011 at 17:58

i remember when IBM made their first 1gig processor it was a pretty big deal, now we got things like quad core's.

p4radox · 6 Sep 2011 at 18:02

Because most new technologies and processes require a new fab to built somewhere, and this takes time and billions of dollars.

QlimaxUK · 6 Sep 2011 at 18:23

make one yourself if you think you can do better

Craig R · 6 Sep 2011 at 18:26

We didn't know how to.

Tbid · 6 Sep 2011 at 18:27

Research and Innovation, it's like anything really - There were aeroplanes over 100 years ago but it took around 40 years to develop the jet engine and break the sound barrier

Longbow · 6 Sep 2011 at 18:43

QlimaxUK said:
make one yourself if you think you can do better

He doesn't, he was asking why it took 10 years to get to the processors we have today.

aardvark · 6 Sep 2011 at 18:44

Craig R said:
We didn't know how to.

Liampope · 6 Sep 2011 at 18:51

Design and development is an iterative process. You improve on what has come before and need similar developments in materials/manufacturing processes in support. Big leaps are possible within reason, but you can't just say 'right guys let's design whatever CPU, car, hoverbike they'll have in 50 years time... right now!' Because you have no idea what it will be like and have no way to make it.

Edit: I think I may have interpreted your question a bit generally and replied likewise (well it is in General Discussion). Never mind you get the idea though.

Pawnless Endgame · 6 Sep 2011 at 19:08

I remember us chasing the megahertz back in the 90s and early 00s, like top clockspeed would be 1.5x more the next year than in the previous year. E.g. 1996 200MHz, 1997 300MHz, 2008 450MHz and so on. I remember looking on OcUK in late 2003 and they were stocking 3.6GHz and 3.8GHz as the top 2 Intel processors. In 2004, the top speed then went back down to 3.4GHz with hyperthreading introduced. In 2006, hyperthreading got replaced by dual-core. Then dual-core got replaced by Core 2 Duo™ then 4, 6, 8 cores in later years. For all of this time, the top speed remains around 3.4GHz, but I guess that processor in dual-core would be like having a 6.8GHz CPU?

SS-89 · 6 Sep 2011 at 19:12

The research for next generation cpus is done years in advance of their release but the logistics of building fabrication plants and testing etc takes many years to catch up.

Edit: Also processor speed is irrelevant. It's all about efficiency and the amount of computational power under the hood.

Ceryndrion · 6 Sep 2011 at 19:19

First of all is the technology to make the chips in the first place, to reliably make smaller and smaller dies, with more complexity, takes a lot of time and effort, and money. As has been said before, many smaller iterations.

To go with that, are the required support structures, new motherboards have to be developed, licenses granted, testing performed.

All in all, it takes a lot of work, between multiple companies, and just isn't something that we have been able to do right off the bat, infact, you will probably find, that in order to create the next generation of processor, we've needed this generation to do so..

Like deep thought said, after ten thousand years, he wasn't powerful enough to answer the question, but he could design a computer that would be.

Andr3w · 6 Sep 2011 at 19:21

I wonder what they will be like in 10 years!

Do you think we will get to some kind of bottleneck or keep advancing at the rate we are?

Pawnless Endgame · 6 Sep 2011 at 19:26

Andr3w said:
I wonder what they will be like in 10 years!

Do you think we will get to some kind of bottleneck or keep advancing at the rate we are?

We'll probably still be stuck at 3.4GHz (like ever since 2004), but with with more cache and cores, and perhaps dual-CPU will drop into consumer pricing range.

Chasser · 6 Sep 2011 at 19:28

oops sorry OP.

gord · 6 Sep 2011 at 19:29

Monserrat said:
but I guess that processor in dual-core would be like having a 6.8GHz CPU?

Not really. Because we've struggled to push the frequency of the processor further (at least feasibly, we could have them go faster but you get diminishing returns on the power/heat) we've gone to multiple cores. Which means the onus now is really on the development of software to use the parallel architecture.

This has inherently been quite difficult, because: a) humans like to do one thing at a time and b) no one has ever had to think about parallel workloads. Only in the last 5~ years have we seen the SDKs take leaps forward. And now we're seeing some exciting automated parallelisation.

Jono · 6 Sep 2011 at 19:38

It's because the companies that make the process control tool can only develop so fast

Jono · 6 Sep 2011 at 19:39

You could always read this

http://download.intel.com/pressroom/kits/advancedtech/pdfs/Kuhn_variation_ICCV2009.pdf

Energize · 6 Sep 2011 at 19:44

QlimaxUK said:
make one yourself if you think you can do better

Random...

Energize · 6 Sep 2011 at 19:47

Monserrat said:
I remember us chasing the megahertz back in the 90s and early 00s, like top clockspeed would be 1.5x more the next year than in the previous year. E.g. 1996 200MHz, 1997 300MHz, 2008 450MHz and so on. I remember looking on OcUK in late 2003 and they were stocking 3.6GHz and 3.8GHz as the top 2 Intel processors. In 2004, the top speed then went back down to 3.4GHz with hyperthreading introduced. In 2006, hyperthreading got replaced by dual-core. Then dual-core got replaced by Core 2 Duo™ then 4, 6, 8 cores in later years. For all of this time, the top speed remains around 3.4GHz, but I guess that processor in dual-core would be like having a 6.8GHz CPU?

Yes, for an embarassingly parallel task like video encoding, performance increases linearly with the number of cores. So 2 cores is 2x as fast and 4 cores 4x as fast.