Intel / Moore's law

AllForIt

Capt'n Midnight wrote: »

The point is that clock frequency on CPUs hasn't gone up much in ages, even though sizes are much smaller and interconnects are shorter.

10 years ago you could buy 3 GHz CPU's in 90 nanometre. Today it's 10nm with 7nm in production and we aren't getting many more gigahertz.

210-GHz in 2001 I want my superfast CPU now :mad:

Well, that's because size isn't directly proportional to clock frequency. That's like thinking if a car was made smaller it should be able to travel faster - it might a bit - but not proportionally.

The frequency is limited by the capability of the physical hardware being used. The physical hardware being used currently is silicone. It's speed has more or less reached it's max, it can't go much faster, because it's beyond the physical capability of the material and that can't be changed.

What has happened though is that more than one processor exists on the chips now which is a clever way to speed things up. Instead of one digger digging a hole, you use 4 or 8 at the same time, and the hole gets dug much faster. So making things smaller has in a way created the illusion that processor speeds have increased but they haven't really.

The only way to really speed up processor speeds is to find a material that is capable of working at a stable frequency higher than that of silicon. If one could discover that material, you'd be a multi billionaire overnight.

Brian?

srsly78 wrote: »

I spend all day running angstrom scale simulations to help reduce critical dimensions, am pretty gutted to hear that this was all invented in the 70s and we are just wasting our time

BTW: Samsung are ahead of Intel today.

Ahead how? I guarantee they aren’t ahead in high end processors like the Xeon.

Brian?

astrofool wrote: »

Not just that, you need to be planning 5-10 years in advance to have the billion dollar fab ready for when the technology is commercially viable to make the smaller chips, one mis-step can give your competitors an advantage, Global Foundries (used to be part of AMD) are actually ahead right now with their most advanced Fab, and Intel are playing catch-up due to a few mis-steps in their latest die shrink. Beyond smaller transistors, vertical stacking is now very popular, enabling transistors to be stacked upwards as well as outwards, and allowing Moore's law to continue for a while yet.

There’s are a host of reasons why Global Foundries aren’t ahead in their process. They’re producing chips on a 10nm process with a higher pitch size than Intel. This means their transistor density is lower than intel’s on 10nm.

I’m half afraid to share other reason as I’m not sure what’s publicly known.

Brian?

AllForIt wrote: »

Well, that's because size isn't directly proportional to clock frequency. That's like thinking if a car was made smaller it should be able to travel faster - it might a bit - but not proportionally.

The frequency is limited by the capability of the physical hardware being used. The physical hardware being used currently is silicone. It's speed has more or less reached it's max, it can't go much faster, because it's beyond the physical capability of the material and that can't be changed.

What has happened though is that more than one processor exists on the chips now which is a clever way to speed things up. Instead of one digger digging a hole, you use 4 or 8 at the same time, and the hole gets dug much faster. So making things smaller has in a way created the illusion that processor speeds have increased but they haven't really.

The only way to really speed up processor speeds is to find a material that is capable of working at a stable frequency higher than that of silicon. If one could discover that material, you'd be a multi billionaire overnight.

Gallium Arsenide grown using epitaxy on an Si substrate?

AllForIt

Brian? wrote: »

Gallium Arsenide grown using epitaxy on an Si substrate?

I'm hoping for single atoms capable of being stabilized in such a way that they can be contained in a designed structure where they can be manipulated to represent 1 or 0.

Capt'n Midnight

AllForIt wrote: »

Well, that's because size isn't directly proportional to clock frequency. That's like thinking if a car was made smaller it should be able to travel faster - it might a bit - but not proportionally.

It was until they hit 3Ghz.
At 3GHz light in a vacuum travels 10cm in one clock. So why are the same propagation delays still there when we are using features a tenth the size on smaller chips ?

The frequency is limited by the capability of the physical hardware being used. The physical hardware being used currently is silicone. It's speed has more or less reached it's max, it can't go much faster, because it's beyond the physical capability of the material and that can't be changed.

SiGe can be grown on Si like the 210GHz transistor above and Intel have been using SiGe for ages to produce "Strained Silicon"

What has happened though is that more than one processor exists on the chips now which is a clever way to speed things up. Instead of one digger digging a hole, you use 4 or 8 at the same time, and the hole gets dug much faster. So making things smaller has in a way created the illusion that processor speeds have increased but they haven't really.

Great if you have highly parallel tasks. Otherwise it's diminishing returns , especially if there's a lot of task switching overhead. And in many cases the clock slows down as the chip heats up when the extra cores kick in.

The only way to really speed up processor speeds is to find a material that is capable of working at a stable frequency higher than that of silicon. If one could discover that material, you'd be a multi billionaire overnight.

Gallium Arsenide is the technology of the future.
Always has been, and always will.
- 1980s quote that's still fairly true.

and besides
Silicon-Germanium transistor at 0.798 THz



And there's still photonics, graphene, and bio-molecules waiting for a breakthrough.

Capt'n Midnight

AllForIt wrote: »

I'm hoping for single atoms capable of being stabilized in such a way that they can be contained in a designed structure where they can be manipulated to represent 1 or 0.

Cosmic rays.

you'll need to provide cross linked redundant circuits , so at least 3 atoms

10000maniacs

Capt'n Midnight wrote: »

Sky dishes have been running at over 10GHz since 1989.

But silicon is relatively cheap and easy to fabricate. You can use it as a base to build other semiconductors on, if you want to.

Intel have been focusing on lower power in laptop processors. But the laptop makers have just reduced battery sizes so it doesn't translate into longer battery life. So today's entry level laptops have gotten lighter but not much faster in recent years.

Yeah but the first Intel 4 ghz 64 bit dual core processor was introduced in 2002 and in 2018 this is essentially what you are still getting with a good home PC. Between 1979 and 1992 clock speeds for home PCs have increased x 4000 and bus width increased x 8. Moores law died in the late 90s.

ED E

10000maniacs wrote: »

Yeah but the first Intel 4 ghz 64 bit dual core processor was introduced in 2002 and in 2018 this is essentially what you are still getting with a good home PC. Between 1979 and 1992 clock speeds for home PCs have increased x 4000 and bus width increased x 8. Moores law died in the late 90s.

Thats throwing away IPC completely which is very misleading.

What 4Ghz 2C chip was out in 2002?

10000maniacs

ED E wrote: »

Thats throwing away IPC completely which is very misleading.

What 4Ghz 2C chip was out in 2002?

The intel Zeon 4.0. And, correction, it was a quad core processor.
IPC and pipeline architecture are not very different in todays chips compared to the 2002 Zeon. You can't indefinitely deepen a pipeline and expect to improve performance. Diminishing returns.

Capt'n Midnight

10000maniacs wrote: »

Yeah but the first Intel 4 ghz 64 bit dual core processor was introduced in 2002 and in 2018 this is essentially what you are still getting with a good home PC. Between 1979 and 1992 clock speeds for home PCs have increased x 4000 and bus width increased x 8. Moores law died in the late 90s.

Moores law, as in transistor size didn't die in the late 90's.

But performance improvements died.

Today's entry level laptops aren't significantly faster than 5 or 10 years ago and are a lot less upgradable. Before that computers would be obsolete before the extended warranty had expired.

Thanks to bitcoin prices for SSD's and good video cards have been daft in recent times too. Not Intel's fault, but they could have been churning them out. Another reason why you haven't been getting more bang for your buck these days.

ED E

Capt'n Midnight wrote: »

Thanks to bitcoin prices for SSD's and good video cards have been daft in recent times too.

Whuuu? Some of your posts are on the money and the same time some are wacko.

SSD prices have trended really well, no upsets there. Crypto doesnt impact them (nobody burstscoins on SSD).

Crypto spiked GPU pricing.

Memory pricing spiked due to market manipulation.

ED E

10000maniacs wrote: »

The intel Zeon 4.0. And, correction, it was a quad core processor.
IPC and pipeline architecture are not very different in todays chips compared to the 2002 Zeon. You can't indefinitely deepen a pipeline and expect to improve performance. Diminishing returns.

Again what? Its Xeon not Zeon.

Fastest Xeon in 2002 was 3Ghz that I can tell. Slight improvement on that in 2003.



Have to say I'm skeptical you're actually at all familiar with this.

10000maniacs

ED E wrote: »

Again what? Its Xeon not Zeon.

Fastest Xeon in 2002 was 3Ghz that I can tell. Slight improvement on that in 2003.



Have to say I'm skeptical you're actually at all familiar with this.

Xeon Zeon. Typo.
I too am skeptical about your knowledge on the subject. Quick question. How and why do you think the current batch of Intel processors have a vastly superior IPC rate than the 2002 Zeon/Xeon?

10-10-20

Ubbquittious wrote: »

I never trusted this Intel crowd.

For the past 40 years or so like clockwork

Tick Tock. [>]