[cipi604]

World chipmakers face most difficult time ever: Hynix
$this->bbcode_second_pass_quote('', 'T')he global memory chip industry is facing the most difficult time ever and could experience a slump for two more years if consolidation is delayed, the head of South Korea's Hynix Semiconductor said Wednesday.

[outcast]

So some of them are having financial problems, how is this related to Moore's law?

[cipi604]

Some today, more tomorrow, Moore's law next year. Think Big!

[outcast]

I'm sure plenty of people said that after the tech bubble burst 8 years ago.

[cipi604]

Only if this was a tech bubble.

[Minvaren]

Moore's law is about the number of transistors on a given die doubling every 18 months, IIRC. In 3-4 generations of chips we hit the limit on miniaturization and electrons start jumping paths if we go any smaller. After that... well, we've nearly maxed clock speed, we're multiplying cores as fast as people will buy them, even though noone's writing multi-core programs yet. For most people, everything except data storage has hit "good enough."

[yesplease]

Bingo! Granted, memory can influence overall production, but Moore's law doesn't say anything about the industry doubling every two years.

[bratticus]

Who cares? Computers are already fast enough for me.

[IslandCrow]

Like oil use and population, in a finite world exponential growth can only go on for a limited time before a physical limit is hit.

[dissident]

The memory makers are already producing some types of flash ram on a 35 nanometer process. Quantum tunneling effects will overwhelm existing IC transistor designs starting at just under 10 nm. So the chip makers are approaching a brick wall quickly. There are only two generations left 22.5 nm and 12.25 nm. Parasitic currents will make the 12.25 nm designs a royal pain.

[outcast]

$this->bbcode_second_pass_quote('Minvaren', 'M')oore's law is about the number of transistors on a given die doubling every 18 months, IIRC. In 3-4 generations of chips we hit the limit on miniaturization and electrons start jumping paths if we go any smaller. After that... well, we've nearly maxed clock speed, we're multiplying cores as fast as people will buy them, even though noone's writing multi-core programs yet. For most people, everything except data storage has hit "good enough."

[mos6507]

Ever since CPUs hit 1Ghz it's been good enough for general computing. Some applications benefit from more speed. Games and content creation (audio recording, video editing, 3D animation) being the biggies. What's happened about 10 years ago, though, is CPUs kept getting hotter and requiring bigger and bigger heatsinks. I had an Athlon 700 back in the day that sounded like a jet plane. So now there is more emphasis on making CPUs run cooler on less power requiring less aggressive cooling. And of course there is a lot of work being done for mobile applications where battery life is key. I think it's time that there not be such a hard line between desktop and mobile hardware. There is a need for efficiency gains just in the motherboard chipsets themselves. Right now you can get boards with Atom CPUs on them where the board itself sucks down a lot more juice than the Atom. So the efficiency is lopsided. Graphics cards are also huge energy hogs.

[lowem]

Here's my Asus Eee Box B202, mounted behind my LCD monitor :



This PC takes up 20W total, and the Atom CPU itself takes up only 2W. It's great for typical Internet surfing and other light duty tasks like Office apps and such. This system actually takes up less power than the monitor it is mounted on (which is 38W).

But then again that's not my primary PC - that distinction belongs to the 64-bit 550W machine with the nVidia 3D card in the living room where it also serves the occasional gaming duty on the big-screen TV.

[bodigami]

There is also the price of making CPU factories; it's becoming bigger and bigger. Probably there will be a time where only 2 or 3 CPU makers remain: Intel and everyone else; or Intel, generic x86 and "the other architecture (either Sparc or PPC).

It's getting too expensive to just produce current gen CPUs; computing can face a serious slowdown (and if that greatly reduces e-waste then I welcome this change).

[yesplease]

$this->bbcode_second_pass_quote('lowem', 'H')ere's my Asus Eee Box B202, mounted behind my LCD monitor :

[lowem]

$this->bbcode_second_pass_quote('yesplease', '')$this->bbcode_second_pass_quote('lowem', 'H')ere's my Asus Eee Box B202, mounted behind my LCD monitor :

[yesplease]

Even then, someone could throw something together that's roughly as efficient for hundreds less. For instance a 64 bit AMD or Intel system could be thrown together for ~$150 that only uses ~20-30W. Granted, I ain't saying that paying for the fit and finish is bad, just that it can easily done for way less cash.

[lowem]

$this->bbcode_second_pass_quote('yesplease', 'E')ven then, someone could throw something together that's roughly as efficient for hundreds less. For instance a 64 bit AMD or Intel system could be thrown together for ~$150 that only uses ~20-30W. Granted, I ain't saying that paying for the fit and finish is bad, just that it can easily done for way less cash.

[yesplease]

$this->bbcode_second_pass_quote('lowem', 'W')hile I kinda doubt your 20-30W claim for a full-sized desktop PC system

[outcast]

Processing power will never be "good enough" for quite a few people. It isn't just the gamers and graphics artists who want more, it is also the engineers and scientists. Take for example the LHC, in order to process all the data they had to design a special, ultra high speed network to feed the thousands of processors needed to processes the enormous amount of data coming in.

link

$this->bbcode_second_pass_quote('', 'T')he Worldwide LHC Computing Grid relies on dedicated optical fibre networks to distribute data from CERN2 to eleven major computer centres in Europe, North America and Asia. From these, data is dispatched to more than 140 centres around the world. Together, these distributed computers provide the power to manage the LHC’s data.

“We can routinely process 250,000 jobs a day,” said Bird, “and we can achieve peaks of 500,000 jobs without problems.” A single job can be a calculation lasting several hours or even several days on a single high performance processor. An estimated 100,000 processors are needed to handle all jobs from the LHC experiments.