User: Password:
|
|
Subscribe / Log in / New account

Small correction

Small correction

Posted Oct 1, 2007 19:30 UTC (Mon) by khim (subscriber, #9252)
Parent article: Memory part 2: CPU caches

This is not due to the fact that faster RAM could not be built, as explained in the previous section. It is possible but it is not economical. RAM as fast as current CPU cores is orders of magnitude more expensive than any dynamic RAM.

Actually it's not true anymore. While it's certainly true that it's possible to build RAM cells which are as fast as CPU cores it does not mean that it's possible to build RAM subsystem which is just as fast. And in fact it's totally impossible today. Not for $1'000'000, not for $1'000'000'000, not even for $1'000'000'000'000. Why? It's easy. Speed of light is 300'000km/sec (exactly 299'792'458m/s since October 21, 1983). It's not uncommon to have 3GHz CPUs today. This means that information can only travel 10cm in one tick - and it should travel back and forward, so only 5cm - less than 2in! If you'll take a look on DIMM you'll find that it's length is 133m. More than 10cm! So even if you plug DIMM module directly in CPU and cells on said model of RAM have infinite speed - you still can not get RAM which is as fast as current CPU core!

Sure - we can make geometrically smaller module and put it closer to CPU, but if we'll want to have gigabytes of RAM - we'll be forced to store only parts of this information in this superfast RAM, the rest will go farther from core... oops - we just invented the cache again! So no, this is not just question of the economics anymore - IT industry stumbled upon first fundamental limitation few years back...


(Log in to post comments)

Small correction

Posted Oct 1, 2007 21:26 UTC (Mon) by epa (subscriber, #39769) [Link]

That may be true, as long as you are restricted to building your circuit in two dimensions... who is to say we won't discover a way to build integrated circuits in a sphere rather than on a flat wafer?

Small correction

Posted Oct 1, 2007 23:20 UTC (Mon) by quotemstr (subscriber, #45331) [Link]

Consider the heat dissipation issues of using a sphere.

Spherical chip design

Posted Oct 2, 2007 2:14 UTC (Tue) by pr1268 (subscriber, #24648) [Link]

Just a thought: implement an array of liquid-cooling jackets. It's worked for years in diesel- and gasoline-powered internal combustion engines.

Hope it doesn't spring a leak! ;-)

Small correction-already in the works...

Posted Oct 2, 2007 0:20 UTC (Tue) by vomlehn (subscriber, #45588) [Link]

There was an article in SlashDot not too long ago (sorry, don't have the reference) about IBM being about to use metal vias (essentially, vertical wiring) for building what are basically 3D chips. This will allow memory to be much closer to the processor, for a given amount of memory. There will, of course, be heat dissipation problems galore.

Small correction

Posted Oct 2, 2007 0:27 UTC (Tue) by Richard_J_Neill (subscriber, #23093) [Link]

Not quite - we aren't limited by speed of light, but by power dissipation. One could certainly fit 100GB of RAM inside the 5cm boundary, by making a 3-dimensional structure. However, there would be no way to keep such a solid "brick" of transistors cool...

[Of course, I hope to be proved wrong in that!]

Small correction

Posted Oct 2, 2007 4:54 UTC (Tue) by AJWM (guest, #15888) [Link]

> However, there would be no way to keep such a solid "brick" of transistors cool...

That's where research into new substrates comes in. Synthetic diamond, for one example (diamond being an excellent heat conductor). (Hey, you didn't think the fancy crystalline computer modules in various SF TV shows were just glass, did you?)

Hot?

Posted Oct 2, 2007 21:06 UTC (Tue) by ncm (subscriber, #165) [Link]

Who says memory has to be made of transistors? In the past, memory has been made of ripples on mercury, ferrous donuts, rotating drums, holes in paper, glow in phosphors bombarded by electron beams... storage need not run hot. In particular, bits not changing state need not consume any power. Changing state may take arbitrarily little power; the faster they must change, the more power it usually takes, but write speed is less critical than read time. Interacting with it is always going to take power and produce heat, but that may be much, much less than with masses of transistors.

Hot?

Posted Oct 4, 2007 13:14 UTC (Thu) by ekj (guest, #1524) [Link]

Actually, that's not true. Bolzmanns constant sets an absolute, physical, lower limit on the amount of power that is needed for causing a permanent lasting state-change. (such a flipping a single bit)

Granted, that limit is *very* low. But it's not zero. I calculated some time back (if you're sufficiently interested, google it) that if we continue doubling computing-power we'll run up against this hard physical limit in aproximately 15-20 years.

That's a long time in computing. But it's not forever. It's short enough that most of us will get to experience it.

Oh yeah, I'm aware of reversible computing. I just don't think that'll go anywhere. I'd be happy to be proven wrong.

Small correction

Posted Oct 2, 2007 8:35 UTC (Tue) by nowster (subscriber, #67) [Link]

What you're describing is latency, not throughput. They're two different measurements of "speed".

Small correction

Posted Oct 3, 2007 2:36 UTC (Wed) by k8to (subscriber, #15413) [Link]

He knows.

The thing about memory is that latency kills your processing throughput. If the CPU needs the data and it isn't available then the thread stalls until it becomes available. There's various techniques which can sometimes hide this problem, which is what this section of course is largely about.

In the context of making memory which does require cache (the topic of the post), the speed that matters is necessarily latency.

Light speed

Posted Oct 2, 2007 12:14 UTC (Tue) by smurf (subscriber, #17840) [Link]

It gets worse.

Light speed is defined as photons moving in a vacuum. Anything else, which definitely includes electrons in a semiconductor, is slower.

It gets worse.

You need to actually switch a transistor or two on the way to+fro (otherwise there would be no point in the exercise ;-), which requires moving a charge, which requires yet more time.


Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds