Moore

[zeta_immersion]

I wish I had 1/3 of the insight these guys have in physics and the universe, research and idea either wrong or right

[Andre S. Veteran]

Actually, if I remember correctly, Moore's law was based around transistors on a chip doubling every 15-20 months-ish. I don't think clock-speed had ever had anything to do with it.

Not directly, but until P4 Moore's law effectively translated into higher clock speeds. That meant you could write a slow program and it would run twice as fast in 18 months. That era is gone, you can't expect future technology to make your inefficient program fast in two years. Crysis was slow in 2007 and it still is, to take a bad example (hence the "Can it run Crysis" meme). Newer CPUs give you more cache, more cores, a somewhat more efficient architecture, but very little single-threaded performance improvements.

[CentralDogma]

I don't think that's exactly true. The atom itself is not a 'hard limit' to anything, really. There are certainly levels of structure below atoms, though that area gets very weird and complex. Whether we can use it is another matter, but I don't think it's reasonable to suggest the atom is a point of finality.

When you get into sub atomic particles, things are governed by a different set of physical laws. You can't build a sub atomic transistor. Someone has managed to build a single-atom transistor, but that is the smallest possible.

Quantum Computing is where things will probably move to after fabrication technology reaches the thickness of a silicon atom.

[Hum]

I've often wondered just how fast do computers need to be ....

For most home users, they reached top speed, years ago. :laugh:

[sc302 Veteran]

This has been well-known for, well, ever. The Pentium 4 was 3.2Ghz in 2004, Intel's latest CPU is now 3.5Ghz. Moore's law has already stopped yielding faster clock speeds a while ago. Since the P4, it's yielding more transistors on a chip hence more similarly-clocked cores. This is great for parallelizable computations, but not everything is perfectly parallelizable. Running things in parallel presents unavoidable overhead that often make it not even worth it. Especially in video games, where frame n+1 is a function of frame n so it's impossible to compute both in parallel; and each frame must be computed very quickly (< 1/30th of a second) so the overhead of spreading the job on multiple cores is often greater than the potential gains. Programmers are only starting to figure out how to program for true parallel architectures, and it's unclear how that approach will evolve.

There might be replacements for silicon, but will they be as flexible, inexpensive, will they really allow the same kind of exponential growth we've had for 50 years? Only time will tell.

See here is the funny thing where the clock speed of a single chip has not changed much, the cores have multiplying the ability of a chip that has a single core by far. So while the clock speed hasn't changed all that much the processing speed has. You can't tell me that a 3.2 p4 is faster than a 2.8 core 2 duo.

[Andre S. Veteran]

See here is the funny thing where the clock speed of a single chip has not changed much, the cores have multiplying the ability of a chip that has a single core by far. So while the clock speed hasn't changed all that much the processing speed has. You can't tell me that a 3.2 p4 is faster than a 2.8 core 2 duo.

C2D has better single-threaded performance than P4 through larger cache and a much more efficient architecture. P4 was notoriously inefficient btw. Even there, the single-threaded performance gap between P4 and C2D, considering 3 years elapsed between them, is not what got between, say, P2 and P3 or P3 and P4. It's quite fascinating that Intel is still able to increase single-threaded performance at all with its new processors, I wonder if and when the point of diminishing returns on investement will be hit.

[DocM]

And we have reached peak oil. The optimistic economist will always trump the pessimistic scientist. At least this has been the case for the past century or so.

Don't be so sure. That assertion is made using the numbers for "proven reserves," basically what we have a 90+% certainty of - a very conservative number. For the US this is stated by the administration as 22 billion barrels - 15th in the world.

Know what's silly? This 22 billion barrel number has changed little since 1946, and since then about 150 billion barrels has been extracted from US sources. Hmmmm....

On the other hand, and using US Department of Energy numbers, there is a very much larger amount of oil available in the US. The below pyramid is pretty much self explanatory, but remember these are DoE numbers using conventional technologies, which are improving so fast it's insane.

[CentralDogma]

C2D has better single-threaded performance than P4 through larger cache and a much more efficient architecture. P4 was notoriously inefficient btw. Even there, the single-threaded performance gap between P4 and C2D, considering 3 years elapsed between them, is not what got between, say, P2 and P3 or P3 and P4. It's quite fascinating that Intel is still able to increase single-threaded performance at all with its new processors, I wonder if and when the point of diminishing returns on investement will be hit.

Small fabrication technology means there's more room for Intel to work with. They can implement what was once software instructions via hardware.

That's not to say they don't make general architecture improvements. With the army of engineers they have, I can only assume they are able to conger up something approximating computing magic.

Anyway, from my understanding, the Pentium 4 architecture was so terrible thanks to it's huge pipeline and bad branch prediction. It would make a guess when it hit a conditional branch that was yet to be computed, pushing it through it's huge pipeline, end up guessing wrong and having to flush the pipeline, literally wasting computation. Intel ended up abandoning the netburst (P4) architecture and iterated on the Pentium 3 to create the Core architecture.

[123456789A]

Don't be so sure. That assertion is made using the numbers for "proven reserves," basically what we have a 90+% certainty of - a very conservative number. For the US this is stated by the administration as 22 billion barrels - 15th in the world.

Know what's silly? This 22 billion barrel number has changed little since 1946, and since then about 150 billion barrels has been extracted from US sources. Hmmmm....

On the other hand, and using US Department of Energy numbers, there is a very much larger amount of oil available in the US. The below pyramid is pretty much self explanatory, but remember these are DoE numbers using conventional technologies, which are improving so fast it's insane.

That's brilliant! Oil based microprocessors!

[Nashy]

If Michio Kaku said it, then I believe it. Not like we couldn't see it coming.

Your theory may be correct, but calling him an idiot? He's a very smart man. Surely smarter than both of us.

Lol. So what you're saying is that you will blindly believe what someone says, even if less than 30 seconds of research finds inaccuracies?

That's why the world's democratic Governments are currently falling apart. Because people blind vote. But hey. Who cares.

[TEX4S]

Lol. So what you're saying is that you will blindly believe what someone says, even if less than 30 seconds of research finds inaccuracies?

That's why the world's democratic Governments are currently falling apart. Because people blind vote. But hey. Who cares.

Dude, I hope you were kidding -

[TEX4S]

delete