Hypothetical: Speed of computers near limit?

[acme]

to my eye, it's been awhile since any significant numerical increase in CPU numbers. The top chips seem to be in the neighborhood of 3.xx Ghz, with threading, and GPUs helping to boost computer performance.

Maybe we're at the limit of Moore's law, not in an absolute sense, but in terms of manufacturable, shippable products.

My question, and it's purely academic...how much more capable or faster can computers get compared to what we have right now..current Macs?


a

[SirDice]

Not much faster I guess. With higher frequencies we're bound to run into some physical limitations. Like the speed of light.

Memory banks for example would need to move closer to the CPU. The physical space between the modules would result in signals arriving out of sync.

NB Moore's Law doesn't really say anything about the speed of the processors. Just that the amount of transistors on a chip doubles every 2 years. Granted speed is heavily linked to this but that's not what Moore said.

[NovaScotian]

The half-wavelength of 3 GHz is about 2 inches for a 180 degree phase shift. CPUs are nearing that size.

[acme]

so, this relates to the signals out of phase issue and basically a physics limitation to cpu clock speeds getting faster?

..so that performance gains will come from where?

or will they? Are we about there on all fronts?

a

[macosnoob]

A few years back, I recall an Intel honcho saying that performance gains for the next decade would come through multiple processors rather than any giant leaps in processor speed. Certainly seems to be Apple's approach with the Intel architecture--dual core, quad core, octo core, etc. How many cores are theoretically--and then realistically--possible?

[NovaScotian]

Quote: Originally Posted by acme so, this relates to the signals out of phase issue and basically a physics limitation to cpu clock speeds getting faster? ..so that performance gains will come from where? or will they? Are we about there on all fronts? a

Where that becomes problematic is when timing signals must be sent from a single source all over the chip. Those a short distance away and those far away might well be out of phase. The multiple core approach can mitigate that. In my view, the next big breakthrough will not be 3D architectures (i.e., stacked cores) but rather compilers that can deal with those problems.

[fracai]

Quote: Originally Posted by macosnoob How many cores are theoretically--and then realistically--possible?

Also, how well can you parallelize a process? How many applications would actually make use of more than even just a handful of cores?

Apple is already countering this trend though. How many people know off hand what the hardware specs are of Apple's iOS devices? They're promoting other features, because the hardware largely doesn't matter; it's currently "good enough".

[acme.mail.order]

Quote: Originally Posted by fracai it's currently "good enough".

Yep. Also, in the individual computer world (largest market) the CPU is no longer the bottleneck so making the CPU faster doesn't make the computer faster.

I have a 2.8GHz machine and the only thing I do that takes significant time is video transcoding. Doubling the CPU speed or adding more cores won't pull data off of the single read head or push it onto the destination drive any faster.

[SirDice]

Quote: Originally Posted by acme.mail.order Yep. Also, in the individual computer world (largest market) the CPU is no longer the bottleneck so making the CPU faster doesn't make the computer faster. I have a 2.8GHz machine and the only thing I do that takes significant time is video transcoding. Doubling the CPU speed or adding more cores won't pull data off of the single read head or push it onto the destination drive any faster.

I agree, CPU speed is not the issue anymore. The rest of the computer is severely holding things up. Memory and I/O speed are the main bottlenecks now.

[NovaScotian]

Quote: Originally Posted by SirDice I agree, CPU speed is not the issue anymore. The rest of the computer is severely holding things up. Memory and I/O speed are the main bottlenecks now.

And it has ever been thus, really. Many years ago now, I was involved in the development of a multiprocessor device using the then newest Zylog Z-80s running at 4 MHz. The bottleneck to work around was that the PIOs would only run at 2 MHz and they were supposed to be daisy chained, i.e., lots of delays between the first and last in the chain. We had to build our own selector to run them in parallel or latency really screwed things up.

A few years ago now, I significantly speeded up a PM G5/2.3 GHz machine by installing a PCIe to SATA card to run a 3G 10,000 RPM hard drive to replace the 2G 7,200 drive it came with.

It's always been thus; that’s why SSDs are so popular. With the advent of cloud computing, of course, the hang-up is your Internet bandwidth.

[Jasen]

Quote: Originally Posted by fracai Also, how well can you parallelize a process? How many applications would actually make use of more than even just a handful of cores?

Aye, that is a huge question. I've done some multi-threaded/concurrent programming. Taking advantage of all cores/processors is hard. You've got to figure out how to break a linear process into parallel chunks (sometimes impossible, if each chunk relies on results from the others), or figure out how to restructure the application so that different parts of it can run independently and yet still communicate.

The biggest benefit of multiprocessing is not to be had at the application layer, but at the OS layer. Having an OS that is aware of how many processors it has available so it can balance all running tasks across them in such a way as to eliminate (or at least mask) latency is paramount. OS X and Windows 7 are both pretty good at that, in my experience. I've experimented with writing simple programs that do some redundant math problem, and spawning multiple new threads doing the same thing. Both OSes were good at moving the threads around to balance the CPU load on all 8 cores on my Mac Pro.

So now, with the new trend in processors, and with the OS capability to load balance, all developers should be trying to keep in mind ways to break intensive tasks up into independent threads in their applications. And I see this being the future state of programming indefinitely.

[benwiggy]

I concur that chip architecture is not the limiting factor.

There are some excellent youtube videos of identical Macs booting up with and without SSDs. The SSD are massively faster than mechanical hard drives. Again, other bottlenecks in GPUs, buses, IO handling -- you can probably double the speed again (after SSDs) on the same CPU with some magical new technologies for all the other stuff.

[acme]

One conclusion from this excellent thread is that, to make a computer already sitting on one's desk faster, simply remove its bottlenecks...

faster drives..more ram...better card..

..up to a point. Several of the bottle necks described here do not sound like user-installable or changeable things..

a