Re: Again: Building my Own Laptop

Den 31-12-2012 10:49, Rod Pemberton skrev:

Of course they will. It's a business. You pay, they Play:

Same story with CAM and 3D-printing - today there are many manufacturers that will run your designs from CAD files.

Was everything I read in the 70s and 80s wrong?

He predicted the number of elements per IC would double every year, and it did from 1959 to the early 80s.

In the 70s, everyone said we would hit a wall, probably in the 80s. It was pointed out that if progress did not slow, we would have billion-transistor chips by 2000. It did take several years longer.

When we hit the wall in the 80s it was said that Carver Meade rewrote the book on processes, and enabled continued progress.

But that progress was reduced to doubling every 18 months, and recently every 2 years.

Every citation of "Moore's Law" in the 70s said "every year", but people keep redefining it. Sure, it will last forever if you redefine it.

We are now nearing these newer limitations.

We will end up on optical computers before long. Then the march starts all over again.

In 1965 Moore observed that growth was doubling every year and he saw no reason for it not to continue for 10 years. So extrapolating beyond

This graph shows little variation from this slope over the last 40 years. So I would say that Moore's law likely held during the 10 years he expected it to and has been slightly declining since, but has been pretty consistent over the last 40 years at this lower rate. I don't see any reduction in the rate in more recent years, in fact it looks like there was some recovery of the rate of growth over the last 12 or

Is there some distortion to the curve because of using only microprocessor data and not memory, etc?

Rick

Sorry, forgot to post the link to the graph. Here is the Moore's Law page with the image.

Rick

Possibly. In I computed the price/bit of DRAM (from 1983 to 2002), and got an average halving in 17 months.

- anton

-- M. Anton Ertl

Regardless of whether the curve is softening a little or not, I am amazed that it has lasted as long as it has. Lately I have been hearing that we may really be reaching the end of what can be done "easily", meaning it is getting a lot harder to keep extending that curve. I understand we will be seeing some fundamental limits becoming real problems around 10 nm... but I have heard that song play on the radio before.

From what I've heard, the real issues are not advances in process technology, but that we don't have any good use for more transistors in CPUs or more CPUs on a chip. Plus, the market has been changing so that more transistors aren't what is needed, but less power consumption for the highly portable devices.

I was a bit surprised by the huge adoption of mobile computing over the last 10 years. Having gotten over my surprise, I expect the PC to become a much smaller player to the handheld and table form factors with the resulting emphasis on very low power and the resulting change in emphasis in processing technology from density to power consumption.

Rick

Indeed.

That's an old song, too.

I've been hearing that song for a couple of decades, too.

The reality is that we can't know what the next "big thing" is. We do know it's not the current "big thing".

Not sure why you say that is a "couple of decades" old. In '93 they were still pushing for "longer, lower, wider" to quote the auto industry's motto during the years when they blithely promoted fancier cars with shiny doodads instead of safety or lower pollution. PCs didn't reach their peak power dissipation until they were over 100 Watts in what, 2000 something? Now they hardly have a PC CPU that uses 100 Watts. I think even the ultra powerful server CPUs try to keep the power consumption down as it costs more to cool the equipment than it does to power it.

Rick

Perhaps not '93, but by '95 the main issue of PCs had turned to power. It didn't get fixed, mainly because the packaging people were better than anyone gave the credit for.

Just because the problem wasn't "solved" (it really never was - people just got bored with balls-to-the-wall performance), doesn't mean it wasn't a primary concern.

FPGAs are programmed in the field not in the factory, every time you reset the device you need to reprogram the FPGA.

I'm not not sure what a "professional motherboard manufacturer' is but the likes of PCBCart will manufacture the boards I doodle up in (gEDA) PCB

Its not hard to open a "cash" account with the three big online suppliers (digikey, mouser, farnell) You may need to register a one-man business first.

throug-hole can be soldered wih a soldering iron, as can leaded SMD (like SOIC). (lead as in wire, not as in Pb) Non-leaded like QFN and BGA require an oven, with a controlled temperature profile. an ordinary kitchen appliance with a conscious operator watching the temperature and the clock can probably fit the bill if he's only making one.

the more more you order the cheaper for each. for the price to be competitive it will need a large order.

only two stacks, and tiny ones at that, if you like stacks why not use a M68000 instead?

forth is an incremental compiler hardware drivers can probably be done in assembler and forth.

I'm not sure what your point is. There was a tradeoff between building chips which were larger and ran faster and chips that used less power. The fact that they continued to build the faster chips and didn't build the lower power ones says to me they are only now being forced by the market to go for low power at the expense of performance.

My point is that power will be the primary issue with CPUs in the coming years. Even today many are not so happy with cell phones that have to be charged more than once a day. Processing speed will be taking a secondary seat to power consumption and the product mix in the market will reflect that; fewer desktops and laptops with more tablets, PDAs and cell phones.

Rick

Who knows, bloated inefficient software may become unacceptable.

Not bloody likely. The demand is for a new generation of phones every six months. That's hardly conducive to any rigorous development cycle and I don't see it changing.

In article , rickman writes: fore.

Servers can always use more CPUs per chip.

Servers are power limited too. That's nice since they don't have to design separate CPUs any more.

I see 2 interesting limits coming soon:

The width of a line will get below the diameter of an atom.

The cost of a fab line will exceed the GDP.

I'm not sure which one will happen first.

Moore's law is "doubling the number of transistors". There is still a third dimension, if you end up with physical limits on a flat die, you can start stacking transistors in two layers - this still would qualify as "doubling".

It certainly shows.

How could they tell? A Gerber file is a Gerber file is a Gerber file. There should be nothing in there to identify its origin. Same for Excellon drill files.

Nonsense.

Nonsense. Hand soldering of SMT components is routinely done every day. I dare say you couldn't spot the difference between one of my hand soldered assemblies and a reflow oven assembly.

Through-hole components were *designed* to be hand soldered (by lines of girls, sitting at benches, with feed guns).

Rubbish. I get quotes for runs of tens all the time. My usual order for small assemblies is "as many as will fit on a standard panel".

And that, folks, is what standards are for!