math junkie thread(how long until 20M?)

[jasong]

I actually don't have the education to do this, IMO, but assuming Moore's Law is true(in this case processor power doubling every 18 months) and assuming no users jump on or off the project but their power increases according to Moore's law, when will we reach n=20,000,000? To simplify things, you can make the rather unappealing assumption that we won't find anymore primes(or you can pick someone's prediction about prime finding and apply that).

I would make a go at this myself, assuming 1.5n=2 time, but I don't even know our rate right now(I'm not sure how cEM's work)

[Jwb52z]

If you ask me, I think you would also have to figure in the socioeconomic status of the people involved in giving their spare cycles to the project. That is a rather large component of how the actual computing power devoted to the project would be affected. Moore's Law is about the actual possible technological speed advancement in computer processing technology as a whole in the industry. It's not about everyone's individual computer every 18 months. If we want to have an actual picture or answer, we would have to somehow get data on every user's finances and how often they historically buy newer and better computers and take surveys on how often they plan to do so in the future including how often they could do it based on their expenses and income.

[pixl97]

Moores law never said computers would double in speed every 18 months, just the number of transistors would double. If you havent noticed, for the past year or more, CPU's have not been getting faster. Mighty Intel has ran into a current process limit just below 4GHz, and speed increses have not occured for a while now. AMD's higest offerings have been around 3GHz. (arguing IPC is another question).

Transistor size is still doubling though, just look at the Athlon64X2 or Intels new dual core offering. If we are suck for another year or two at the 4GHz limit, only adding extra cores, it will be quite abit longer then most have imagined to reach the 20M mark. There are some that are suprized that we dont have 5GHz desktops currently. This should be the wakeup call to most distributed computing projects though. That you should make your client program processor aware and execute a copy on each core. Future speed gains in projects like this will come from adding more clients, adding more processing cores, or algorithmic changes, Not from faster processors.

[vjs]

I was at a conference on semi-conductors a couple years back were the Keynote address bascially discussed this topic from a purely fundamental aspect.

There were several points I've taken with me over the past couple years.

1. Computers are basically switches, where gates "gaps" are either open or closed.

- closed (on) basically means electrons can flow through open (off) electrons shouldn't flow through.

2. Due to noise how can we tell electron are flowing through a closed switch.

Conc. 3 electrons should pass through the switch inorder to say it was closed.

Problem: as electrons flow through the switch they create heat due to resistance.

3. Electrons travel accross the gap close to speed of light --> how small can we make the switch (shortest distance)

Problem: As the switches get smaller electrons will pass between the switches not the contacts. Inaddition as they get smaller more heat smaller size increased temperature --> increased resistance (terrible cycle).

I could go on, what everything boils down to is a maximum switching frequency and the removal of heat.

If I recall correctly, the conclusion is it was not impossible to have switching frequencies (This is switching frequency not raw MHZ) within computers of 20GHz (Modern tech).

Add to this tech advances such as ddr, qdr, multipliers, dual core. I don't see a reason why we shouldn't expect to have quad core 10ghz per core machines the norm within the next decade.

The current problems are more; cost associated with production, heat removal/power consumption.

And of course the biggest problem is lack of competition.
- release CPU's 100 mhz faster than the previous every other month.
- The thought that the average individual doesn't need a 10 Ghz computer, or more than 64k of memory.

Edit --> Since I'm ranting dreaming etc --> I think the next major advances will be in heat removal from the die. If you actually look at the thermal output per cm^2 cpu's make great heaters.

My prediction, attached layer/upper most layer of die will have some sort of heat pipe or column. Bucky tubes look promissing, or perhaps heat pipes within the core itself.... also why are we not using liquid gallium or mercury (O.K. the mercury is obvious).

[jasong]

Referring to vjs's post about heat:

Does anyone think Intel's announcement about transistors that use significantly less power to do the same amount of work will affect the increase in GHz? In other words, do you think Intel is holding back right now because they don't want people complaining about cooked processors?

[Matt]

I was reading in the Tom's Hardware mother of all cpu charts that apparently AMDs more recent processors are massively more efficient in terms of power consumption (and therefore heat production) than previous AMDs and Intels.

[vjs]

Does anyone know if they are actually using strained substrates yet???