Connect with us

Where's cpu makers going ?

Discussion in 'Electronic Design' started by [email protected], Dec 15, 2006.

Scroll to continue with content
  1. Guest

    They shrink the die, packed more and more transistors, going from 130
    nm, 90 nm, then 65 nm... technologies, sound very very good, lots of
    improvements !

    Yes, then what ? In order for the CPU to work properly they add an
    1-cubic feet :)) of the metal heat sink and a noisy fan to cool the cpu
    down !
  2. Guest

    If you're talking about x86, then yes. If you're talking about CPUs in
    general, no. Not all CPUs are like that. That's only in the
    MHz-competitive PC world. On the RISC side ARMs are quite impressive
    especially if you look outside the Xscale family. Some of the low end
    ARMs are smaller than what PICs used to be. If you look at
    microcontrollers then it's more impressive. People are starting to
    package micros in small 3 or 4 legged ICs the size of an SMD resistor.
  3. Guest

    That's what we're getting at. The point is they don't need to shrink
    it no more, just improve the power/thermal issues, that will beat the
    competitors !!!
  4. I've come from 8K core store, 1 M disc drives, diode microcode matrix
    boards, the 4004 etc. 1 micron technology developed in Rutherford labs...
    I see fully integrated microcontrollers, as in ram rom I/O in high density
    being the norm.And I don't mean 1K :) Then bond it to an LCD panel.....
  5. Guest

    ehhh no,
    what drives the power consumption up is speed and transistors, not
    shinking should make it use less power, but it also means its capable
    of faster
    speed and room for more transistors. So everytime something is gained
    with a
    shink it is lost in running it faster and with more transistors to get
    more perfomance.

    shinking yesterdays cpu and doing nothing else just makes it smaller
    and cheaper
    to produce ..

  6. Guest

    Agree, but what's the advantage when the heatsink + fan occupies a
    volume of thousands times bigger than the cpu itself ?

    If there exist a CPU just few hundreds time bigger than let say the C2D
    for example, having same performance with no extra heat sink or aided
    cooler, can you tell who's winner ?
  7. Look at the last few models of processors. Each has used _less_ power than
    the ones before. What then happens, is that the speed is then ramped up as
    far as they can go within the bounds of the existing cooling. The cooling
    requirements of current CPU's, are almost identical with those of seven or
    eight years ago.
    If you look at units like the AMD64, the slower models of these, can run
    without fans at all.
    The smaller dies make the chips cooler, but then 'marketting' pushed the
    speed up as far as it can go within the bounds of the acceptable cooling.
    A lot of the decision is yours. Take a standard CPU, and _underclock_ it
    to about 1/2 to 3/4 of it's rated performance, and done properly, you can
    turn off the cooling fan. You get performance that is much faster than
    anything available five years ago, and a tiny fraction of the power
    consumption as well. There are several specialist suppliers doing exactly

    Best Wishes
  8. Guest

    Look at the last few models of processors. Each has used _less_ power than
    I would like to have such thing, the CPU can run same speed same
    calculation power as the current cpus with no heatsink + noisy fan
    We can see they "push" the performance and "pull" out the thermal, in
    this case the desktop industry "might" bite the bullet
    What about the heatsink ? After all, we have no improvement? :)))
    Ooop, yes indeed they do have liltle bitty gain to five years ago

    Thanks for sharing your thoughs
  9. Gareth

    Gareth Guest

    I imagine that a bent bit of metal is a lot cheaper than a similar sized
    area of processor silicon, so making the actual processor smaller makes
    it cheaper.

    For the likes of Intel and AMD, the main reason for making the chip
    smaller is to make it run faster. The signals propagate across the
    processor at a finite speed, so when you are clocking your processor at
    a few GHz the time taken for the clock signal to propagate across the
    chip is actually a significant fraction of the clock cycle. Also,
    making the thing physically smaller will reduce parasitic inductance and
    capacitance, which again will allow it to run faster.

    For the PC market, the physical size of the processor isn't really an
    issue since most of the PC case is empty space anyway. If you want
    physically small and/or low power processors they do exist, though
    obviously they don't have the raw power of Intel's latest devices.

    This is one of the smallest I know of.

    If you want more processing power without the need for a heatsink have a
    look at these:

  10. Gareth

    Gareth Guest

    Just a few hundred times bigger would probably mean a lot more expensive
    to make

  11. Why are you so anti the heatsink?. Even the smallest chips have these, and
    need them. They are cheap, and small.
    Again, why so worried about the heatsink?. I have three systems here based
    on this sort of technology. Look at industrial embedded PC's. The
    heatsink, in most examples is the case. Less than half a dozen cubic
    inches of aluminium. Even a chip dissipating a couple of watts, needs a
    heatsink of some sort. It may simply be the PCB itself, but the heat has
    to be got away from the die. Making the die bigger, makes this problem
    worse, not better, and makes the chip vastly more expensive (much more
    than linearly to the area, since the probability of making a good chip,
    falls). Try buying a large die chip, like a 10cm^2 CCD element, and you
    will understand why this is not done...
    Best Wishes
  12. colin

    colin Guest

    Every instruction takes a certain amount of energy,
    although the energy per instruction might be falling they are doing more
    instructions per second
    with a lot of instructions per second you need a big heatsink, no matter how
    big or small the cpu itself.

    The same processor can be run at a lower voltage and speed and not realy
    need much of a heatsink.

    The problem is software that is going backawrds, getting larger and larger,
    needing more and more instructions to do much the same thing.

    Colin =^.^=
  13. David Brown

    David Brown Guest

    That's true for dynamic power, used when switching the transistors. But
    static power is depedant on the leakage, and that goes up with smaller
    transistors. Depending on the type of device and application, static
    power can be more relevant than dynamic power.
  14. Guest

    The advantage of the heatsink + fan is speed. You can have a very
    compact system without a single heatsink if you're willing to run at
    200MHz - look at handphones & pocket PCs.
    A bigger CPU running at the speed of the Core2 Duo would actually need
    an even bigger heatsink. Core2s are extremely cool in comparison to
    other similar performing CPUs like the P4s. That's the effect of a lot
    of things, process shrinkage is just one of them.

    To reduce heat you have to reduce MHz. But marketing isn't willing to
    and newer games keeps insisting on using up more. There are
    architectures out there which kicks ass at low MHz - the PowerPC is
    one, Niagra is another one and of course the ARM family. They thrive in
    markets which requires high performance but where the consumer isn't
    told or is generally ins't aware about the processor's MHz - markets
    like high speed DSL modems, smartphones, set-top boxes etc. But they
    fail utterly in the follow-the-herd PC market where consumers usually
    opt for higher MHz even if the competitor can perform the same or
    sometimes better with less "MHz".

    Engineers have lots of solutions to the heat problem. But not much for
    the x86 because marketing and the consumer isn't willing to give up on
    the MHz myth (Intel is trying to but consumers still isn't buying the
    idea. Besides, benchmarks are often confusing, it's easier to compare
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day