Will Intel or AMD sell a cpu faster than 4GHz in the next five years?

[ET_]

I saw water-cooled processors for PCs running well over 4 GHz last year...

Luigi

[cheesehead]

Originally Posted by axn

Five years is waaay too conservative. Should be two years (three at the most) before 4 GHz is done-and-dusted, most likely by /both/ Intel & AMD.

Lessee ... "01 Apr 07" plus three-at-the-most-years is ... next Thursday.

By that time, both the core count as well as GFLOPS rating/core would have also tripled.

Man, can I predict them or what!

I love youthful optimism!

- - -

And, to have the energy of your avatar, axn ... !

[CADavis]

Well I missed the boat on this one... but starting from NOW my answer is no, if Turbo Boost doesn't count. If Turbo Boost counts, then my answer is Yes.

[Uncwilly]

Originally Posted by Historian

For 99% of people, there's little to no difference between 4 and 6 cores because most programs aren't multi-threaded. Because of that, it makes more sense to increase clock speed than to increase the number of cores, but the main question is when Intel and AMD will finally realize that.

But having your virus checker running 24/7 in the background, the OS, and various other programs all at once (IE and a media player and whatever else), the number of cores will be noticed. Everything will run faster and smoother.

[jasonp]

Originally Posted by Historian

it makes more sense to increase clock speed than to increase the number of cores, but the main question is when Intel and AMD will finally realize that.

Is it your position that Intel and AMD could give us a 5GHz processor now, or at least soon, but because of their naivete actually choose not to?

Robert Colwell was the chief architect of the P6 design team at Intel, and in his book 'The Pentium Chronicles' writes, about his leaving Intel after helping design the Willamette,

I felt that as one of the principals who had led the company to a high-
clock-rate x86 strategy, I should have been able to lead it away from that
strategy when it became necessary (and we knew from the beginning that
eventually it would). But it seemed to me that that time came around 1998,
and over the next two years I was unable even to make a dent in the
product road maps
[...]
Beyond all of that, however, was a looming thermal power wall that was
no longer off in the distance, as in P6, but instead was casting its long, ugly
shadow directly over everything we did. That experience was primarily why 
I was so sure I did not want to work on any high-clock-rate chips beyond
Willamette. I just did not think that there would be enough end-user
performance payoff to justify the nightmarish complexity incurred in a 
high-performance, power-dominated design.

See also the Forrest Curve

[R.D. Silverman]

Originally Posted by Historian

The 3.866 GHz speed is only available with turbo boost, but I'll assume that counts because nothing in this thread says that a 4GHz computer has to have that speed sustained when all cores are used. However, the Wikipedia article doesn't mention processors coming out after Q3 this year, so the one and a half year period between then and April 2012 will be filled with wild guesses and speculation. Here are my thoughts:

?

Go read my paper from IEEE Computer: Exposing the Mythical MIPS Year.

Note that my I7 processor as 2.4GHz is faster per core than my 3.4GHz
older P IV.

Merely looking at clock rate and assuming that faster rates
means faster processors is just plain stupid.

[Uncwilly]

Originally Posted by R.D. Silverman

Note that my I7 processor as 2.4GHz is faster per core than my 3.4GHz older P IV.

That is i7 not I7.

[cheesehead]

Originally Posted by jasonp

Is it your position that Intel and AMD could give us a 5GHz processor now, or at least soon, but because of their naivete actually choose not to?

Speaking for myself (Historian may differ):

" naivete "? No ... and No to the first part, too.

Beyond all of that, however, was a looming thermal power wall

That's a real problem, not naivete.

See also the Forrest Curve

... and that's another non-naive problem.

The reasons for being stuck at sub-5GHz nowadays are real (but won't last forever).

[Historian]

But having your virus checker running 24/7 in the background, the OS, and various other programs all at once (IE and a media player and whatever else), the number of cores will be noticed. Everything will run faster and smoother.

Are 2 cores better than one? Yes.
Are 4 cores better than 2? Yes.
Are 6 cores better than 4? Probably not, for most users.
Are 8 cores better than 6? No, unless you're running a server.

For most people, the law of diminishing returns starts to kick in above 4 cores. Think about it, if it were not for DC projects, how often do you actually need to use more than 4 cores? It's pretty rare for people to have more than 4 programs running at once.

Originally Posted by jasonp

Is it your position that Intel and AMD could give us a 5GHz processor now, or at least soon, but because of their naivete actually choose not to?

My position is that Intel and AMD cannot give us a 5 GHz processor right now or in the near future even if they want to. 4GHz? It depends*. But if Intel can, it won't release a 4GHz processor not because they're naive, but because of business reasons. Their main competitor, AMD, isn't a big threat yet, so there's no need to release a faster processor at the moment - you'd basically be competing against your own products at the high end.

*Yes if they continue developing dual and quad cores, no if they're trying to increase the number of cores beyond 6 cores instead of bumping up clock speed.

[jasonp]

The crowd that visits here is in the odd position of belonging on the horizontal axis of the Forrest curve; give everyone on Mersenneforum a 32-core processor and they'll all run Prime95, or some other computationally demanding application, 31 times. Likewise, give me a processor with lots of cores and I'll find something to do with it.

The Forrest curve post appears at regular intervals in the usenet archives, back into the late 1990s, and your concern that real-world performance eventually won't improve has always followed it. If you literally only use email, browse the web and use office applications then you haven't needed to buy a new computer for the last decade (my wife's computer dates from 2001), and the number of cores you 'need' is irrelevant, it just has to be nonzero.

On the other hand, if you do have demanding computational needs, and rely on commercial software to address those needs, then you should indeed hope that the people who develop that software can incorporate multithreaded or parallel primitives into updates or new products, or at least use standard libraries (BLAS, FFTW, MKL, Accelerate, Intel's compiler tools) that do this already. If they do not, and the incremental single-thread performance gain that the next generation of processors provides is not enough for you, that's a business opportunity and not a liability.

It has always been true that software which is computationally demanding cannot be written once and then expected to accelerate forever. Creating demand for innovative software products is something the software industry as a whole should welcome; why are they getting paid otherwise?

Of much more interest to me are other issues: suppose that the amount of compute power and memory you have to solve your problems is suddenly much larger than you are used to. How does that change the way you solve the old problems? Also, what sort of currently impossible solutions become feasible, opening up new application domains?

Filesystems that never delete anything? 3-D raytraced home movies that are rendered in the background? There was a lot of hand-wringing about how the internet would make all these desktop computers obsolete because they are just platforms for a browser most of the time. Is that really true? Nobody can find anything to do with all this storage and low latency computation that couldn't be done through a web interface?

[Mini-Geek]

Originally Posted by jasonp

give everyone on Mersenneforum a 32-core processor and they'll all run Prime95, or some other computationally demanding application, 31 times.

Only using 31 of 32 cores constantly...
What a bizarre concept...