This article was posted on slashdot -
http://www.gotw.ca/publications/concurrency-ddj.htm

Are we really nearing what is the limit of microprocessor speeds?
There seems to be some truth in it, cause the past year we have been stuck with just going from 3.0 ghz to 3.6 ghz while the year before i guess we did from 900mhz to 2.4 ghz or something.
Posted on 2005-01-07 15:01:48 by clippy
It might be getting harder to push the GHz barrier upwards... but that doesn't mean we're reaching a speed limit, we should just throw away the x86 junk and use an architecture that has better design and instructions per cycle. Alas, AMD has sorta spoiled that by introducing AMD64 and thus extending the x86 lifetime even further :roll:
Posted on 2005-01-07 15:09:53 by f0dder
Yes, thats true.
But discovering such an architecture and making it mature enough to go commercial will take a lot of time.

Also the cost of shift from x86 to any other totally different architecture will be *immense* for the software industry. Far greater than the shift from 16bit-32 bit cause we have a whole *lot* more software now anda lot more systems dependent on that software.

Thats why no processor company would want to take such a huge risk of coming out with an entirely different architecture which would be incomatible as it would be almost bound to failure, as has already been the case with the Itanium processors.

Amd are doing the right thing as the company by making their 64 bit processor compatible with x86 architecture.
Posted on 2005-01-07 15:31:26 by clippy
Itanium could have been a success if...

- they were priced lower
- they weren't exclusively targetted for servers and very high-end workstations
- Intel hadn't screwed up with the hardware x86 emulation in the first itaniums

Imho, the hardware x86 emulation was a bad move. The die space could have been used better (more cache? :P), and instead they should have focused on good *software* emulation for legacy apps.

As long as you have an operating system, drivers, and a well-working software emulator, moving over shouldn't be a big problem. If the platform was made by Intel and supported by Microsoft, the large software vendors would port over their applications as fast as they could... and the software emulator should be fast enough to handle the legacy apps that aren't ported.

Moving to a *new* platform would also mean we could get rid of a lot of the other junk from the PC platform, not just the patchwork x86 processor.


Amd are doing the right thing as the company by making their 64 bit processor compatible with x86 architecture.

Nope, they're just extending the lifetime of this aged platform. I bet Intel wouldn't have made their 64bit extended P4 if AMD hadn't started, and AMD is probably the only company big enough except Intel to actually do this.
Posted on 2005-01-07 15:44:35 by f0dder
I think clock speed wise we have hit a barrier, but as f0dder said, we can improve performance of PC's immensely if the x86 was ditched.
Posted on 2005-01-07 16:14:24 by x86asm
There seems to be some truth in it, cause the past year we have been stuck with just going from 3.0 ghz to 3.6 ghz while the year before i guess we did from 900mhz to 2.4 ghz or something.

A 2.4 GHz Pentium 4 is not 2.7 times faster than a 900 MHz Pentium III. Actually my 1.4 GHz Pentium M can keep up with my brother's 2.4 GHz Pentium 4. So in terms of clock frequency there has been a huge jump 'thanks' to Intel's market strategy and unbelievable will to minimize the work done in one clock cycle. By moving too fast in that direction they've hit some limits (power consumption), which now slows them down.

If we look at AMD I believe its performance index has grown steadily for the last years and with dual-core processors I don't see this coming to an and any time soon. 65 nm technology is in the labs and they're starting to understand just as much about power consumption than about performance optimization so I expect some great things for the future.

It's true that, especially Intel, had some problems in the last year, but we had that with the 1 GHz barrier too, now it's the 4 GHz barrier. It's inevitable that the physical limit is getting closer, but seriously we won't hit a wall in just a year time. I expect the performance curve will get some bends but it's definitely not over for at least a few decennia.

Layered diamond chips anyone?
Posted on 2005-01-07 17:46:17 by C0D1F1ED
I think clock speed wise we have hit a barrier, but as f0dder said, we can improve performance of PC's immensely if the x86 was ditched.

It's ironic to say that on an x86 forum of course. :wink:

I don't think x86 really has to be ditched. Let's face it, in a way we all hate some aspects of it but it's undoubtly the most popular ISA ever. If anything significally better exists I would expect we've heard about it by now. I see little indication that there's any revolutionary ISA out there.

Let's take for example Intel's very own Itanium. It's more powerful than a Pentium 4, we know that. But what do you expect? It has gigantic caches, about three times more functional units or so, produces 150 Watts of heat or more, and is insanely expensive. Now imagine they scaled it down for the desktop market, with smaller caches, less execution units, etc. Would it still have that lead over our Pentium 4?

Ok so it might be a little faster, maybe 20%. But seriously, what's that if it would take longer to develop such chip than to optimize x86 processors to be 20% faster? x86 proves time and time again to be very extendable, and this is an extremely important property. In the embedded world some nice ISAs exist, but they wouldn't survive the desktop market for long.

Just like everybody hates Microsoft because everybody uses Windows, x86 is disliked because it's the first thing to bitch about. Yes it's not perfect but of all the alternatives it's really not that bad.

Last but not least, x86 is a CISC ISA, but since the Pentium Pro the processor core is RISC. With micro-instruction fusion and such there's even more abstraction between the ISA and the actual execution core. Transmeta even proved a VLIW core is possible. So I'd even dare to say the ISA doesn't matter all that much, as long as it's extendable.
Posted on 2005-01-07 18:11:43 by C0D1F1ED
I think clock speed wise we have hit a barrier, but as f0dder said, we can improve performance of PC's immensely if the x86 was ditched.

It's ironic to say that on an x86 forum of course. :wink:

I don't think x86 really has to be ditched. Let's face it, in a way we all hate some aspects of it but it's undoubtly the most popular ISA ever. If anything significally better exists I would expect we've heard about it by now. I see little indication that there's any revolutionary ISA out there.

Let's take for example Intel's very own Itanium. It's more powerful than a Pentium 4, we know that. But what do you expect? It has gigantic caches, about three times more functional units or so, produces 150 Watts of heat or more, and is insanely expensive. Now imagine they scaled it down for the desktop market, with smaller caches, less execution units, etc. Would it still have that lead over our Pentium 4?

Ok so it might be a little faster, maybe 20%. But seriously, what's that if it would take longer to develop such chip than to optimize x86 processors to be 20% faster? x86 proves time and time again to be very extendable, and this is an extremely important property. In the embedded world some nice ISAs exist, but they wouldn't survive the desktop market for long.

Just like everybody hates Microsoft because everybody uses Windows, x86 is disliked because it's the first thing to bitch about. Yes it's not perfect but of all the alternatives it's really not that bad.

Last but not least, x86 is a CISC ISA, but since the Pentium Pro the processor core is RISC. With micro-instruction fusion and such there's even more abstraction between the ISA and the actual execution core. Transmeta even proved a VLIW core is possible. So I'd even dare to say the ISA doesn't matter all that much, as long as it's extendable.


I do not dislike the x86 ISA. I am with you it is very good and has proven itself to be quite scalable, but for each generation we are adding hack after hack, I believe they can only do this for so long without running to performance barriers.
Posted on 2005-01-07 19:51:33 by x86asm
When I was doing hardware design we cursed the day the x86 was born as we designed our boxes using the beloved 68000 and Z8000 while drooling after National Semiconductors NS32000. Others were lucky enough to do work using the MIPS risc chips or even delved into "Transputers" from INMOS.

There were plenty of good architectures out there and still are.
Posted on 2005-01-08 14:19:52 by drhowarddrfine
i'd be tempted to say the graph in this article is BS, at least the lines he draws to "prove" the curve is flattening (and the dotted line that goes up).

that doesnt mean the rest is false and that doesnt mean the trend doesnt exist, of course, but i fucking hate when ppl try to do such things. basically, i feel i m being taken for a moron.
Posted on 2005-01-10 12:16:27 by HeLLoWorld
Finally :)

Would be nice to buy a computer and have a life expectancy of more than 3 years. I wouldn't mind it a bit if everything stalled for a decade or so.
Posted on 2005-01-10 12:40:41 by JimmyClif
Yes they did,

And the so called "better procesor" will not help much. Multiprocessing will not help much either.

Everybody that went on a computer University knows that there is a law that states that the advantages from multiprocessing are rapidly lost in inter comunications and twisting applications for this.... (basically greater than 4 CPU is a loose of time)

Only special crafted matrix operations from some scientifical and graphical imaging might benefit from again special crafted transputers layouts...but that is all and it is not working for general user level computing.

Such advanced procesors are preached by "elites" that have disregard for market rules when it suits them ... interesting how fast people forget about capitalismus and free market rules when they feel like they are inteligent :D . The x86 architecture was proven by the market... and by your own rules (that i do not like) the market always knows better. Again the AMD did the exact RIGHT thing (that Intel just forgot how to do) and again it won the markets.

Anyway RISC procesors with huge ammounts of registers will not help much today bloated compilers... I have worked with them and they are not even by far so elaborated as x86, in terms of cache management or pipelines and such.

having many registers will help a little because compiler have problems choosing registers... a few extra registers would help humman with x86 also ;)

But the real problem is the knowledge of the algorithm and you can not teach a compiler or a CPU that...

Because the essence of inteligence is making very very small steeps. Each time you make a bigger jump ... the Universe slaps you back almost instantly ;)

IF having an elite microprocessor mattered that much, then MACs should have ruled the market. Instead they have a fixed quota but not very big.

Personally i loved the Zilog CPU's but again the markets decided diffrently .

I might note that the x86 has a more human readable and humam useable assembly language that other microprocessors. other CPU's use a more compiler friendly and anti-human assembly language ... could this be a reason? (just joking here)
Posted on 2005-01-10 14:34:54 by BogdanOntanu
Bogdan, more than 4 CPUs might not be that advantagous for running a single applications (or rather, a single algorithm in one application), but it could surely benefit when you run a lot of processes (think server terms). And as for bus speed problems with lots of processors, I guess NUMA architecture will help.
Posted on 2005-01-10 14:51:24 by f0dder
I'd like to see a multi-processor hot plate to keep my tea hot. It should be possible with 4 CPU's and one large heatsink. A seat warmer would not be a bad idea either - wherever we what the heat at the time is where the most computations come from.

:lol:
Posted on 2005-01-10 21:07:55 by bitRAKE
I think clock speed wise we have hit a barrier, but as f0dder said, we can improve performance of PC's immensely if the x86 was ditched.

Should we form a company? I don't know but MS and INTEL are understimating the quality of the tools. :-D First we must make ms fall
then intel. Recruit as much linux users to form a company for hardware.
Posted on 2005-01-10 22:04:22 by Statix Star
Maybe this will eventually be the processor that reaches 10ghz.

I do wonder how far Steve Bush has gone ahead with it.
Posted on 2005-01-11 16:35:01 by clippy
Bwaahhaha ;)
Posted on 2005-01-11 17:49:30 by f0dder