The Year 2004 in Microprocessors

DeanMan writes "From spintronics to clockless CPUs, 2004 was a year of process and research in the microprocessor industry. As a way to transition into the new year, this article offers a month-by-month look at the highlights of the 2004 microprocessor timeline."

Comments

Clockless CPUs

[koreaman]

how does that work, someone enlighten me please.

Re:Clockless CPUs

[Wiser87]

There's an interesting article about it here.

Re:Clockless CPUs

[Spitfire75]

From TFA: http://www.geek.com/news/geeknews/2004Nov/bch20041 104027700.htm Asynchronous processors are capable of allowing each of their units to run independent of a global synchronizing clock, saving the power consumption--not to mention the design life cycle--of a complicated and usually power-hungry clock route scheme. The clock is increasingly the source of a large amount of power consumption, because of both the increasingly long relative wire length and the buffers (extra gates) required to repeat the signals in high-clock-speed devices. Obviously, the elegance of this low power design comes at a cost, in fact a barrier cost to high volume manufacturers. First of all, there is a great reliability issue for high-speed devices. No clock means potential race conditions and other performance/functional conflicts.

Re:Clockless CPUs

[tloh]

It implements the refinement of a 10 year old technology invented in Belgium.

Re:Clockless CPUs

[verus vorago]

asynchronous CPUs - different parts of the CPU go at their own speed and the results are co-ordinated in other ways (i.e. without lockstep execution across the whole chip).

This is much more complicated to design and to mass produce but the power savings may make it worthwhile.

Re:Clockless CPUs

[iamnotacrook]

The very first CPUs were designed without a clock. However the introduction of a clock made motherboard design far simpler, at the expense of flexibility on the chip itself. Now that technology has progressed forward as much as it has, the original clock-less type cpu design is more viable.

I for once am glad to be rid of our GHz-overlords.

FPGAs?

[paithuk]

No mention of FPGAs?

Re:FPGAs?

[stratjakt]

FPGAs have been around since the 70s, what's new about them?

So....

[FuturePastNow]

IBM is a news source now, eh?

Re:So....

[FuturePastNow]

True. Google shows a lot of other articles by him, in other places. But, while I like IBM, linking to stories like this sets a precedent I don't like. Most companies will flat-out lie to make themselves look better.

And- Kane? Did anyone else look at his name and picture and think of Command & Conquer?

The future of silicon chips

[paithuk]

Intel's plan all along has been to reduce the size of their pipeline stages in order to increase the possible clock rate. However, with the halt of the 4GHz processor, and their new found interest in multicore chips, it'll be interesting to see how they'll compare with FPGAs in the upcoming years since both offer what the other is looking for. Intel want to be parallel, and FPGAs want to be sequentially quicker. The only difference is that Intel has been researching how to be quicker for a lot longer than FPGAs have been around so the guys at Xilinx shouldn't have too much difficulty following Moore's Law, whereas Intel might have more difficulty expanding into multiple cores since their chips are already huge. Who will win out in the end? Will Intel start snatching up companies like Celoxica and Xilinx in the coming years?

Re:The future of silicon chips

[wafflemonger]

Intel's plan all along has been to reduce the size of their pipeline stages in order to increase the possible clock rate.
Unless this is a new plan the word reduce should be increase.

Re:The future of silicon chips

[CajunArson]

No, actually it should be reduce... each stage is smaller---> more overall stages. Witness Prescott.. it has ~10 more pipeline stages than Northwood did, each one is doing less (and hence can be clocked faster)
Intel is now moving away from this since the performance gains just aren't there and the power consumption is getting terrible (like they said in the clock-less posts... you must distribute clock to all those stages amongst other power-sucking things)

Summary

[af_robot]

Quick summary for slashdot's readers:
1.AMD rocks (generally good, til their CPU prices are lower Intel's and we can overclock cheap Athlons to save some $$$)

2. Intel sucks (Pentium IV = really bad, PentiumM = good but pricy, but we still hate intel today, because they are evil)

3. IBM rocks (good boys, cause they support Linux and can beat SCO's ass)

4. There are some companies in a world but we don't give a shit until Linux can run on their processors.

Re:Summary

[paithuk]

Intel doesn't suck. AMD wouldn't be here today if it wasn't for Intel, and probably nor would be the machine you're sitting at right now. Intel have some amazing guys working for them, and have hosted brilliant minds in the past. Linux may be open source and beautiful in some respects, but it hasn't done for operating systems what Intel have done for processors. Come on, be serious pal.

Advantages of clockless CPU

[nizo]

The space savings in the clockless CPUs is worth it, plus you don't have to keep winding them up all the time.

Be less optimistic about clockless design

[slashdot_nobody_nowh]

One should remember that clockless design
poses two huge difficulties:

1) verification (both logical and timing);
2) in-chip noise.

Clocking allows oscillations created
by generating edges to fade out before
the sampling edge.

In clockless designs signals change whenever they
want in a sense, so sampling may occur while
the noise (parasitic oscillations) is still high,
and wrong values will be stored/used.

Re:Be less optimistic about clockless design

[slashdot_nobody_nowh]

Look, I tried to be short in my post.

When I'm saying signals change more or less
whenever they want, I mean there is no
generating clock to relate their changes to
and thus to avoid interference created
by signal changes.

With modern frequencies connectors in chips
behave to big extent like transmission lines.

I worked as VLSI designer and STA (Static
Timing Analysis) methodology engineer,
so I hope I know what I'm talking about.

Re:Be less optimistic about clockless design

[bob beta]

I'm not disagreeing that it's a new area and there aren't tools for it at this point in time. Pure 'clockless' and 'not-all-totally-syncronous' are two different things, though.

An annoying example from my past was when a 'glitch' problem was found in a complex TTL circuit I designed, and the boss wouldn't let me do it right, i.e. back the clocking out another layer and syncronize something that shouldn't have been left hanging on the edge. The boss insisted I just throw a fricking cap across the signal to get rid of the glitch. Bleah! Sucky kludge!

Analog folks count in 'linear' and 'unsaturated signal' issues all the time. Granted, most logic and digital circuit designers aren't prepared to work at that level...

Wish they had these 5 years ago.

[k4_pacific]

Just think how much trouble a clockless CPU would have saved leading up to Y2K.

Re:Wish they had these 5 years ago.

[forceflow2]

Yeah, it probably would have kept those thousands of planes from falling from the skies and our toasters from rebelling.

Unlike IBM's previous microprocessor history...

[AtariDatacenter]

Remember IBM's microprocessor history that was posted to Slashdot a week or so back? I have to say, this one is far more even handed to the competition. Quite a lot of mentions of SPARC for the first time.

But this one line cracks me up...

American Technology Research predicts that Sun® and IBM® are well positioned to capture the 64-bit desktop market since both use the Opteron processor as an integral part of upcoming product lines and both have initiated flexible CPU roadmaps.

Sun? IBM? Capture the desktop market? My, these folks at American Technology Research much be geniuses! Or is that genusi?

FWIW, Sun has been doing 64 bit computing for quite some time now with the 64 bit SPARC chips it has been putting out for ages. But Sun Microsystems and IBM, masters of the 64 bit desktop? Oh boy.

IBM, marketing department, line 1...

[AtariDatacenter]

"IBM debuts Cell processors, designed to be used in workstations, Sony PlayStations gaming consoles, and in Toshiba televisions. Programming the processor is said to be relatively easy."

How much did they have to couch it? "Relatively easy?" "Said to be..." ?

Translation for the technical crowd:
"Programming a cell processor is hard."

Why aren't OPTICAL-Electronic chips being made ??

[zymano]

A hybrid might ease us into all optical chips.

Why is everyone dropping this field ? Quantum is way off in the distance and so is spintronic.

Using optical buses would reduce wiring complexity too.

Re:They forgot

[Anonymous Coward]

You'll all not be laughing come 2006 when we're all playing Duke Nukem Forever on our Linux desktops we bought from Dell!

Incorrect information

[raptor21]

The article claims that Sun is outsourcing Niagara, which is a 65Nm process to Fujitsu. This is absolutely false. Niagara is to debut in 2005-2006 according to Sun and on 90Nm technology not 2007.

http://blogs.sun.com/roller/page/jonathan/200409 10

Since the chip is already in the Sun labs how can it be 65Nm? No fab, in my knowledge, is ready for 65Nm yet,

http://aceshardware.com/read.jsp?id=65000293

Also sun never claimed to outsource all chip manufacturing to Fujitsu. The article is based on blurbs from unreliable sources, example geek.net.

This is the second IBM article to calim that Sun is outsourcing all chip desgin and manufacturing to fujitsu. Is this some sort of FUD IBM is trying to spread?

2004: Rest of desktop computer gets faster.

[MtViewGuy]

And that is the real story of desktop computer technology in 2004.

It's no longer how fast you can crank up the CPU speed, it's now how fast the rest of the system runs. Look at what we have now on desktop machines:

1. The development of faster motherboard interconnects with improved chipsets and things like HyperTransport and its competitors.

2. The wide availability of PC3200 (DDR-400) DDR-SDRAM system RAM, with even faster RAM coming over the next 18-24 months.

3. The development of AGP 8x and new PCI Express connections for graphics cards with 3-D processing ability that would be the domain of ultra-expensive workstations only a few years ago.

4. The development of ATA-100/133 IDE, Serial ATA and soon Serial ATA-II IDE, and UltraSCSI 160/320 interfaces and 10,000+ RPM drives with 8 to 16 MB on-drive memory caches for very fast hard disk access. Even optical disk drives are benefiting from these faster interfaces.

5. The very wide availability of 100Base-T Ethernet connections on most motherboards, plus some motherboards now sport 1000Base-T Gigabit Ethernet connections.

6. The near-universal availability of USB 2.0 connections and increasing use of IEEE-1394 connections to external devices, which makes the use of external disk drives to back up data and connect to digital camcorders possible.

All of these developments have resulted in vastly faster computers in terms of overall speed even if you don't have the fastest CPU installed on the motherboard.