CPU Wars

msolnik writes: "Whether you say "0.13-micron" as most of us do, or "130-nanometer" as PR flacks prefer, the phrase is weighing heavily on both Intel's and AMD's minds. Indeed, each company's timeline in reaching that mark may determine who calls the CPU shots in 2002. Read more here at Hardware Central." Other submitters noted that AMD and Motorola have both updated their development roadmaps.

Comments

Intel 4004 anno 1971

[Hougaard]

The 4004 consists of 2,300 transistors based on 10 micron technology fitting on a 12mm2 area.The microprocessor has 46 instructions. The 4040 is an enhanced version of the 4004, adding 14 instructions, larger stack (8 levels) and 8K program space. It can address 640 bytes. Documentation is written by Adam Osborne. The chip is introduced to the public in Las Vegas by Wayne Pickette. The sales price will be US$ 200 per piece.

This was the news of 1971

Re:Intel 4004 anno 1971

[Hougaard]

76.923076923, but the interesting thing is, that its both horizontal and vertical so its actually 5917.1597633 times better on the same square-inch.

Re:Intel 4004 anno 1971

[renehollan]

stud9920-speak:

Your error: "#@$@#$ @#$@#!4 fuc@#@# years!!!"

My error: "oops... sorry"

/me can't help but laugh. (Oh hell, my karma's so high, I could stand a few -2, pedantic).

Mac Hype

[BiggyP]

bloody hell, they really are hyping the G5, and they haven't got any confirmation of what technologies it will use, they simply assume that motorola's latest chip will be the basis, how much would you have to pay for a mac for them to make returns on their production process?

Re:Mac Hype

[Spruitje]

bloody hell, they really are hyping the G5, and they haven't got any confirmation of what technologies it will use, they simply assume that motorola's latest chip will be the basis, how much would you have to pay for a mac for them to make returns on their production process?


Well, actually.. A G5 powermac will cost almost the same as the current G4 machines.
The PPC8500 is a 64 bits processor which is 100% backwards compatible.
I've seen some preliminary SpecFP and SpecINT figures and if those are correct a PPC8500 running at 1,6 Ghz is equal to a P4 running at 3 Ghz.
It is twice as fast as a Itanium running at 80 Mhz. and uses only 15 watts peak.
Compare that with around 60 watts for a P4 running at 2 Ghz..
The difference with this chip is, that most of design work was done by Apple itself.
This chip uses 0.13 micron technology and SOI.
So ,actually the dye size is almost equal to that of the G4 which means that it will cost almost the same.
Don't forget that the PowerPC chip is based on the Power architecture of IBM.

Nanometers ahoy!

[Visoblast]

Before we get to 0.09 microns, lets start using nanometers to get rid of those preceding decimal places. Plus, unlike micron, a nanometer is an accepted SI unit (see http://physics.nist.gov/cuu/Units/prefixes.html). Strange the PR people should use it first -- could this be a sign of the Apocalypse?

Next year

[The Gardener]

Next year looks like the best time ever to buy a new performance PC.

Well, duh. Just exactly like every year since they were invented.. And just like every computer magazine pundit has said since day one

The Gardener

Re:Next year

[blair1q]

And have they ever been wrong?

--Blair

Smaller die == less heat?

[tnak]

Ok, I admit it. I'm confused. I thought a smaller die size increased heat. Less surface area to radiate from.

Gotta love the last line:

Next year looks like the best time ever to buy a new performance PC.

Next year is always the best time to buy a new PC.

Re:Smaller die == less heat?

[NerveGas]

Ok, I admit it. I'm confused. I thought a smaller die size increased heat. Less surface area to radiate fromm

You're confusing temperature with heat. : )

Both points of view are actually right. In the ideal world, having smaller transistors lets them operate at a lower voltage, so you use less power, and generate less heat. And if the die is smaller because of a decrease in transistors, you still use less power.

In the real world, though, you don't see Intel shrinking their dies and then leaving them at 650 MHz. When they shrink the manufacturing process, they also increase the frequency, offsetting any decrease in power usage. And, over the long run, they also increase the NUMBER of transistors, making it use even more power. So while the textbooks say that the new chips will use less power, they're likely to use MORE power, especially when they've had time to ramp them up to the higher clock speeds.

steve

Re:Half a mil?

[Anonymous Coward]

A mil is approximately 592 meters.

You get that dividing a "mile" (1609 m) by "e" (2.7183).

Re:Fast CPUs might be bad.

[GroovBird]

Well IMHO you're not really correct. All this is relative to what you're used to work with.

You're only naming two games, both using the same engine, that are now approx. 5 years old. These days all games are trying to be as immersive as possible, using 3D graphics and sound, enhanced with special FX, and playing against an army of bots trying to mimic our behaviour. They are already using dedicated coprocessors (called GPU's these days).

GUI's have evolved from crappy crammed black and white boxes with hourglasses to 24-bit 1280x1024 alpha-blending anti-aliasing semi-intelligent "interfaces". This all takes memory, memory bandwidth and CPU cycles.

I find myself amazed, even as a software developer, that these days I can take pictures with my digital camera and send them to my mom using e-mail. I predicted this could be done a long time ago. But now I'm doing it I have to stop at moments and find myself simply stunned by the world we live in. We're ordering pizza's from our PCs using broadband network connections. My audio software (Propellerhead's Reason) can emulate a jampackked rack of synths and samplers, and the sound is generated in realtime. I don't have a digital camcorder, but if I owned one I'd spent my nights making my own movies. Picture this 10 years ago.

If you think OO is what makes softwar bloatware then you don't understand OO, in my opinion. OO is one of the ways to achieve true code reuse, which is what we're all striving to do because we are all lazy asses. Code reuse means you get a lot more done in less time, and if done right it should take less space all at the same time.

What really makes software 'bloatware' is the addition of functionality beyond what is needed by the majority of the users. But then again the markets have widened and software has become one of the biggest business in the world today. More users want to find software useful and software vendors respond with more and more features which will always sound like bloatware in the eyes of a few geeks who like to hack together their own kernel and run it on your average pocket 'PC'.

Sure games were fun 20 years ago just as they are fun today. I like to play tetris myself a lot of times but if you really think about it, same now as back then, only 5% of all games are classics and 95% are crap. We're all just spoiled now and the only reason we'll play pong is because it makes us feel nostalgic.

In 10 years you'll say that you don't need the latest AMD XP 22000+ (16Ghz nominal) with 512GB of battery-backed-RAM and a semi-optical harddisc of 600TB ... but then again you'll always be saying this.

I say, keep 'm coming.

Dave

Personally I'd never go back to the days where i had to wait

Re:Fast CPUs might be bad.

[Weh]

for every more powerful cpu to come out there's people saying we don't need it. I wonder how long it has been going on? Windows 3.11 was nice on a 486 66Mhz but I'm sure happy that chip development didn't stop there.

I think it would be interesting to see the effect of CPU power on software pricing. With faster CPUs software might be less optimised thus costing less programmer time. It's just a thought...

My Packard Bell P75

[Knunov]

I bought my first computer in 1995. It was a Packard Bell P75.

Go ahead. Laugh. If you told me you actually paid money for a PB, I'd laugh, too.

PB actually used good motherboards in their systems. It was the components that sucked.

Anyway, to this day, I *still* have and use my PB computer. Yes, it went from a P75 -> P133 -> P200 MMX, and went from 8MB -> 32MB -> 64MB -> 128MB and the hard drive went from 1GB -> 4GB -> 20GB, but it's stll in use.

Admittedly, I've bought other computers since and I no longer use it as my main machine, but I *could* if I wanted to. I only bought faster machines because I wanted to, not because I needed to.

It runs Win98 like a charm and runs Linux even better. It has always been stable and still is, 6 years later.

If people would cater to their needs instead of their wants, the CPU industry would either wither, or they would start offering REAL improvements. These 100MHz increases are BS.

They need to start with a minimum 1GHz jump and better internal architecture. Everything else is just them going wallet fishing.

Knunov

No way, I'm supersticious.

[snatchitup]

No friggen way I'll ever own a 13 nanoM chip. I'm just too supersticious. I've got enough to worry about with my data, and (jpg)'s to trust them to an unlucky number. It's worse than a hat on the bed!

They should switch to Angstroms.

Oh wait a minute, my calculator tells me that 0.13 Microns equals 666 Angstroms. Holy Ess, The end is Nigh.

CPU speed Nuts...

[oconnorcjo]

I just don't get the desire for machines faster than 600 mghz. The CPU is going at least twice as fast as any other component on a PC machine. What I did recently was buy a DDR motherboard to get ram that ran at 133 mghz (advertised at 266 mghz) and so I got a AMD 1.4 gigahertz cpu with it. One of the nice features of the motherboard was the ability to change the clock rate of cpu and bus. I LOWERED the clock rate of my CPU to 800 mghz and my machine is as responsive as I would ever want it to be. When I hear that Intel is charging twice the price for thier 2.0 gigahertz CPU as thier 1.8 ghz and people go out in droves to buy the 2 ghz anyway boggles me! Most of them don't need the speed of either CPU AND people are willing to pay 100% more money for a measly 10% performance boost. Ten years ago, Most PC's came with a "turbo button" on the case with the idea that only when you really had to use the the cycles would you press turbo and the CPU would go twice as fast. Back then, the button was pointless because when computers were going at 66 mgz, processors would regularly be very busy. But today the Turbo Button would actually be a nice feature. When doing word-processing or surfing the web, have the machine go slow but then when playing quake 18 (Revenge of the killer CPU), press the turbo button so the bloatware can look sweet. However, for people who REALLY NEED more power (all of the time) *couph* *couph*... SMP looks to be the far better alternative than these monster single cpu solutions.

Re:CPU speed Nuts...

[ivan256]

Let me guess, you slowed the chip down to 800 Mhz just so you could tell everyone that you don't need such a fast machine, and that they dont need it either. You're just like one of those obnoxious people that feels the need to tell everybody that they don't have a TV, like that makes them better then you or something.

If you have the speed you might as well use it.

Also you may have lowered the stability of your machine by slowing it down that much. Certain parts of the logic need to 'refresh' to maintain their state, and when the designer assumes that the minimum speed a CPU will be sold at is 1.something Ghz, they might not make sure the charge sticks around long enough to work at less then half of the intended clock speed. But you're so smart...

Re:CPU speed Nuts...

[oconnorcjo]

Also you may have lowered the stability of your machine by slowing it down that much. Certain parts of the logic need to 'refresh' to maintain their state, and when the designer assumes that the minimum speed a CPU will be sold at is 1.something Ghz, they might not make sure the charge sticks around long enough to work at less then half of the intended clock speed.

Actually, I underclocked my machine because I was trying to insure stability (not that I had any anyway but then it never hurt). I am told that a lot of servers are underclocked for the same purpose. I also had the idea that my cpu would run cooler thus reducing any chances of overheating and at the same time saving electricity since I keep my main machine on 24/7. As for instability, I have had none of that and I would not expect to. That is usually an OVER CLOCKER problem (which I believe AMD and Intel do which is indicated by the need for huge heat sinks and fans).

Re:CPU speed Nuts...

[ivan256]

Underclocking may increase stability in small doses. 10% is probably fine, but I have worked with CPUs in the past that don't work as documented when underclocked more then 25%. Don't trust your data with an asumption. Manufacturers publish specs for a reason.

war

[VEGETA_GT]

Well to be honest, I have been watching this sine the first Athlons came out and proved that Pentium was not all that. So far AMD has been able to bask down Intel at every turn. The 2 ghz p4 is still slower then amd's athlon 1900 (1.6 GHz). Its not the speed, its what you can do per clock cycle, and amd chips dose do a lot more.

But reading the article, I find that the again go after the ghz number,

Of these, the P4 Northwood could be the most compelling CPU release of 2002

Their reasoning, the p4 well be unto the 4 GHz barrier in a few months. The Athlon is planning to make some jumps as well which, makes this sounds to me like the article is written by someone leaning towards the users who love big GHz numbers and not real speed.

What makes this even funnier is the fact that most users could buy a 1 GHz and still play the latest games and the other things in 2 or 3 years.

my 2 cents plus 2 more

CPU is not problem anymore

[El_Nofx]

Unless you are ripping Divx movies left and right or a Seti@home freak you don't need a faster cpu, It will do nothing for you. Anyone notice that you pretty much have the same Harddrive as you did with your pentium 1 120, the size has increased but if you go IDE it is still 7200rpm and the data transfer rate isn't any faster.
It is funny, Xp Pro runs the exact same on my PII 400 with 384 meg of ram as it does on my PIII dual 1 gig with a gig of ram machine. The 400 actually boots faster!. So what does processor speed to for you in every day apps? everyone here knows exactly what i am saying. I am just complaining becaue we always hear about the new processor that is supposed to be so great that is coming out next year or whatever. WHEN AM I GOING TO SEE A SOLID STATE HARD DRIVE? Sure Serial ATA is coming up but the transfer rate on that is only starting at 166MB/s. ok. show me a harddrive that actually needs anything better than ata 100 first.
The bottleneck in every modern computer is still the hd, and the bus, we should fix those first and then jack up the mhz..

Re:CPU is not problem anymore

[NerveGas]

Unless you are ripping Divx movies left and right or a Seti@home freak you don't need a faster cpu

Actually, these MHz wars benefit me in a very nice way. I'm still using a PIII-650 at home, but my servers have much more substantial hardware - dual CPU's are the *smallest* machines in the stack. And these MHz wars have made desktop machines that can best high-end servers of only two years ago.

Two years ago, I spent $4,000 on a chassis and motherboard that would use quad Xeons. Add in $2800 for the processors, and that's a lot of money. Today, I can spend $200 on a dual Athlon motherboard and $500 on two chips, and have a machine that will either rival the quad Xeon or beat it in almost any situation.

I remember when I was amazed that opening up an 800x600 JPEG took less than three seconds on a new machine. The funny thing is, at the time, I didn't mind waiting three seconds, and really hadn't even noticed that it was a wait. But once I'd used a faster chip, going back to the three-second wait really cramped. Even though you don't *need* a faster procesor, chances are that the next time you upgrade, you'll start noticing little things like that, and say "Wow... this is nice."

As an interesting side note, if you're looking for longevity out of a computer, go dual CPU's. I had a dual Pentium-133 w/ 64 megs sitting around that I bought for $40. For fun, I put NT4 on it, and in nearly every situation, it was almost or more responsive than a P3-650 with Windows 98. Yes, computationally-bound processes took a while, but in sheer responsiveness, it really impressed me. I think that a dual 1.6 GHz Athlon would have a tremendously long usable life span.

steve

Re:Fast CPUs might be bad.

[Junks Jerzey]

With the recent entrance of the seccond generation MMORPG on the gaming front, processors that were considered decent no more then six months agao are now not even worthy of the title 'decrepit'. Games like Anarchy Online, Starwars Galaxies, and Dark Age of Camelot, only to name a few, are pushing the benchmark higher and higher for gamer.

MMORPGs are more dependent on *bandwidth* than anything else. You're just talking about the 3D side of things.

look at a few new less eppic games such as Giant: Citizen Kabuto from early last year, people couldnt run that with all the widgets and gizmos cranked up to the maximum level, I could barely even play it on my PIII 500, and that was almost a year agao.

That's because 90% of the time in games like that is spend inside of a 3D driver. Switch the game to pre-assembled display lists (i.e. "use the transformation capabilities of the card") and you'll get a 10x speedup on the same machine. The trouble is that game developers can't assume such a card, as there are lots of entry level machines that are shipping with bare-bones 3D capabilities.

Re:get your terms correct

[Tet]

I meant that no one ever uses MICROMETER

No, but we do use the micrometre. The same way we use microfarads, microseconds and microvolts. I guess in the US you still use microns, but then you still use feet, inches, pounds and ounces, too. You have a perfectly good system of SI units, so why not use them? At least micron is just another name for a valid SI unit. Unlike Angstroms, which are just an abomination against nature (they should have just used nm or pm as appropriate).

Stop the train!

[niekze]

Am I the only one who notices that every week /. posts a news article about Intel or someone coming up with supar-dupar-mega-fantabulous technology that we never hear about again?

Like New Optical DSPs With Tera-ops Performanc
Or Intel Cites Breakthrough In Transistor Design
Perhaps Clockless Chips
Not forgetting Intel Promises A Cool Billion (Transistors)
Notwithstanding Intel Claims Smallest, Fastest Transistor
But who could forget Intel Claims 10Ghz Transistor
Which looks a lot like Intel Says 10GHz By 2005
But is just as vapor as Intel Creates 30-Nanometer Transistors
or my personal favorite: Intel Goes for Display Encryption

How can they get any work done when they're too busy telling us what they predict in a bajillion years?

Could faster processors lead to better programs?

[entrox]

I'm looking forward to ever-increasing clockspeeds, as this could get us away from programming applications in a low-level language like C/C++. Let's face it: Most of the bugs in current programs stem from the fact that C was not designed to handle sloppy or lazy coding. Dangling pointers, buffer overflows, memory leaks etc. result from the low-levelness of C (that's OK - for it to be efficient it needs to have the ability to do all kinds of things with the hardware directly). C should only be used for developing operating system kernels and device drivers, as no other higher language would handle the task well.

Faster processors and more memory would make higher languages such as Lisp or Python viable for applications (such as Browsers, Desktop environments etc.), which in turn would result in less bugs and increased stability when applied correctly. The current state with software makes me sick. I don't blame it on C per se, but on programmers using the wrong tool for the wrong job.

Writing in such a higher language would probably even increase portability (which C can't fulfill by a far shot) as you would program at a higher abstraction level. No need for autoconf/automake or ugly defines scattered throughout the code, making maintainance more difficult.

I hope that more coders switch to some better suited language than C/C++ for application development. I've switched to Lisp myself.

The Pentium 4 is now just a puppy with big feet.

[Futurepower(tm)]

I consider the Northwood to be the "real" Pentium 4, just as other second-generation products like the 100MHz Pentium and "Coppermine" Pentium III have proven to be the "real" versions of Intel processors in the past.

I agree with this. The Pentium 4s we see today are just puppies with very big feet. They will grow up and become something much more impressive.

Will this help Redhat Bloatware?

[renehollan]

Yeah that was a bit of a troll.

I'm currently awaiting my first new PC in a long-time: Soyo Dragon+ mobo, AMD Athlon XP 1600+ with 512 MB DDR RAM, ATA/100 WD 100GB disk (yes, /me likes SCSI, but likes $$$ more), generic DVDROM, and Netstream2000 H/W MPEG2 board. In preparation of it's arrival I downloaded a copy of Red Hat Linux 7.2 with the intention of installing it on an old spare 1.5 GB drive I had free in my old, ailing PC (Intel P200, 80 MB RAM), just to give it a whirl.

Well, things were real tight with the small drive, and my on-board IDE controllers were acting flaky anyway, so I ended up getting a "spare" 20 GB ATA/100 Maxtor drive and Maxtor (re-labeled Promise) ATA/100 PCI controller. The 20 GB Maxtor was now UDAM5 hde on ide2.

The point of all this history is to illustrate that I now have a "soon to be spare" computer where the limiting factor is CPU and to a lesser extent RAM. I go ahead an install RH Linux 7.2 on the new drive.

After a bit of farkling around with kernel boot options (ide2=d000,c802 is your friend!) I boot into RH Linux 7.2, in all it's X 4.0.1 glory.

.... and it struck me as slower than RH 6.2 on the same box, running from a slower drive on a slower, flaky mobo IDE interface (prolly not even ATA/33). Not much slower, but slower nevertheless.

I'm fairly sure that the new box would make the speed difference between RH 6.2 and 7.2 imperceptible, but the experience left me wondering about the extent of bloat in RH Linux releases, not that I'd want to run anything significant on the P200 anymore, but I might want to use it as some type of low-duty server, with an up-to-date kernel. In a nutshell, what got slower?

No doubt, the new machine will be welcome.

Re:Will this help Redhat Bloatware?

[renehollan]

Point noted, except you're comparing apples to oranges and I compared apples to apples.

In my case, unless I'm doing some long-term processing, the key isn't "fastest", but rather fast enough. I wouldn't spend much $$$ to get a kernel build down to 30 seconds from a minute, for example -- a minute is fine for me for the few times that I build a kernel. 30 minutes, of course, is annoying.

Re:Will this help Redhat Bloatware?

[Shabazz]

I was just giving another example. RH6.2 on my computer was faster than Win98SE. I recently wiped everything and went Win2k, RH7.2 without changing any hardware.

Everyone who wants to see Linux succeed on the desktop (including myself) needs to recognize that all those bad words people hurl at MS won't change the fact that Linux + XF4.0 runs significantly slower on the same hardware.

A lot of the advantages of Linux on the desktop start to disappear when you realize that it takes a lot of power to run it. It's not agonizingly slow on my computer, but it's pretty frustrating. Especially when Win2k just hums along on a slower disk with an "inferior" interface.

Re:Will this help Redhat Bloatware?

[renehollan]

A lot of the advantages of Linux on the desktop start to disappear when you realize that it takes a lot of power to run it.

True enough. I remember the days, probably up to RH 6.2 when [GNU/]Linux distros were generally snappier than bloated Microsoft offerings, even as the user productivity apps were less mature. It would be a sad day indeed, when standard GUI and productivity apps available under a [GNU/]Linux distro were slower, just to get more features "out there" -- stick to the tried, true, and efficient, until the polished can compete with it's peers on performance.

Cyrix and Transmeta

[Anonymous Coward]

Yawn. Another article purely pushing Intel's and AMD's chips going another notch in clock speed. In the meantime, Cyrix and Transmeta have both shipped CPUs based on new cores, the Cyrix one at .13 micron, and no one bothers to mention it.

Re:Fast CPUs might be bad.

[Tackhead]

> has anyone else been telling their 'we want to get on the internet' family and friends to buy OLD PCs?

Yes, yes, yes, yes, yes!

For those of you who have a nearby surplus store, go there.

Sample upgrades: I saw a $30 P166MMX system that happened to have an Asus TX97 motherboard. A free upgrade to the beta 0112 BIOS from the 'net. A $30 K6-2+-450 laptop chip, or a K6-3-333 are drop-in replacements for the P166MMX, and offer performance comparable to a PII in the same speed range. Such a system is a great place to toss that stick of 64M PC100 SDRAM you're not using anymore, as well as that 8.4G hard drive you just replaced.

I did that upgrade for my own box and it's capable of doing all-software DVD on a cheap-azz 4M ATI TV-out card from 1996 with no DVD hardware support.

Want monitors? Surplus stores rule. I was in one yesterday and picked up a 19" Sony true-flat CRT for $120. (Pricewatch: $400-500). They had 17" Sony flat-CRTs for $70 (Pricewatch: ~$300). There were also several 21" monitors (Viewsonic P815, Pricewatch $700 new, $325 refurb) for ~$200.

Did the 64-bit AMD technology slip?

[Animats]

The AMD roadmap shows their 64-bit CPU in late 2002. Is that a delay from previous announcements?

That's too late. They need it sooner to compete with the Inanium.

Re:Please don't take it personally...

[Just Some Guy]

I'm thinking about writing agent ai code in *asm*, because compiler generated code won't be fast enough.

You're in a position to write a thesis involving AI, so I assume that you already have a bachelor's in CompSci.

Did you learn nothing from it?

I mean, do you honestly believe that you can increase the speed of your programs enough by abandoning an easier-to-maintain language to make it worthwhile? What happens when the next generation of compilers comes around that's faster than your hand-tooled assembler, and you have to re-write your code yet again to squeeze out those extra cycles? What if your code gets executed on a modern processor with deep pipelining, advanced branch prediction, and out-of-order execution? Are you that confident that your manual re-write will take full advantage of the hardware it's running on, moreso than a computer-optimized version?

I'm sorry, but unless your AI consists of a very few tightly-rolled loops that you can super-optimize, I just can't see the benefit of throwing away 30+ years of compiler design experience for a theoretical gain that may or may not appear.

Re:Please don't take it personally...

[MarkusQ]

by Anonymous Coward on 10:57 03 December 2001 (#2648773)

but I'm sick of this "C for kernel, bloatlang for everything else" BS. I'm writing a ALife sim of language evolution program (for my thesis) in C and I'm thinking about writing agent ai code in *asm*, because compiler generated code won't be fast enough.

I assume you registered at the university instead of attending as an "Anonymous Coward"--otherwise, the diploma isn't going to do you much good.

-- MarkusQ

it's cooler

[geekoid]

I have a 1.4 MgHz chip. thats plenty fast. the reason I want .13 is so it will run cooler

Re:it's cooler

[msaavedra]

I agree wholeheartedly with that sentiment. Unfortunately, I have a feeling the chipmakers plan on bumping up the clock speed to the point that the chips run just as hot. I personally would rather have a 1 GHz chip that dissipates 10 watts than a 2GHz chip that dissipates 70. I don't think these new chips will be available at 1GHz, though.

Re:Could faster processors lead to better programs

[geekoid]

personally I would rether drum all the sloppy and lazy coders out of the business.
screw ease of programming and code readability. I want someone who is smart enough to figure it out to be coding, not some high level coder wanna-be.

Re:Please don't take it personally...

[Just Some Guy]

That assumption is wrong, my undergrad has nothing to do with compsci, but my master is cogs. Nevertheless I assume I know enough for undertaking such a project. Fell free to prove me wrong, I can only benefit.

The issue that I see is that even a 40% performance increase is still well below what I would consider to be the threshold of useful optimization. I mean, that's less than one year's hardware progress. Sure, it may be tempting now, but wouldn't it be better to leave your research in a form that's more accessible to a large audience? Granted, C may not be the lingua franca of AI research, but I'd be willing to bet money that more AI researchers can read C than AMD assembler. Wouldn't you at least like to have the option to have your code peer-reviewed by researchers who don't know 3dnow?

I'm not the moron you are thinking of.

If you're implementing ANNs, then I'll give you the benefit of the doubt and assume that you're not. :) Still, I disagree with the conclusion you reached. The ability to maintain programs I wrote 6 months ago and share them with my colleagues is much more important, IMHO, than squeezing out every last bit of performance.

Already happened...

[srvivn21]

Look at the transition from processor specific assembly to (possibly portable) C code. (Were computers not originally programmed with direct binary instructions?) More software is being written in interpretive languages (such as Perl and Java) than ever before. The transition you are asking for is occuring, whether you realize it or not. It has been occurring since computers came to be.

Intel ads on slashdot?

[blair1q]

Should I be surprised that I've never noticed the P4 ads until I saw the one popping up at the top of this thread?

--Blair

Premature optimization

[Convergence]

Premature optimization is the root of all evil. I also was considering making the same types of choices you are looking at.

My code deals with building massive 3d arrays containing tens of millions of cells and manipulating them. Obviously, the inner loops of the manipulation would be the bottleneck.

So I ran my trusty profiler.... And found out that 90% of my time was being spent READING THE DATA IN.

It took two lines of code to make that three times faster, making my program 2.5x faster.

Interesting... Then, a couple of weeks later, I took a large deployed system with an active developer community, www.squeak.org, and ran that through a profiler, and found out how changing one line of code lead to 4% speedup in the core intererpreter, and lead to other simple changes that were just as valuable. I also ran the benchmarking in the interpreter, and sped up syntax hilighting by 40%.

If I was doing something like what you were, I'd probably go all-out at using a more dynamic language (Smalltalk) for the extreme flexibility.

Only devolve into C/assembly for the critical parts.

Many times, the bottlenecks aren't where you think or might predict they are. Why spend weeks guessing incorrectly and optimizing code that won't help you go faster when the profiler will tell you exactly what magic bits to re-examine.

It can also find O(n^2) artifacts and all the rest.

If your code is currently running, run it through gprof and see where the CPU time is really going.

Your personal favorite

[Convergence]

The display encryption idea from intel, that is the much ballyhooed HDCP, that was the subject of Niels Ferguson a few months ago. (also on slashdot), and one week ago, slashdot posted a news story announcing that it is completely broken.

BTW, its not vapor, Apparently, a ten thousand bux 42 inch rear-projection TV from JVC actually is using the piece of digital control crap.

Re:Could faster processors lead to better programs

[benedict]

On the contrary, laziness is one of the three chief virtues of a programmer.

It's especially good when it keeps them from writing code in the first place ... less code means fewer bugs!

Re:Could faster processors lead to better programs

[benedict]

C is *so* not a cure for laziness. Some of the most lazy and careless code I've ever seen has been C.

It could be argued that a language that lets you express yourself tersely is less prone to laziness problems. If there aren't a million t's to cross and i's to dot, then there's no way to have a million uncrossed t's and undotted i's laying around at the end of the day.