Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
Intel Hardware

Intel Roadmap Update: The Art of Naming Processors 239

Posted by CowboyNeal from the chip-by-any-other-name dept.
THG writes "CoolTechZone.com has compiled a list of Intel processors from its roadmaps, and discusses Intel's naming convention. According to the article, 'Gone are the days when processor names were something as simple as their clock speeds. If you wanted a nice and powerful 3GHz processor, you simply asked for a P4 3.0GHz and that was it. Ever since Intel has decided to revamp its naming conventions, the confusion makes you wonder if the whole idea of renaming was a smart move. Moving on with Intel and it's desktop endeavors, the problem is that if the names were as simple as stated above, we would've somehow managed to figure them all out. But someone at Intel obviously wanted to ensure that we don't remember processor names without having a 100-page manual on product families, so there are modifications to each series, which may or may not be consistent across different series.'"
This discussion has been archived. No new comments can be posted.

Intel Roadmap Update: The Art of Naming Processors More

Intel Roadmap Update: The Art of Naming Processors

Comments Filter:

  • Intel's naming scheme is convenient (Score:5, Funny)

    by nizo ( 81281 ) * on Thursday November 10, 2005 @06:51PM (#14003346) Homepage Journal
    Actually I like Intel's complicated scheme; instead of looking up which CPU is which I just remember to go buy an AMD [amd.com] processor instead. Probably not what Intel had in mind when they came up with an overly complicated naming scheme however.

    • Re:Intel's naming scheme is convenient (Score:2, Funny)

      by Anonymous Coward
      I think I speak for everyone when I say this - thanks for linking to AMD's website. I certainly had no idea where to find it.
    • hahahaha, pwned.

      Totally agree.

      Would be neat to see a new core from either camp though just for the sake of science [e.g. what new designs are out there? I'd like to think the Athlon isn't the "best" design ever for a CPU].

      Am I the only one noting the "improved" presler cores take about as much as the AMDX2 processors do TODAY? ...

      Tom

    • Re:Intel's naming scheme is convenient (Score:4, Insightful)

      by Nik13 ( 837926 ) on Thursday November 10, 2005 @09:18PM (#14004315) Homepage
      AMD's naming system may not be the best, but I like their rating system. Clueless n00bs have been complaining that they were "cheating" by not giving the actual clock speed (which alone means very little). You can tell approximately how much faster is a specific chip over another one they sell using that (and an equivalent P4 somewhat). It's not totally accurate, but you know a 4200+ will be about twice as fast as my old Athlon XP 2100+ or a P4 2.0GHz. Anyone can buy a chip using a system like that.

      Whereas with the current Intel chips... Model numbers (a 519? how fast is that really?), different sockets, different FSBs, different cache sizes, different cores, different intructions sets (SS3 or not, EMT64 or not), dual core or not... You can't easily tell how fast one is over the other ones (nor can you tell easily which ones run cooler). They're finally victim of their own GHz ratings and they got nothing to go by anymore (as a measure of relative speed) it seems. Unless you're following their offerings closely (most people aren't), then it's pretty hard to pick one.
    • You're right. That's probably a side-effect they didn't expect. It stems from the marketing division.

      Joe: I bought me an Intel Cele-ron a few months back.
      Marky: Oh yeas, well I just got me an Opteron.

      The latter clearly sounds more superiour, yet in the same "class".

  • Why would you want a 3Ghz CPU? (Score:5, Funny)

    by temojen ( 678985 ) on Thursday November 10, 2005 @06:52PM (#14003358) Journal
    Wouldn't you be more concerned with the performance than a specific clock speed? This is of course assuming you're not using it as a RF source or something.
    • Intel's architecture is deep pipe but few IPC. So pretty much clockspeed is performance.

      The problem isn't the idea of renaming, its how they did so. AMD's renaming is simple- higher number, higher performance. Dual core is called an x2. ^4 bit is called just that.

      Intel's renaming scheme- umm, I really can't find a pattern in it.
      • The naming (numbering) of AMD processors is, at best, loosely tied to performance.

        Not that it's unique to AMD.. the value version of almost any video card, while carrying a higher model and often price tag than it's full-blown predecessor, is usually lower in performance.
      • > ^4 bit is called just that.

        Grammar hint: Don't try to capitalize numbers at the beginning of your sentence :)
      • I agree that Intel's new naming scheme is confusing at best. Perhaps they should name their processors based on performance instead of clockspeed. The equivilent of an Athlon 64 3800+ could be called Pentium 3800++ Codename, Ass-kicker (Extra plus added by the Marketing Department) (Codename also created by Marketing)
    • I can quote numbers of CPU cycles as "nanoseconds" without conversion if the CPU runs at 1 GHz.
    • To be fair, until the P4 came along MHhz was pretty synonymous with "speed" (except for AMD processors which were slower - don't flame me, I owned a K6233 and I speak the truth of its disappointing performance). And even today the MHz "myth" is almost universally true within processor families, and still mostly true between processor families; it's rare that a 2:1 advantage in MHz fails to predict which chip will be faster. In other words people who think MHz is unrelated to effective speed are too clever
      • But the WHOLE POINT of the article is that you can't even judge that anymore. Even going by the names instead of the model #s, is the Pentium M in the same family as the Pentium 4-M? or the Celeron M? A Pentium 4 at twice the clockspeed is still in the same performance ballpark as a Pentium M, and future faster chips will branch from the Pentium M tree, so I think your assertion is wrong now and getting wronger by the day.
    • I would be, but sadly, the number does not express this. Some of the 5xx series outperform the 6xx series, and the 840 and 840EE have quite different performance. Add to that, the 7xx Pentium Ms, while possessing a higher number than the 3xx, 5xx and 6xx processors, are outperformed by them for the most part (some of the celerons might be slower). And its not even based on price! The nice 6xx's cost the same or more than the 8xx's!

  • All Intel has been doing... (Score:5, Funny)

    by grub ( 11606 ) <slashdot@grub.net> on Thursday November 10, 2005 @06:53PM (#14003370) Homepage Journal

    ... is take a few letters or a small word and add "ium" to it. They had a chip which gave off a musky odour but was irresistable. Unfortunately the "Cuntium" never made it out of the lab.

  • Intel (Score:4, Insightful)

    by Anonymous Coward on Thursday November 10, 2005 @06:53PM (#14003377)
    This is only going to get more and more confusing with multi-core. Users are going to have to distinguish not only based on clock speed, but number of processors, and with HT (number of logical processors). Add to that the fact that it is unclear what advantages these multiple cores have with current client operating systems, given that there aren't too many true multi-threaded applications out there, and this becomes bewildering for even a savvy consumer.
    • This is only going to get more and more confusing with multi-core. Users are going to have to distinguish not only based on clock speed, but number of processors, and with HT (number of logical processors).

      That's just the point, I think Intel and others want to move away from having typical consumers have to wade through all this crap. Using names that are associated with the intended use of the processor, instead of the specs, is supposed to achieve this.

      If I go to buy a car, the only metric I'll use to ev
      • If I go to buy a car, the only metric I'll use to evaluate the engine is the number of cylinders.

        This is not a good analogy as your car engine performance metric is extremely flawed. I think HP and torque are a better ways of measuring engine performance. In your view the 145HP 3.0L V6 Ford Vulcan engine with two valves per cylinder is about as good as a 200HP 3.0L V6 Ford Duratec engine which has four valves per cylinder. On the other hand, Saab's sophisticated 250HP 2.3L 4-cylinder engine must be necessar
    • Kind of like the pharmaceutical companies and the oil companies. Let's confuse the people and _tell them_ what they need to purchase.

      * ducks *

      At least at that point we might be able to join in on the fatcat profit sharing.

      Perhaps there is a new industry emerging - "Purchase Consulting"?

  • Real speed != clock speed (Score:4, Insightful)

    by JimBowen ( 885772 ) on Thursday November 10, 2005 @06:54PM (#14003384)
    The reason for scrapping clock speeds in favour of these 'strange' naming conventions is not confusion, it is to help people realise that clock speed does NOT indicate how fast a processor is.
    If people thought that a 3GHz celeron is as fast as a 3GHz P4 with HT, or indeed a 3GHz Athlon64, then they would be very confused indeed.
    Many people did think this though, before the new naming conventions applied, so I think it is a good thing.
    • That's also a good sign that people don't have a fucking clue WHAT they need and they only buy the higher number because of small penis syndrome.

      I mean honestly try to explain to someone that a $50 Sempron running at 1.4Ghz would do them just fine for writing Word documents, playing solitaire and doing email. Then keep a straight face when they ask "why don't I buy this 3.2Ghz Dell computer?" Sure there are a lot of gamers/developers out there that need the juice but there are still a huge amount of peopl
      • I would agree, but what you fail to realise is that these consumers a planning ahead. They know that they'll use this machine for 2-3 years at least before upgrading. They also know that they'll be so riddled with spyware at the halfway point that they NEED that extra 1.7 GHz just to pull through.

        on a serious note though, I keep trying to convince my mother to switch to linux on her crappy little laptop. All she does with it is surf and email and if she'd just ditch that bloated MS and put in a nice light-w

        • What consumers don't seem to realise is that PC's don't get slower as they age. If it's fine for using Word and surfing the web now, it will be fine in two years' time. Unless the consumer anticipates their needs changing, they are wating their money by 'planning ahead'.

          Probably about to start a 'reverse penis size' thread here, but I can get by very nicely with my 466 MHz G4 tower which I use mainly for web, email, terminal sessions and occasionally Word/Excel. It's every bit as fast as the day I got it.

      • I have to echo the planning ahead post above. Yeah, you might be able to do those chores you mentioned, but whay happens when they buy a digital camera and now want to process images? What about a future video iPod and the desire to transcode video?

        I'm more in the buy future capacity camp. Yes, it's possible to have too much, but then again, that beats out not having enough and having to trash everything and start over...

        • I don't buy this. Furthermore I think it's just a symptom of gross waste. We had symlinks and advanced networking in Linux 5 years ago [and more but let's be fair]. Back when the average box was a 386 running at 25Mhz with 4MB of ram.

          It takes Microsoft until 2006 to come out with an OS that fundamentally isn't better than Win3.11 [just shinier, more doodahs] and finally some proper shell, etc. And a computer running 3Ghz with a GB of ram, etc...

          As another poster pointed out, if you can run your wordproc
      • What does your small penis have to do with Opteron chips offering fantastic performance?

    • Well, since we need to compare to BMW according to the article: a series 3 maximum speed is 250km/h, so is the maximum speed of a series 5, and 'lo and behold also of a series 7! Wow, they all are equivalent! ;-) (For those that don't know: stock BMW's are electronically locked to 250km/h, they can do more if the enigine allows is and if they are unlocked)

      All in all: CPU speed doesn't matter... especially not when talking Intel ;-)

    • And of course AMD has been doing something similar for ages (although their scheme is somewhat simpler). For example, an Athlon64 3000+ doesn't run at 3GHz - it only runs at 1.8GHz.
    • If people are confussed by Intel's naming convensions, and its causing problems for Intel's sales (say people are just buying AMD instead) then it is only what Intel deserved! They've been peddling the clock speed myth for years and so if it is starting to hurt their bottom line, they made the bed ... Anyway, based on recent talk from the company and Apple's switch, I'm sure the next metric they will be selling everyone on as being the only one to worry about is performance per watt. These awkward names

    • Intel is being evasive about true performance (Score:4, Insightful)

      by tomcres ( 925786 ) on Thursday November 10, 2005 @07:39PM (#14003748)
      The problem is that Intel is using an arbitrary numbering scheme which has absolutely nothing to do with any objective measure of performance. Their numbers simply refer to the relative number of features and relative speed of processors within the same family. This makes it impossible to compare processors across different brands. A Celeron-D 560 is not going to necessarily be as fast as a Pentium-D 320, but you'd never know by the numbers. On top of which, you'd never know that the -D in Celeron-D is for "desktop" whereas the -D in Pentium-D is for "dual-core." Of course, we techies know this, but this is Intel's way of deliberately misleading consumers.

      AMD, on the other hand, uses a P-number which is directly comparable across processor lines and uses an established standard of a 1GHz Athlon Thunderbird = P1000. Everything else is relative to that. So you know right off the bat that an Athlon64 3000+ is only marginally faster than a Sempron 2800+, you don't have to play games like with Intel.

      • AMD, on the other hand, uses a P-number which is directly comparable across processor lines and uses an established standard of a 1GHz Athlon Thunderbird = P1000.

        Where did this come from? I went to a presentation by an AMD engineer a couple years ago and he did not say it was this way. He said it was a comparison against the clock speed of Intel's prevailing product, that a 3000+ was supposed to be roughly equivalent (on average, based on an average of a large number of benchmarks) to a 3.0 GHz Intel Penti

  • I thought they used... (Score:5, Funny)

    by sczimme ( 603413 ) on Thursday November 10, 2005 @06:55PM (#14003395)

    I thought Intel just put a regional map over a dart board:

    *thunk* - "Williamette"

    *thunk* - "Tillamook"

    et cetera, et cetera

    • I thought Intel just put a regional map over a dart board:
      *thunk* - "Williamette"
      *thunk* - "Tillamook"

      Maybe they should consider something more suitable to their latest offering and switch to the names of cheeses.

      *thunk* - "Vieux Boulogne"

      *thunk* - "Stinking Bishop"

      *thunk* - "Limburger"


      • Maybe they should consider something more suitable to their latest offering and switch to the names of cheeses.

        They're way ahead of you! Have a look. [tillamookcheese.com]

        Scary, n'cest pas?

        *thunk* - "Stinking Bishop"

        I don't know why, but that was absolutely bloody hilarious. :-)

    • I thought Intel just put a regional map over a dart board:

      That's funny, I thought the same thing about AMD:

      *thunk* - "Venice"

      *thunk* - "San Diego"

      *thunk* - "Manchester"

      *thunk* - "Denmark"

      *thunk* - "Toledo"

      *thunk* - "Venus" (damn, that one sure flew off the board!!)
    • *thunk* - "Williamette"

      *thunk* - "Tillamook"

      *thunk* - "Cedar Mill"

      Hey, all in Oregon! I can see three options here:
      1. The dart thrower is precise and likes Oregon.
      2. It is just a map of Oregon.
      3. The dart thrower is precise and hates Oregon.

    • Re:I thought they used... (Score:4, Informative)

      by JeremyALogan ( 622913 ) on Thursday November 10, 2005 @08:06PM (#14003917) Homepage
      It's worth mentioning (for the uninitiated) that the parent is refering to areas in and around Portland, OR. Intel's main campus is in one of the Portland suburbs. Some of the names they've used so far (off the top of my head):

      Willamette - A river in Oregon. Runs north-south through Portland.
      Prescott - A city in Oregon. Also a major street in North and North East Portland.
      Madison - A street in Portland... not sure it it's much else...
      Tualatin - A sothern suburb of Portland. Also a street in Portland.
      McKinley - A city in Oregon.

      Tillamook is a town in western Oregon known for it's cheese factory. ALSO a street in Portland :)
      • Other Intel chips and chipset names can give you an indication of where they were developed. Most parts are designed in Oregon. But, for example, most of Intel's low power parts have names of Israeli geographical features (Banias, Dothan, Merom, Gilo, Jonah, Dimona), and this probably means they were developed at Intel Israel. Expect to see some Indian rivers show up on the list as soon as that development site is up to speed. See the huge list of code names [erols.com] for a geography lesson of the Pacific Northwe
    • Intel's official code name scheme is a place or geographic entity in the United States or Canada. Every project picks one of these code names and gets them OK'd by legal. Since a good portion of the engineers live in the Portland, OR area they often pick names from this area. However, McKinley is likely named after Mt. McKinley, and Madison is probably named after the capital of Wisconsin. But the dart method is a good idea. I think I'll try that for my next project.

  • The loser always wants to hide. (Score:5, Insightful)

    by r00t ( 33219 ) on Thursday November 10, 2005 @06:58PM (#14003425) Journal
    If you are behind in GHz, avoid discussing it.

    If you are behind in benchmarks, avoid discussing it. (Look! GHz!)

    If you are behind in low-power, avoid discussing it.

    If your expensive flagship "server" CPU is only 2% faster
    than the gamer version, avoid discussing EVERYTHING that
    could possibly matter.

    Grrrr.... I wish I could force them to include SPEC benchmark
    numbers in the processor names. Put the lowest number first,
    then a "-", and then the highest number. Slimy bastards always
    hide from the light.

    • SPEC is not ideal (Score:4, Informative)

      by vlad_petric ( 94134 ) on Thursday November 10, 2005 @07:39PM (#14003747) Homepage
      Two problems with it:
      • not really indicative of anything. Some of the workloads in SPEC are what a unix hacker would run (perl,gcc,bzip/gzip), but most other are very obscure pieces of software.
      • "good at SPEC" is totally different from "good at server workloads". The former are generally CPU-bound, whereas the latter - memory bound (so for the former you want high clock and high machine width, for the latter, however, you want sh*tloads of caches). Also "good at SPEC" is different from "good at media" (the reasons are more complex)
      • Being "not really indicative of anything" issue is a lame excuse. Sure, not many people want to run SPEC all day, but not many people want to run any other benchmark all day either. All benchmarks, except ones customized to your own specific usage, will suffer from this problem. Despite this, we still use benchmarks.

        SPEC is pretty decent for server workloads, especially when you consider a recent SPEC*RATE benchmark.

        If you are thinking about multimedia instruction sets when you mention "good at media",

        • My point was - if you play games, you're probably interested in FPS numbers or 3d marks (sure, in conjuction with a video card, etc); if you do server stuff, then you probably want to see how good it's doing at SPECweb/TPCx. But what does SPEC really tell you ? How can you use it to approximate the behavior of other apps?

          Furthermore, SPEC*RATE is *not* a good idea for server stuff. From the SPEC benchmark suite, only one program is memory bound (mcf). OTOH, database workloads are pretty much always memory

    • > wish I could force them to include SPEC benchmark
      > numbers in the processor names.

      ALL BENCHMARKS ARE FLAWED! We talk about faster processors, but faster for what? This isn't a consistant thing. A command could take 3 clocks on one processor and 4 on the other. That single CPU command could make a difference in one benchmark and not another when those add up. Multiply this by tons of instructions of differing proportions. There are different methods for doing a lot of things in every CPU. Di
      • Gamer benchmarks greatly rely on the video card. That's no way to test a CPU.

        SPEC is a very nice mix of stuff, much of it similar to what I do all the time.

        If I wanted a whole-system benchmark I might use SDET, but since we are
        discussing general-purpose CPU performance here, SPEC is most appropriate.
        • I used gamers as an example, but it was a general benchmark statement.

          ANY benchmark on two different processes is entirely different. A car can be entirely outpowered or it can be stronger in some areas and not others. It's the sum of all of those that shows the performance... but what does that mean?

          -M
          • OK wise guy. Whatever. Let me sell you a nice 486SLC2-66 with a 33 MHz external coprocessor. Most people don't like the benchmark numbers, but I'm sure you won't be swayed by that foolishness. The external coprocessor helps with math. You don't get that with most systems, so this is a special deal.

            Go read up on SPEC [spec.org]. It's not like BogoMIPS.

  • "Do you want fries with that?"

    Processor names, and their various compatible motherboards and heatsinks has been a thorn in the sides of custom computer builders for some time now. I understand we can't pin companies down to certain standards so everything works with all other available equipment, but wow it would be nicer if we could. You wouldn't have to learn the latest tech trends to get a 3GHz upgrade, by matching a CPU with a processor, with the right heat sink, and read up on the compatible RAM etc,
    • I don't see the problem. Once you know which form factor your CPU is you're 90 percent of the way to figuring out which CPUs you can use. When I could find a Pentium with AMD64 (screw you and your EM64T shit, Intel) for a decent price it took about five minutes for me to figure out if it would work with my motherboard (BIOS update necessary) before I ordered it. This stuff isn't rocket science.

  • Duron, Athlon, Itanium, Centrino, Xeon, Centron - ARGH!

    Seriously - the name list is so long we could start naming HURRICANES after these!

    I just want to know which one is the latest. Is something like "P9-MMX2" too much to ask? That way I can know I'm not being scammed because the processor would read P8 instead of P9.
    • the latest what? 64bit? dual core? low power? low end consumer? etc . . .

      yes, I too mis the days when you said:
      "what's the fastest thing out?"
      "486/50"
      "I'll take one!"

  • Comment removed (Score:5, Interesting)

    by account_deleted ( 4530225 ) on Thursday November 10, 2005 @07:07PM (#14003502)
    Comment removed based on user account deletion

  • ye good olde days of chip numbers (Score:5, Informative)

    by G4from128k ( 686170 ) on Thursday November 10, 2005 @07:08PM (#14003514)
    In the olden days, we didn't even have words. It was 8086, 8087, 80286, 80287, etc. No Optiums or Penturons back then.

    Trademark issues drove Intel to make up processor names -- Intel couldn't stop competitors from selling non-Intel 80486 chips because chip numbering was a generic identification scheme in the electronics industry.

  • ...they failed to call the P6 the "sextium".
    • AMD already came up with the antidote to the "Sexium" should Intel decide to use that name... Sempron!

      (for the benefit of those that do not speak Portuguese... sem pr0n = without pr0n.. although, due to a peculiarity of the Portuguese language, words cannot end with the letter 'n'.. IIRC, the whole Inquisition was started because some heretics started using the letter "n" at the end of words.. it's true! really!)

    • Given the clock speeds of the time, we could have had a "Hexium 666".

  • AMD (Score:2, Insightful)

    by Serff ( 183148 )
    Didn't AMD start the whole thing? We could also blame Apple for starting the conversation of the "Mega Hertz Myth" too. But I like those 2 companies, so lets just blame Microsoft for the whole thing instead. ;)

  • All this naming confusion (Score:5, Funny)

    by n6kuy ( 172098 ) on Thursday November 10, 2005 @07:09PM (#14003529)
    ..makes me reach for my Nexium.

  • Code Names (Score:4, Interesting)

    by SmartSsa ( 19152 ) on Thursday November 10, 2005 @07:13PM (#14003566) Homepage Journal
    Code names are just as bad as the official part numbers.

    However, if you haven't figured out already, Intel is moving away from directly selling CPUs based on their speeds and starting to bundle 'Platforms.'

    This started mostly with Centrino (the platform), since it's not a CPU. And is now continuing into the Desktop and Server marketspaces.

    It's their hopes that end users won't ask for "pentium 4!" but rather (insert catchy platform name here). It's worked well with Laptops. People want Centrino! And it'll likely work with Desktops, but probably not so much servers.

    With that their naming conventions for individual parts are also going to get even more screwy...

    But, on the other hand, Intel is not the only one to have evil codenames. They, as well as their competators, should just stick with sequential numbering so one can say "higher number is newer!"
  • I think the reason for naming conventions change is relatively simple (though cetrtainly, IMHO, not too solid): Marketing. Gone are the days when only people asking 'what processor is in that box' were tecchies, geeks, nerd and co - to them, something like P4-2/3.2-800/2 , meaning Dual core Pentium 4@3.2 Ghz frequency and 800 MHz freq + 2 MB cache made sense as an understandable, technical naming convention. To Joe Schmoe, it is confusing. He likes to hear "Yes, sirree, the finest gen-yu-wain Pentium 4 Smit
  • Having recently started looking at Intel laptops, I was hoping for a little information on the Laptop processors, especially as the clock speeds stated and actual are so varied. Sadly, the article only touched on Laptop series processors and failed to provide any real depth to this point.

    GrpA
  • Two guys walks into a computer store, buying for processor chips.

    One person, searching the counters, grabs a staff member. "Excuse me," he says, "do you have an Intel..."

    The staff member groans, faces towards the person, and says "Okay, an Intel..."

    The person jyrates to some unheard music as he says "Killamanjaro Quad-Core Ultrahyperthreaded Coochie coochie Low Watt Midtown Bus 314159629 processor."

    The staff member asks "That's the Tango model or the Disco model?"

    Meanwhile, the other person is already at t

  • It's irrelevant (Score:4, Insightful)

    by drsmithy ( 35869 ) <drsmithy@gm[ ].com ['ail' in gap]> on Thursday November 10, 2005 @08:16PM (#14003975)
    The only people who really care about a processor's specifics (bus speeds, cache sizes, clock speed, etc) are techies. Techies don't just walk into a store, grab a CPU off the shelf and pay for it. They actually research what they're buying and decide on what they want to buy *long* before they actually begin th purchasing process. So, since it's easy to find out which marketing names match up with which processor features, it doesn't really matter what their marketing names are.

    Added to that, any techie for which it's a matter of importance (eg: the bloke at your local computer fixit shop, 14 year old gamerz) will have memorised which marketing name has which processor features within hours of them being released, lest they not appear to be l33t enough.

    Everyone else just picks a price point and then buys whichever machine is at that price point the salesman tells them is best.

  • Today I was at a meeting discussing computing issues for numerical simulations of particle physics being done on computers at CERN. One big complaint was that they're having a hard time finding places to put some of their machines because of thermal issues.

    Now, these machines all run Linux, do server duties and loads of floating-point math (Monte Carlo calculations), and are in a situation where cooling is an issue.

    Nonetheless... on all the machines I've used there, /proc/cpuinfo says they're running P4's o

    • Re:And, yet, people continue to use P4's... (Score:4, Informative)

      by Jerry Coffin ( 824726 ) on Thursday November 10, 2005 @08:46PM (#14004127)
      Today I was at a meeting discussing computing issues for numerical simulations of particle physics being done on computers at CERN.
      [...]
      Nonetheless... on all the machines I've used there, /proc/cpuinfo says they're running P4's or P4-based Xeons. Wouldn't running Athlon 64's or Opterons (in 64-bit mode, since they're all on Linux) give better performance and less heat?

      That depends on their code. Numerical simulations are mostly floating point that's often quite vectorizable. In that case, they could be using SSE2 quite a bit, which generally works better on the Intel chips -- but they probably won't get much benefit from this unless they're hand-optimizing at least a few of their inner loops. Most compilers can do some automatic vectorization, but they don't make good enough use of the capability to overcome the Intel chip's shortcomings elsewhere, as a rule.

      OTOH, if they're doing a lot of vector math, they'd probably get considerably better performance still by writing the code to execute on the GPU instead. The obvious shortcoming of that would be accuracy problems -- the GPU's floating point is engineered far more to maximize speed than accuracy.

      --
      The universe is a figment of its own imagination.

  • Does having two cores make hyperthreading unncessary, or was it dropped because they couldn't build hyperthreading and dual CPUs into an LGA775 package?

  • Why not use megaflops in the CPU name? (Score:3)

    by master_p ( 608214 ) on Friday November 11, 2005 @06:00AM (#14006490)
    Talking about megaflops will certainly make it easier to compare processors and remember which model is which, as well as capture architectural differences. For example, Athlon 600 (for 600 megaflops), Pentium 500 etc.

Slashdot Top Deals

I cannot conceive that anybody will require multiplications at the rate of 40,000 or even 4,000 per hour ... -- F. H. Wales (1936)

Close