AMD’s TPI will not produce performance metric

I confess I’ve known about this for a little while, but I haven’t seen it reported elsewhere, so I suppose it’s still news. You will recall that AMD dubbed its model number-based rating system, upon its introduction with the Athlon XP, a “bridge metric.” The company pledged to replace the system with a more robust set of performance metrics to be produced by its True Performance Initiative, or TPI.

We decided to hold AMD to this promise, and thus I had a conversation with AMD’s TPI director, Hal Speed, back in December ’02. He said at that time that TPI was still on track for early ’03, with lots of new partners and new momentum backing the effort. However, 2003 came and went—as did the launches of AMD’s Opteron and Athlon 64 processors—with nary a peep about TPI’s new performance metric.

But here’s the kicker. I spoke to Mr. Speed again shortly before the official launch of the Athlon 64, and he revealed to me that the True Performance Initiative will not produce a performance metric. Speed said AMD was never able to get a full contingent of market leaders to sign on to the effort, and so the company decided to nix the project. Apparently, AMD actually thought Intel might be willing to join TPI. Speed said AMD had picked up “signals” that Intel might be receptive, perhaps in part because of the difficulty of marketing its own Pentium M processors. Obviously, things didn’t work out.

Mr. Speed disagreed with my characterization of TPI as a failure, though, claiming the discourse about processor performance had changed, and in that sense, TPI had succeeded. He cited the intentionally vague model numbers assigned to Opteron processors as healthy departure from older MHz-oriented thinking.

For most Athlon 64 products, AMD continues to use its model number rating system, shakily making reference to the mythical AMD Athlon “Thunderbird” processor against which the system is purportedly indexed. I say the reference is shaky for two reasons: one, no T-bird CPU has ever run at much more than 2GHz (consumer versions topped out at 1.4GHz), and two, AMD now apparently assigns Athlon 64 model numbers almost at will, based on a trio of factors: cache size, number of memory channels, and clock frequency. We have shown in our Athlon 64 3000+ review, published today, that the 3000+ and 3200+ products perform identically in many tasks.

In other words, we think the veneer on AMD’s “bridge metric” is wearing a little thin—even if we do still appreciate the need for some consumer guidance in the matter of comparative performance. Unfortunately, without the new, industry-standard performance metric AMD promised to produce via TPI, Joe Consumer may simply have to take AMD at its word about model numbers and relative performance. And taking AMD at its word, we have learned, may not always be the wisest thing to do.

Comments closed
    • Anonymous
    • 16 years ago

    “And taking AMD at its word, we have learned, may not always be the wisest thing to do.”

    Granted, AMD’s comparative metrics have not always been accurate, but at least an attempt to be accurate is made. And, is usually “pretty good.” It sure beats augmenting the pipeline to present a bigger MHz number, while decresaing the actual performance. Need I remind you all of the P-3 clock-for-clock against the P-4? A CPU doen’t get built by accident, and marketing campaigns don’t just happen either.

    There are lies, there are damn lies, and then, there are benchmarks. It always depends on tuning, this is a fact proven in just about every test run for either side’s claims.

    If it may not always be wise to take AMD at their word, who’s would you prefer? Another company that you can be absolutely certain is lying (and if you’re an Intel groupie, puh-leez, you’d be mad to even try arguing with that and you now it!)? At least AMD makes an effort to be comparative. There’s always a benchmark somewhere that will make it look wrong.

    It seems reason must always be anonymous. If your name was on it, would you be objective?

    • AMD64Baby!
    • 16 years ago

    You Intel guys are a bunch of idiots.

    Once you have used an AMD Athlon XP / 64, you never go back to an Intel. Intels are expensive, slow, flaky, and flat out crap.

    Intel should be embarassed that a 2ghz Athlon 64 can whip up on its 3.2ghz processor. And this is just in 32-bit. You’d be a stupid f**k to go out and pay double the price for an Intel that gives you the same performance as the AMD Athlon 64 3000+. And do you people realize that this Athlon 64 3000 gives you 3-3.2ghz P4 performance today, BUT, when 64-bit software comes out, you could get up to a 50% increase or more. You’d be f**kin stupid to buy pay double for the p4 3.2 if u knew any better.

    So, you people can slam AMD all you want. But I think its Intel who should be being slammed for their high prices, mhz brainwashing, misleading celeron performances. Intel should not be where they are today… If only more people were aware of AMD…..

    • Anonymous
    • 16 years ago

    Could you imagine what a TPI would do to Intel’s product line?

    Think Celeron. A Celeron 2.8GHz with a TPI of 1000+ would look truely tragic.

    • Anonymous
    • 16 years ago

    Hey thats great WaltC, instead of one huge long post post 3 med size ones in a row!!

    Only joking, but it does look funny in flat view.

    • Anonymous
    • 16 years ago

    The PR number is based on performance in a range of benchmarks and gives a general performance rating. we know what those benchmarks are and it would be easy enough for a site, like TR to test the claims.

    If you have a specific task in mind like DivX encoding then don’t use a general performance rating to make your buying decision use DivX. (that includes a MHz rating)

    The Athlon64 3000+ is slower than the 3200+ on the vast majority of benchmarks, therefore deserves a lower general rating, which it gets. The rating is equivilent to being 6% slower. In games this is consistant with the difference between the performance delta of the P4 3.0 and 3.2 (also 6% “performance rating” difference) at about 4-5%.

    In media encoding/3D rendering the scores are a lot closer between 0 and 4%, this would suggest that a rating of 3100+ is more applicable (PR diff 3%, performance difference 2%).

    AMD clearly decided to use the LOWER of the two options in order to not take flack for the chip not performing like a 3100+ in gaming situations and are still taking flack for the chip performing better than the name implies in certain situations in comparasion to its siblings!

    A similar situation has occured with the 3400 but nobody seems to have complained, this chip has had an increase of 10% clock speed (8% performance increase? can’t be bothered to look at that, can’t afford one!) and only 6.5% PR increase. IMHO it should have been a 3500.

    The fact is that chips with similar or even the same names have always performed differently (or even fitted different slots/sockets) P3 600, Athlon 1.0, P4 2.0, P4 2.4. Never mind the fact that 2 same “rated” chips from different makers could perform vastly different, K6-2 500, P3 500

    The PR system is not perfect but it is no worse than using MHz as a performance yardstick and is unfortunatley suseptable to the same problems.

    AMD had to do something to show the relative performance of their chips V intel for Joe publics benefit, I know somebody here thinks JP does not buy AMD anyway but, in the UK at least, thats not true, PC world the largest high street retailer here sells quite a few AMD PCs and Time (the 2nd largest?) sell huge numbers of AMD systems and JP given the choice between 2.0GHz and 3.0GHz will take the bigger number most times, How do you think Intel sells all those celerons?

    • Anonymous
    • 16 years ago

    tpi is a stupid idea because the amd cpus at a lower clock can be faster in some areas and slower in others. Therfore depending on what the consumer is willing to do with a computer, the tpi can be a good benchmark for one user and not for another user.

    For example i need cpu speed mainly for divx encoding where intel slaps amd in the face. So in my case TPI is bullshit, but for someone playing games it could have a meaning.

    Hence why no firm outside amd wants to support that TPI. it considers every users the same

      • Anonymous
      • 16 years ago

      You could say this no matter what you use to measure performance. Any time you use one number to indicate how fast a particular general purpose CPU is, there’s the risk that a particular task’s speed won’t be correlated well with the single performance number. This is true even for clock speed comparisons between chips with the same architecture – if I have a memory bound benchmark, a 2.4GHz P4 with an 800MHz FSB might very well be noticably faster than a 2.53GHz P4 with a 533MHz FSB.

      As others have pointed out, if you’re going to use a general purpose CPU for one specific task, it’s a good idea to measure the speed of that particular task. If you’re going to be doing lots of typical general purpose computing tasks, something like TPI would be more useful than raw clock speed, since you’re measuring actual performance in tasks you’ll likely use the computer for.

      That being said, there is a lot of room for discussion about what the correct suite of “typical general purpose computing tasks” is…

    • indeego
    • 16 years ago

    Take TR’s cachemem graphs and somehow add it to the end of a processor name. Sure it”s long, unweildly, and difficult to pronounce, but think of how easy it would be to tell the difference between processors and their performancesg{<.<}g

    • Anonymous
    • 16 years ago

    A universal performance rating will happen one day! Just after world peace, DNF in the stores bundled with a Bit Boys graphic card and my grandma is using linux.

    • Anonymous
    • 16 years ago

    If Intel was ahead of AMD in performance, I think they’d agree to a universal metric. Until that happens TPI is outta luck.

      • Anonymous
      • 16 years ago

      Good point!

      • Anonymous
      • 16 years ago

      Um, so how come TPI didn’t go into action sometime in the last year?

    • WaltC
    • 16 years ago

    Scott, in order for there to be a “performance metric” it is necessary that the major cpu suppliers all agree to it. That’s not negotiable. Simply put, if Intel won’t agree, and the record is clear that AMD has been trying to get Intel to agree for years to an official performance metric, then there can be none. And, of course, you can’t hold AMD responsible for what Intel refuses to do.

    You’d think, based on the fact that both its Banias and Itanium cpus are faster per clock than the P4, that Intel would see the inherent wisdom of establishing a performance metric which moves everybody off of the old, discredited “MHz” notions as to performance among cpus built on differing architectures. You’d certainly *think* so, as using the MHz metric (a left-over relic from Intel’s heyday as a monopoly x86 cpu supplier), isn’t suitable for convincing a potential Itanium customer that a cpu clocked 1/3 as fast as a P4 is 2-3x faster. I would think there’d be some difficulty there. A performance metric would explain it; a MHz metric simply cannot.

    So now that we’ve established that it’s Intel which is deserving of your ire instead of AMD, as it is Intel blocking the adoption of an industry performance metric and not AMD standing in the way, I think we have to consider that AMD will of course be cautious and careful about the application of its own performance metric, which Intel’s non-compliance is forcing AMD to apply unilaterally. Would it do AMD much good to exaggerate the metric for its own cpus versus Intel’s? I can’t see how, since it would surely be found out, exposed, and hurt rather than help AMD in the general markets. Note this is especially true in the server markets where Opteron is aimed–at professionals. So, I would think if any error is to be made that AMD would err conservatively, and *understate* its metric, if anything.

    Another way to look at it is that at least AMD has a performance metric, whereas Intel has none at all, apart from raw MHz numbers, which mean just about nothing when consumers are comparing cpus of differing architectures, and as I said that applies to Intel’s own family of cpus–AMD need not be mentioned to see the merit of a performance metric among Intel cpus. Basically, it seems to me that you’re upset with AMD because of something Intel refuses to do…:)

    From another point of view, however, the market has become accustomed to the idea that MHz numbers shouldn’t be used to compare performance among competing cpus of different architectures. Intel’s own Itanium has helped this conception along to no small degree, not to mention IBM’s G5’s, and of course AMD’s Opteron/A64/Athlon. So, truthfully, whether or not Intel adopts a performance metric is simply not as important to the market as it was a year or two ago, because the markets much more clearly understand when you can use MHz as a performance metric (only when comparing cpus of identical architectures) and when you can’t (when the cpu architectures are different.) I think some of the pressure relative to the markets confusing MHz with performance in the wrong ways is simply no longer present, and so in a sense it’s no longer as important whether Intel agrees to a performance metric or not, from the standpoint of Intel’s competitors.

    I really do not know what you mean when you talk about “trusting AMD” to get the performance metric right. When Intel throws nothing but a MHz number at you–what’s the good in *that*?….:) How is it that you figure Intel’s “MHz metric” is inherently more reliable than AMD’s “performance metric”…? It seems to me that a performance metric, even a unilaterally applied one, is more informative than a MHz number.

      • Anonymous
      • 16 years ago

      You’re forgetting Intel is the bigger company here. I wonder what happens to all of you retards spouting your mhz myth comments when Intel kicks the crap out of AMD. Companies aren’t the ones using MHZ as a way to compare themselves to other products, consumers are

        • Anonymous
        • 16 years ago

        Consumers, like you, are generally idiots.

        • WaltC
        • 16 years ago

        Not trying to be uncouth here, but it seems to me that…

        You’re pretty “retarded” yourself if you can’t see that Itanium runs in terms of MHz a whole lot slower than a P4, but runs software a whole lot faster at the same time. If you can’t see that, and you can’t understand it, then I’d stop any interests you might have in pursuing a technology career.

        MHz numbers are fine metrics to use when you talk about cpus built on the same architecture–such as when you compare one P4 with another P4 or one A64 against another A64 or one Itanium against another Itanium. MHz numbers, however, tell you *nothing* about how a P4 compares with an Athlon, or an Itanium or a Banias or a G5, and several other cpus I can name. You may not like that, or understand it, but it nevertheless is a fact which is indisputable.

        Edit: I wanted to add, as well, that I am as much a “consumer” as I am an employee of a company, and the sales records for Athlon in all its forms since its debut prove pretty conclusively that lots of “consumers” are capable of thinking in terms of a performance metric divorced from simple MHz clock numbers–lots of them. If that was not the case then Athlon would never have taken off–especially interesting when you consider that in 1999 the Athlon was launched into the consumer markets and not the “company” or “corporate” markets. In my opinion, Intel’s refusal to cooperate in establishing an industry performance metric will only disadvantage the company over the long term in relation to AMD, which provides both a MHz number and a performance metric for its cpus.

      • Anonymous
      • 16 years ago

      I always smile when I read your responses because they’re so fricken long. You literally write a page when a paragraph will do, let me sum up what you’re saying in easy to follow points for the folks with ADHD:

      1.) Intel is banking on the ignorance of the consumer to sell their P4s.
      2.) Intel does not want TPI because it may inform the consumer that MHz isn’t everything.
      3.) Thus Intel will not sign TPI so Intel should be to blame for the failure of TPI, not AMD.
      4.) Intel could stand to benefit from TPI to help sell more Itanium and Banias/Dothan processors.
      5.) Blame Intel, not AMD.

      There, my post is much shorter, gets the point across without the fluff. 🙂

        • WaltC
        • 16 years ago

        Heh-Heh…:) I don’t assume that everyone who might read a post of mine understands the basic issues involved, so I like to explain them a bit more fully as I see them. It’s like the person above who thinks that people are “retarded” if they don’t see MHz as a valid performance metric among different cpus–I provided concrete examples of even how MHz is inapplicable as a performance metric among Intel’s own cpus (Itanium/P4)–but this person still fails to grasp the issue. Plus, of course, when I wrote my own post I didn’t have the advantage of having it pre-written in front of me so that I could summarize it…:) But, point taken, and I concede that I have a tendency to wax prolific…:)

      • Anonymous
      • 16 years ago

      First, I’d like to say that blaming Intel for something AMD once promised to implement but failed to deliver to this very day is wrong. Heck, even announcing something that relies on the cooperation of your largest competitor is bad enough to earn AMD years of justified ridicule.

      y[

        • WaltC
        • 16 years ago

        Excuse me, but how could anyone expect AMD to “promise” anything on behalf of Intel? When people say, “AMD promised to set an industry-standard performance metric,” it is understood, or certainly should be understood, that voluntary industry participation within such a metric is a requirement, and that includes Intel. And, we don’t have a case where AMD and Intel disagree over performance metrics–we have a case of AMD wanting to establish *a* performance metric and Intel not wanting one at all. As such, it’s hardly surprising to see AMD continuing its own performance metric unilaterally. Certainly doesn’t surprise me, and it’s a heck of a lot more informative, I believe, than simply throwing out a MHz number.

        I read awhile back of some university boys putting together the “world’s fastest clocked cpu,” which ran at an astounding 500GHz or so. Problem was, it was neither useful or practical, as it executed something like 1/1000th of an instruction per clock (it was some incredibly low number–I can’t recall exactly), the net effect was that it would run current software slower than most current desktop cpus. It was more of a publicity stunt to call attention to the school than it was anything else, and the effort had zero technological value. But I thought it made the point very well that MHz alone is not a suitable performance metric for comparing the performance of cpus of different architecture.

        As to your comments about Intel’s approach to market segments, and the assumptions the company evidently makes about the general intelligence and knowledge within those market segments, I completely agree. However, it does not alter the fact that a MHz metric does not explain the performance differences between an Itanium and a P4. In fact, in that comparison, the “MHz metric” works exactly in reverse with the much lower clocked cpu being the much faster cpu between the two.

        So, the question remains: why would *anyone* wish to compare cpus of different architecture only on the basis of their MHz clocks? Just doesn’t make any sense, I’m afraid.

          • Anonymous
          • 16 years ago

          Certainly doesn’t surprise me, and it’s a heck of a lot more informative, I believe, than simply throwing out a MHz number.
          ———-

          Once again , you don’t know who the bigger company is here. AMD is not going to tell Intel to compare Intel’s own chips to theirs. Why is it a surprise to you that Intel doesnt want it? Intel can do fine without any crap confusing their customers because AMD told them to, Intel seems to be doing orders of magnitude better than AMD in every segment financially and market-share wise, without a performance metric. before you talk about AMD and Intel you need to go learn how business works

            • Anonymous
            • 16 years ago

            y[

            • Anonymous
            • 16 years ago

            y[

    • 5150
    • 16 years ago

    Life would be so much simpler without marketing hype.

    • Turkina
    • 16 years ago

    I cant believe the guy’s name is Mr. Speed, ROTFL

    • Anonymous
    • 16 years ago

    I don’t care /[

      • LiamC
      • 16 years ago

      LOL! I vote for Gerbilnator – Model 53. Much better than FX-53. Maybe TR could run a poll on it 🙂

    • daniel4
    • 16 years ago

    Why did I mark that as a reply to #2, hehe oh well :P.

    • Anonymous
    • 16 years ago

    Their model numbers are getting out of hand. I just wanna know how fast the damn thing really is and now what they think it is compared to Intel. I’ll let the review sites like this place tell me how the CPU really stacks up against the competitors offerings.

    Besides… who is AMD really trying to impress? Average Joe or Alpha Geek?

    Average Joe will always want to stick with Intel just because that’s what Average Joe knows. Most computers you see for sale at popular consumer stores are Intel based anyways. Average Joe won’t think twice about AMD because Average Joe wants a computer with “Intel Inside” because that’s what he sees on TV.

    Alpha Geek will at least take the time to do some research as to what is out there and how each CPU performs among each other. Alpha Geek will understand that MHz to MHz ratings is no longer a valid way to compare processors between companies. Alpha Geek will be the only one who will seriously look at AMD as a viable competitor to Intel. Alpha Geek doesn’t care for the lame model numbers AMD is associating with their CPUs.

    It’s like the posers who put V-TEC stickers on their Honda Civic when it’s not. Or the poser who put an ///M5 badge on a car that is clearly is 3-series BMW (not even an ///M3). Who do those posers think they’re fooling?? The people who would even care what kind of engine is in your car are the same people who would know that you’re a wannabe idiot poser for making it look like your car is something it’s not.

      • Corrado
      • 16 years ago

      Uhm, they really aren’t out of hand. Look at the benches in the 3000+ article. In fact, the 3000+ often bests the 3.2C, and if it doesn’t beat the 3.2C is in between the 2.8C and a 3.2C… with would be…. TADA! 3000mhz p4c. How much more accurate can you get?

        • Anonymous
        • 16 years ago

        Um, but it says its compared to the thunderbird, which would make the 3200+ AXP and the 3000+ A64 not make any sense at all

          • daniel4
          • 16 years ago

          The benchmarking suite is continuously being updated, you can’t possibly think that they would keep the same reference from year on year. The current suite would probably put an Athlon XP 3200+ @ around a 2600+, but existing products aren’t changed by the update in performance measurement. Either way one can still make a similar comment in as to why a 2.4C P4 is faster than a 2.4B P4. And plus will people end that damn Thunderbird crap, we all know it’s compared to a P4 and that Thunderbird comment is close to 3 years old!

      • daniel4
      • 16 years ago

      “AMD now apparently assigns Athlon 64 model numbers almost at will”

      I wouldn’t say that, they do seem to use a standard set of benchmarks in order to come by their model numbers. For the Athlon64 they use the most recent versions of the very same benchmarks that I see a lot of review sites. A nice TR article would be one that verifies whether using the same set of benchmarks that AMD uses will come by the essentially same conclusion that AMD has. Haven’t seen anyone do this yet, so it would be interesting. BTW everyone knows they don’t measure them towards Thunderbirds and that whole Thunderbird comment was something that was said around 3 years ago. Either way my article idea is super cool and you can find their list of benchies here “http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_9487^9500,00.html”.

    • DrDillyBar
    • 16 years ago

    I had completely forgotten about this. Good catch.

Pin It on Pinterest

Share This