Intel Plans CPU Naming Change

Jemm writes "According to The Globe and Mail, Intel will start using performance numbers rather than clock speed to number their chips. 'Under the model number system, processors will be given numbers to describe their performance, in addition to being described as running at 2GHz or other speed.'"

The Megahertz Myth

[Liselle]

Good news for the average computer idiot who wants to upgrade or buy a new machine. I think it's past time to undo the damage Intel's marketing has done with the Megahertz Myth [amd.com]. I'm weary of explaining it to people. It will be nice to have something more helpfully descriptive to a consumer than "cache" and "bus", or at least clarify that they don't refer to paper money and vehicles that carry children to school. :P

Follow the leader

[oingoboingo]

FIrst Intel adopts the x86-64 ISA in their new chips, and now they start using performance ratings. What next? Jerry Sanders to replace Craig Barrett as CEO? How times have changed.

So how will AMD name their CPUs now?

[Anonymous Coward]

Their naming convention will be 2 steps away from gHz performance now!

Presenting the AMD XP 5500+, which runs at 4 gHz, but is equivalent to a Pentium V 5.5EE, which is equivalent to a 4.0 gHz!

What are they going to compare to?

[GarbanzoBean]

From what I understood, AMD got the numbers by comparing itself to the latest Pentimum chip running at that frequency. Now, what is it going to be. AMD 128 at 100GHZ has performance 150000+ measured in units of Pentium X that has performance 50000000+ of Pentium 9 running at 1THZ.

Seriously though, the perfomance numbers are beginning to be as confusing as the speed numbers. In the end it is what you "feel" gives you a better performance. Or more scientifically, which benchmarks you choose to run to fit your expectations.

Perecursor to a change in design strategy?

[Random BedHead Ed]

Intel has long coasted along on what Apple likes to call the "megahertz myth." The power of a processor is more than just its clockspeed, as Apple and AMD have struggled to point out for years. Intel ignored the debate because they were ahead in clockspeed, so it was a convenient metric that always showed them to seem ahead of the competition. This change in CPU naming might indicate a recognition that its rivals may overtake it in clockspeed. Perhaps they're planning strategic changes that could take them below Apple or AMD in clockspeed and want to jump on the "clockspeed ain't everything" bandwagon as soon as they can.

You can't "measure performance" with one number

[0x0d0a]

The problem is that you can't measure processor performance with one number. There's just no way to do so.

Before, AMD and Intel used to use clock rates. They didn't pretend to actually be summing up their chip's performance with the metric they slap on the box. It was even okay when just AMD had a performance number, because there was no sense of putting an industry-wide metric on a box. Now, one of two things will happen:

Possibility 1) AMD and Intel will decide upon a standard benchmark suite to determine "performance" and processors will be optimized around that benchmark instead of around real world software (i.e. consumer loses).

Possibility 2) AMD and Intel will come up with *different* measurements to determine their "equivalency number". AMD will focus on chip feature X and Intel on chip feature Y, each probably choosing the one that best supports their case. Both will accuse the other one of using an inaccurate and artificial metric. Each one focuses on improving their score in their chosen test. The performance profiles of the two chips diverges more. Since most software must be least-common-denominator, all developers except those few that choose to include custom-compiled or assembly bits and processor-specific support will make software that runs slower on average. (i.e. consumer loses).

I liked it much more when Intel and AMD's marketing departments stuck with slapping stupid stickers on boxes and making deals with OEMs -- neither one directly affected me.

Bad for the consumer? Bad for some

[Dachannien]

With more than one company providing relative performance indices as "names" for their processors, and none really providing a basis for these relative ratings, the consumer will now be forced to rely on product review sites like Tom's Hardware or Anandtech to evaluate the real performance of processors.

That's a good thing in as much as the numbers will stop meaning anything to those with the technical know-how to get useful information from Tom or Anand.

But there are a lot of Stupid People out there using and buying computers every day, and they will be completely in the dark when it comes to evaluating their choices. For them, the deciding factor when choosing a processor in their premanufactured desktop machine will be only what a further descent into Marketing can tell them. ...Which is probably exactly what Intel wants.

Problem..

[Chordonblue]

When Intel abandons this scheme, what precisely will a 4500+ processor actually mean? It's bad enough trying to quantify it now, but at least we have the actual P4 GHz to compare against.

Something will clearly need to be done - independant benchmark-wise - to prevent abuse. It's going to get bad folks.

The good news: I think we're going to see '5000+' processors before the end of the year now.

The bad news: They will run like 4 GHz models.

What's old is new again

[fuzzy12345]

My Pentium(TM) Family User's Manual, Volume 3: Architecture and Programming Manual shows, on the front cover, a hand holding a chip marked "intel pentium iCOMP(TM) Index=815 (m)(c)INTEL '92 '93

It is either a 90 or a 100MHz part, don't know which.

The practice of inventing a silly(TM) performance index that looks better on your chips than your competitor's, or can't be used without a license, is pretty old.

Re:Payback

[MC_Cancer_Pants]

I like this explination personally. Very technical but try and keep up

Two children are playing on a beach, filling up a plastic pail with sand. The first child uses a teaspoon to scoop sand into the pail. The second child uses a much larger toy shovel, moving a great deal more sand with each scoop and working more efficiently.

The same concept also applies to processor performance. A computer with a processor that does more work per cycle, like an AMD Athlon processor, can out perform the same computer with a less efficient processor

Well then...

[Loki_1929]

Guess the rumours of Intel's problems with 90nm, Prescott's severe ramping problems, issues that even 775 can't solve, and the incredible heat dissipation of the newer chips are all true. This seems to be yet more confirmation, even moreso than the release of 2.4GHz Prescott chips this week. Gee, boys, guess we should have listened to Bob Colwell when he was standing around screaming about the unsustainable clock ramping and heat dissipation curves.

When the architect of the P6 says something, you usually ought to listen. Perhaps next time you'll get off your high horses and follow the suggestions of the smart people. Now he's gone, you're fucked for '04, and you're in serious trouble on the desktop front if Tejas doesn't turn out to be a rabbit out of a hat.

Pentium M

[PhotoBoy]

This seems to bear out the rumours that "the next big thing" from Intel on the desktop will be based on the Pentium M which is a chip which ably demonstrates that more Megahurtz isn't necessarily better.

I guess Intel is starting this change in numbering early so it doesn't debut a new chip and a new way of labelling the speed of the chip at the same time. Launching both at the same time might look suspicious to less informed buyers, especially if Intel goes from selling 4Ghz chips to 2.4Ghz chips with a PR of 4500+. By starting early hopefully people will be more accustomed to the new numbering scheme and less likely to think they are being conned. A friend recently told me he had bought a new 3Ghz Athlon XP, he was ready to take it back to the shop after I explained what the 3000 meant!

I wonder how compatible this will be with AMD's PR ratings? What would the equivalent to an Athlon 64 with a PR of 3400 be? I hope Intel doesn't invent a PR system that deliberately uses bigger PR numbers than AMDs. I can see confusion amongst consumers who will think an Athlon 64 4000+ is not a match for a "Pentium 5 6000" even if they are equivalent performers.

While Megahurtz has long been a poor way of determining the speed of a chip, I think having two different PR systems that aren't compatible could be worse.

Re:The Megahertz Myth

[Shichinintai]

Does AMD's "Megahertz Myth" page sow Fear, Uncertainty, or Doubt?

No, I think the word you're searching for is propaganda. And despite it's negative connotations, propaganda is simply, "An organization or plan for spreading a particular doctrine or a system of principles.". So while all FUD is propaganda, not all propaganda is FUD.

I really hate this "PR" crap

[Sycraft-fu]

I think VIA started it, but I'm pissed at AMD for continuing it, and now Intel for jumping on board. Mhz are a useful and TRUTHFUL stastic. It tells you how fast a given chip cycles at. This is a fact, not a bunch of marketing BS. Further, for within chip comparisons, it is a useful number. For example:

I have a P4 1.6ghz, I know that the max my board supports is a P4 2.4ghz. Supposing I want to upgrade, how much speed will I gain by maxing my processor? Answer: A bit less than 150% of my current performance. When all else is held equal about a chip, performance scales slightly less than linear. So if you need to double you performance, you need to a bit mroe than double your clock speed.

But PR numbers seem to just come out of the ass of marketing people. When AMD first went to their PR system they claimed it was based off of some benchmark comparison to their old Athlons. In reality the formula was increase the PR number 100 for every 66mhz in actual clock increase. This, of course, meant there PR numbers become more and more BS the higher they went. Chips can get, at best, a linear imporvement out of clock speed increase. It is simply physically impossible for a doubling in clock speed to result in more than a doubling in performance without an architecture change. I also recall when AMD moved to a new core, I think with the 2800+, that for a lot of things ended up being slower, hence making the PR seem even more like BS.

There just isn't a singular way to measure chip performance. Different designs are good and bad at different things. What's more, it depends on how something was written and compiled. Some apps may be well optimised for Intel processors, not for AMD, so they seem to run slower than numbers might suggest on AMD chips.

At least with Mhz you have a real, factual, non-BS number that is useful for internal comparisons. PR numbers just turn it into total shit and confuse the situation.

Re:Problem..

[afidel]

Well since all the GHz measurements from AMD are vs an equivilant performance Intel chip, and now that reference is gone, why not use the industry standard benchmark. Name chips based on the SpecIntBase score and be done with it!

Egg Grading

[VoidEngineer]

What I find interesting about this article is the inherent variability inherent in the way that modern chips are made.

For those of you less familiar with how chips are made, there is a standard sized silicon "wafer" which Intel uses... I forget the exact diameter, although it's round and about the size of a large diner-plate. Anyhow, it comes as a large cylinder, and they slice off diner-plate sized wafers, and try to fit as many chips on it as possible.

Now, making a chip involves lots of chemical-etching and photo-chemical reactions using ultraviolet light. The interesting thing about all of this is that they'll print hundreds of chips with each go, and each print doesn't create the exact same patterns. It's really alot like using an old typewriter... Ever notice how one of the keys might get bent or out of alignment and it types letter's inconsistently? Same thing happens with printing chips, apparently.

Anyhow, because of photonics angles, chemical flow dynamics, atmospheric pressure, and all sorts of odd little variables within the clean room, the chips are variable, even though they're printed from the same wafer. In the end, a 2.0 Ghz chip may have come from the same wafer that a 2.2 Ghz chip, or even a 2.4 Ghz chip (for example). As I understand it, chips from the outer edges of the wafer are more likely to be slower than ones in the center (increased angle from the lasers, chemical and atmospheric turbulence effects from the edge of the container, etc.) Apparently, the technology is getting to the point where slight changes in entropy within the chip production process will get magnified into performance differences in the end product. Butterfly effect of sorts, actually...

In the end... it's the same chicken producing eggs, but sometimes the eggs are different. And the eggs eventually get graded (A, B, C, etc).

note: I've never worked in a chip production facility, so my post is bound have some technical errors in it. Feel free to supplement my post; try not to flame. Just paraphrasing other articles I've read about the process...

I'd like to see an "open" designation

[Stonent1]

How about we as the technical computer consumers come up with our own designator? We could start by basing it on a known quantity, for example a 1GHz P3 with a 133MHz bus. Then we benchmark the different parts of that CPU. FPU intensive, Integer intensive, MMX intensive, SSE intensive, cache hit intensive, cache miss intensive, and a mix intensive. Then whatever score is produced is weighted and collectively called 1.00 Then from that point on all CPUs are to be referred to by their number based on their weighted scores. So perhaps a 2GHz Pentium 4 is only a 1.5 when compared to the P3. Or even better, I'd love to see the individual scores of the different sections. I'd like to make it really easy for people to get specialized processors that best suit their needs. In some cases, it is hard to determine what would be the best cpu for the application. You may need one that can fly through compiling software but you don't really give a crap about SSE, MMX or FPU.

This may suggest that Moore's law is at it's end

[mc6809e]

When I saw Intel was doing this I immediately thought "that's the end of Moore's" law. Intel has been trying to win the clock rate race for years. But, consider there newest Pentium, Prescott. This chip now has a 31 stage pipeline and is built for high clock rates. Yet, it still is clocked at less than 3.2 Ghz -- the highest speed of the older Northwood. Why is this? Even the earliest Pentium 4s were able to greatly out-clock the pentium III's when they first came out. They weren't faster overall, but did have higher clock rates than the PIII. But now we have the 31 stage Prescott and the about same clock rate.

If Intel thought it could keep bumping the clock rate up, they wouldn't move to something like AMD's performance rating. Yet here we are.

Something has changed.

Re:Does anybody remember iCOMP?

[Roguelazer]

Original [ideasinternational.com]
A very good explaination about Intel's iCOMP benchmarks can be found Here [cpuscorecard.com]

However, all of Intel's recent benchamarking references to "SysMark 2004" results.

Is reading comprehension a skill lost on ./?

[Sycraft-fu]

I quote myself (emhpasas added) "That doesn't mean it's the be-all, end-all benchmark, just a useful (and truthful) was of evaluating CHIP performance within a line."

I KNOW that the chip isn't the only thing in a computer. There is a reason why I'm still running a 1.6ghz P4, I spend my money on other subsystems since for me, they are the ones that make the most difference. However when evaulating CHIP performance specifically when evaluating, again quoting myself "a single architecture (meaning one kind of one brand of processor)" Mhz is an effective comparison. A P4 Northwood at 2.4ghz on a 400mhz bus will be able to do calculations roughly 150% the speed of a P4 Northwood on a 400mhz bus at 1.6ghz.

Now if you compare different bus speeds (533mhz vs 400mhz) different architectures (Northwood vs Prescott) or ESPICALLY wholly different architectures (P4 vs Athlon) it breaks down. But SO DO PR NUMBERS! There is NO gaurentee, and in fact a high degree of probablility, that AMD and Intel will have DIFFERENT BS schemes that have nothing to do with each other and less to do with reality.

I am not saying that Mhz is the ideal benchmark. I am saying that it is turthful and facutal and useful in limited in-line comparisons. PR numbers are the dream of a marketing department and have shit to do with shit and are worthless, even in comparing like chips.

Flops

[nycsubway]

That sounds similar to how AMD names their CPUs, and frankly I never understood what they really meant in terms of one being better than another. How about giving the power of a CPU in gigaflops?

While no measure can be truely accurate, the number of floating point operations a CPU can do per second is a more accurate judge of cpu power than the clock speed.

I'm glad Intel is choosing to use a different naming convention, hopefuly it will be something more meaningful.

Re:Payback

[BWJones]

Apple isn't developing the PowerPC, IBM is.

Believe it or not, there is this alliance called AIM. It used to be Apple, IBM and Motorola, but given Moto's problems, they have essentially dropped out for the embedded market. At any rate, the G5 was very much co-developed by Apple and IBM with some chip design and fab positions solely at Apple.

Apple is basically just an upscale systems integrator.

Without getting too much into the oft hashed out facts, just think about where the computer industry would be without Apple to do the R&D? I am not saying we owe everything to Apple Computer, but think about what you are saying before you type. Off the top of my head, here are a few things we owe to Apple: 1) Integrated motherboards consolidating most functions into a few chips with the Apple ][, 2) Plug and Play compatibility with NUBUS, 3) GUI with the Lisa, 4) First to use small form floppies with the Apple ][c, 5) First to implement CD-ROMs with Macintosh, 5) First to support on board sound and graphics with Macintsoh, 6) First to include built in networking with Macintosh, 7) First to develop the laser printer and postscript printing with the Laserwriter, 8) First to develop the PDA with the Newton, 9) First to develop the laptop form factor as we know it with the Powerbook, 10) First to leverage the GPU for routine interface with OS X, 11) First speech technology with the Apple ][, 12) First virtual programming environment with Hypercard, 13) Developed Firewire, 14) First company to ship a consumer digital camera with the Quicktake, 15) First cross platform standard for multi-media with Quicktime, 16) The first "multimedia" PC with the MacTV that integrated a television with stereo CD back in 1993 or so. We could go on and on here, but you get the point.

Apple's Xserve and G5-based machines are niche machines and they don't really offer compelling performance advantages

There is a reason that the number three supercomputer in the world right now is made up from off the shelf G5 hardware. It provides the performance for less money than the alternatives.

And OS X is severely handicapped in the market relative to Linux and Windows--OS X just isn't used very widely as a server operating system.

Well, that depends upon what you mean by handicapped. Marketshare? Sure. Useability? Not on your life. I've used Solaris, IRIX, Linux, Windows and others and nothing comes close to how easy, secure and convenient OS X is to administer for servers. Even the base desktop OS includes Apache that is as easy to use as dropping your html into a folder and pressing "Start" to function as a webpage and it can handle the traffic with the best of them. In fact, I am running a retinal anatomy site on an old G3 iMac that gets upwards of 45.000 hits/day from about 3000 unique users. The site is multimedia rich and yet, I never have to worry about it. When it was being hosted on W2k, I was constantly screwing around with it to keep things up and running smoothly and when it was on IRIX, it was stable, but IRIX was expensive and arcane as can be whenever changes were needed.

But the threat to Intel is AMD, not PPC.

Give it some time as the G5 really just came out. Between Apple running OS X and IBM running Linux shipping on systems now with the G5, there is going to be some significant market share being gained by those two companies.

bits

[minus_273]

will they start takling about the new revolutionary 16, 32, 64, 128 etc bit systems? Will we finally see the Intel64?
this will bring isssues like if NEC comes out with a system that has 2 128 bit processors, is it a 256bit system? Bitness is where it is folks. Im'm glad Intel finally got it.

In addition

[metalhed77]

AMD never kept thier spec numbers close to intel. They just wound up that way. Their benchmarks were all based off performance relative to a duron 1000 mhz.

Re:This may suggest that Moore's law is at it's en

[Weirsbaski]

Something has changed.

Intel has changed. Pentium 4 was specifically designed to have high frequency: performance-per-MHz was a secondary requirement. But now, intel is in the early stages of designing their next-generation part, and they have two choices- even-higher frequency, or lower/same frequency but better architectural performance.

I suspect they found out (or are finally starting to admit) that pure frequency doesn't buy as much performance as people thought, so now they have to fight the inertia of their own "GHz is king" mantra.

Re:This may suggest that Moore's law is at it's en

[danila]

If Intel thought it could keep bumping the clock rate up, they wouldn't move to something like AMD's performance rating.
I hope I do not sound extremely naive, but I like to think that Intel is not led by marketing people. And Intel's engineers do not directly care about selling more chips, they care (I hope so) about making ever faster (for actual applications, all of them) processors. Thus if they decide to concentrate on other things than upping the frequency for a while, this is probably a sound technical decision. The best Intel's marketing can do is reflect this good decision in a better performance metric.

Intel could have increased the GHz, but if they decided another approach is better, I tend to believe them.