Slashdot Mirror
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 AMD numbering system has never been directly related to Intel's (officially) but is instead related to the performance of older Athlons... in effect saying "The XP 2700+ is roughly equal in speed to an original Athlon running at 2700 Mhz"
Help Brendan pay off his student loans
Really, the technical community needs to sit down and figure out a universal cross-platform benchmarking method.
Well, there's SPEC and TPC. Other than that, benchmarks are both overrated and the best metric we have for evaluating performance. Then you have cases when a CPU is optimized for a particular benchmark to inflate performance numbers (hence the term benchmarketing).
Doesn't FUD imply it's untrue? If you don't like AMD (I don't, they just had a convenient explanation to link to), it's a similar situation with Apple, though they have a different architecture entirely.
Auto-reply to ACs: "Truly, you have a dizzying intellect."
Actually its not the origional Athlon. Its the Tbird that they compare it to.
Moo!
Older intel CPUs used a performance metric named iCOMP which was stamped on many CPUs. A bit of googling suggests this is still around. Perhaps this is another case of reinventing an old idea?
From what I understood, AMD got the numbers by comparing itself to the latest Pentimum chip running at that frequency
You understand wrong. AMD Performance ratings are as compared to a Thunderbird core Athlon. In other words, a "PR 3200+" chip is eqivilent to a Thunderbird running at 3.2Ghz, and not a 3.2Ghz Pentium 4.
What part of "shall not be infringed" is so hard to understand?
I think the article got this wrong. If you read the Anandtech Report, they believe it is going to be an Opteron-ish number scheme, not an AthlonXP-ish one.
Quote from the report:
News broke earlier today that Intel will most likely change its current "Megahertz" strategy in favor of a more subdued "Model Name" approach. This does not necessarily mean Intel will change its processors to a PR rating, like "3000+". Rather, the new model system sounds very familiar to AMD's Opteron approach, with three or four digit numbers replacing the product name.
The unofficial
Intel has NEVER stated that Mhz equals performance. Go look on their site. They like to quote SPEC, and their own performance tests, and all the rest of the BS that companies do. Never Mhz. The Mhz myth comes from two places:
1) Fanboys. I first remember it gaining real popularity among the Apple fanboys when Apple went PPC. They claimed that the PPC showed a positive second derivitave (growth of growth) in Mhz where Intel showed a negative second deravitive and how PPC could scale to huge speeds that CISC just couldn't handle. That of course, neve came to pass. Which lead us to:
2) The anti-Mhz myth. That Mhz don't mean anything. This is just FALSE. When you compare a single architecture (meaning one kind of one brand of processor) mhz give a VERY good idea of how performance will scale. If something gets X on a processor at 500mhz, you can with confidence say it will get nearly 2*X with the same kind of processor at 1000mhz. 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.
PR numbers are just a bunch of crap. So far, I've never even seen any that are reliably based off of benchmarks. Even if they were, it wouldn't matter. Show me any benchmark, I'll show you how it's not relivant to things a lot of people do. Like take SPEC. It is a big industry standard benchmark. People doing scientific and engineering work place a lot of faith into it since it benchmarks what they do.
Well Intel LOVES SPEC, their processors when mated with their compiler do very well at it. Does that mean we should use it? Hell no. SPEC isn't applicable to everyone. It's got nothing to do with games, audio, video, bussiness, servers, etc. It's a science and engineering benchmark. What's more, it's a benchmark designed to come form source code, so to bench the compiler as well as the system. It's a good, open, standard benchmark, but it won't work as the single number to completely describe chip performance (nothing will).
PR numbers improve nothing, and just confuse and BSify the situation. At least Mhz are factual numbers and have some basis in reality. From what I've seen of PR numbers, they are mainly a dream of marketing and don't apply to the real world.
Uninformed consumer goes to the local discount electronics store. Looks at a computer based on Intel's CPU, sees 2000 megalobangerz. Looks at the AMD based computer right next to it, sees 1800 megalobangerz for a hundred bucks less. Decides the Intel is "better", so its ok to cost more. Reality, the computers are pretty much the same.
AMD did this becuase their chips simply do more work per clock cycle, this was done at the expense of not being able to scale the clock nearly as high as Intel. A 2000+ AMD is *roughly equivillent* to a 2.0Ghz P4.. it wins some, it looses some.
The jump you talk of was at 2600+, when AMD went from a 2.0 Ghz at 266 mhz FSB (called a 2400+) to a 1.833 Ghz at 333 mhz FSB, called a 2600+ Barton. Performance #s goes up, clockspeed goes down.. but FSB goes up! Yes, it's annoying, but this was done as to give most consumers who do minimal research a "fairer" basis for comparison when shopping for computers.
MHz is an absolutely useless metric for comparing processors today when FSBs range from 200 mhz to 800mhz and cache from 128kb to 1MB and higher. Intel and AMD went different routes when designing their offerings, and as you say, it's very difficult to come up with a single number to describe their performance. The problem is that MHz is the number that has been 'historically' used, and it just so happens that AMD went the route that yielded a smaller MHz (and god bless them that they did); so they made the transition to a BS-marketing-numbers system.
DJ kRYPT's Free MP3s!
If you don't get this joke, you haven't been here.
and please mod this person up. (S)He is correct in stating that the AMD model numbers are derived NOT from the Pentium 4, the Athlon classic, the Centrino, Celeron, PIII, Crusoe, 8088, or any other God-forsaken chip, but from the Thunderbird core Athlon CPUs. Those were the last Athlons to advertise the clock frequency, and thus were the obvious choice for a comparison chip for the next generation of processors. If I just bought a 1.4GHz Thunderbird Athlon (common chip for the time), I would expect that an AthlonXP 1500+ would perform better than it, and I would be correct. An AthlonXP 1500+ under the new rating system, were it to be compared to the Athlon classic core (far less efficient than Thunderbird) would probably run at about 1.1GHz. As it is, the AthlonXP Palomino core 1500+, being a relatively minor revision to Thunderbird, ran at 1.33GHz.
So mod this guy up. He's right, the post he's replying to is wrong.
Have a nice day.
-- "Government is the great fiction through which everybody endeavors to live at the expense of everybody else."
here's one of many found on google
I should also note that last year when AMD tweaked the formula they ended up with a CPU that had a higher PR number than the previous model, but was slower in realworld performance.
Try as you might, you will never find any AMD benchmarks that backup the 1GHz Thunderbird myth.
Because the baseline isn't an Intel chip? The baseline for AMD's pro-rating scale is a 1GHz Duron. IE: A 3200+ is 3.2 times faster than a 1GHz Duron.
Doesn't FUD imply it's untrue? No. FUD is fear, uncertainty, and doubt. It's what results when a fact is unknown, yet decisions have to be made based on that fact. If we knew that the FUD-source was false, it wouldn't produce any FUD. Unfortunately, we can't be so sure.
How about an explanation of the numbering system right from the horse's mouth
mirror
from AMD's athlonXP site (doesnt' seem to be working right now)
web archive of AMD's site
My server
Now that's a trend I think is broadly continuing. Multi core CPU's are a part of it. We may also see async processors coming out with zillions of transistors, but no central clock.
That's nothing, the adc and sbb (add and subtract with carry) instructions take 8 cycles! That is 8 times what it was like on the Pentium 3. At least the add, sub, xor, not, and, or, neg instructions take only 0.5 cycles.
Not Duron, but older Athlon Thunderbird. And it does not mean "3.2 times faster than Duron", it means it is fast as 3200MHz TBird would be.
http://en.wikipedia.org/wiki/PR_rating
If we knew that the FUD-source was false, it wouldn't produce any FUD.
Even if "we" know that the FUD-source is false, the targets may not know. Often, "we" regular Slashdot users are not the targets for specific negative advertising campaigns. Rather, companies aim for the PHBs who control purchasing in large enterprises. PHBs seem to respond more readily to commercial attack ads than do those who actually use the products in question.
AMD have already started an entirely new naming scheme anyway with Athlon64 FX-51. Presumably they're going to go FX-52, FX-53, etc.
And if you wanted AMD to use the MHz in their marketing, then I suspect it wouldn't be much better, for instance with the AthlonXP 2600+, there were two different clock speeds which ran at about the same real speed.
Karma: It's all a bunch of tree-huggin' hippy crap!
Why does this incorrect info keep getting posted (and modded as "informative" at that)? AMD stated several times quite publicly that their rating initially was meant to compare against the "Thunderbird" Athlon chips. More recently they've simply said that it's relative performance between the AthlonXP line and that it can "outperform it's closest competitors". Here's a direct quote from AMD's AthlonXP FAQ
Q: What does the 3200+ model mean?
A: This is a model number. AMD identifies the AMD Athlon XP processor using model numbers, as opposed to megahertz. Model numbers are designed to communicate the relative application performance among the various AMD Athlon XP processors. As additional evidence that performance is not based on megahertz alone: the AMD Athlon XP processor 3200+ operates at a frequency of 2.2GHz yet can outperform an Intel Pentium(R) 4 processor operating at 3.0GHz with an 800 FSB and HyperThreading on a broad array of real-world applications for office productivity, digital media and 3-D gaming.
AMD's model numbers not rated against Intel's P4 chips? You might want to tell AMD that!
Xerox PARC did the R&D for modern GUIs. The Lisa was Apple's first attempt to copy the Xerox PARC GUI work, and it failed. Then, Apple tried again with Macintosh, and by cutting a lot of corners made the system cheap enough to make it a success.
I have seen the early GUI development by PARC. MUCH more R&D was required to get that concept up and running for a machine that could serve as a "personal computer." Yes, the Lisa failed, but it was the first personal computer that had a GUI.
The laser printer was developed at Xerox PARC. Postscript was developed at Adobe, based on a more complicated PDL developed at Xerox PARC. Apple just happened to create a successful product based on those technologies.
PARC "invented" the laser printer, but it was Apple who heavily underwrote a new company by the name of Adobe and co-developed the laser printer for use with the personal computer.
The Psion predates the Apple Newton by nearly a decade, and I think it wasn't the first PDA either.
I'll give you that technically, but I used an early Psion in 1986 or so and it was not really a functional information manager. The Newton 120 that I owned a couple of years later was a true PDA that allowed for word processing, information management, communication for email and early Internet via modem and IR, and more. The Psion was more of a glorified address book or flat data file keeper.
Not even close; you can find the history of the laptop here. In fact, the idea goes back to Alan Kay's work on Dynapad--late 1960's or early 1970's.
Laptop form factor!(not laptop) with palm rests in front of a full sized keyboard with trackball or (later) trackpad was the innovation there. All of the previous laptops I have owned have been awkward with keyboards up front with no place to rest your hands and no pointing device integral to the laptop.
The Apple II was irrelevant to speech recognition research and development
My point still stands, that the first speech synthesis was developed years before anybody else on the Apple ][.
Not even close. (Digital Camera)
Consumer digital camera! is what I said. I remember the MavicaPro series and they were hideously expensive. The Quicktake was actually affordable by the consumer.
You other examples either refer to system integration issues (e.g., supposed first use of a 3 1/2" floppy--developed by Sony), or are vague and meaningless from a technological point of view.
Hey, I remember installing Microsoft Word or Office using a skyscraper of little floppy disks and I for one, am grateful that Apple began shipping computers with CD-ROM drives in them for just this reason.
Plug and play compatibility is something that is also a huge time saver. Do you remember setting all of those damned DIP switches when installing a video (or other) expansion card and constantly rebooting every time you changed something? Come on now, even now with a modern SGI Octane, when I install a new video card (or other expansion card) I am down for at least a half hour configuring things. Plug and play revolutionized the personal computer industry.
First to include built in networking is meaningless? There is this thing you are using called the Internet.........
Firewire is meaningless? I guess you don't use any significant amounts of data.
Look, don't get pissy and there is no call to be offended. I am simply giving credit where credit is due. I grew up using TRS-80s, IBM PCs, Heathkits, Apple ][s, Sun Solaris boxes, SGI IRIX boxes etc.... and I find Apple really does make the easiest to use yet most flexible kit. OS X pushes that flexibility even further.
Visit Jonesblog and say hello.
Gigaflops is only a tiny fraction more useful than GHz, if at all.
Gigaflop tests come in three basic varieties. First are ones that fit entirely into the L1 cache of a processor, making the memory subsystem totally irrelevant. This is no good since the memory subsystem plays an important role in performance. In this sort of test a 2.8GHz Celeron processor with 128K of L2 cache and a 400MT/s bus speed would get a score essentially identical to a 2.8GHz P4 with 512KB or 1MB of L2 cache and an 800MT/s bus speed. In 90% of real-world applications though even a much slower 2.0GHz P4 would beat the pants off a 2.8GHz Celeron (the current Celeron chips are absolutely abysmal perfomers).
The second type of gigaflops test has a slightly larger dataset, so performance is almost entirely determined by what level of cache it fits into. For example, if they used something like a 60K dataset, an AthlonXP or Athlon64 would blow the doors off any P4 because it would be running everything in L1 cache while the P4 would be running out of (the much slower) L2 cache. Clock for clock the AthlonXP chips could easily be twice as fast in such a test. Things would get even worse if your data set fit into the L2 cache of one chip but not another, ie if you had a 750K data set, a "Prescott" P4, with 1MB of L2 cache, could be HUGELY faster than a "Northwood" P4 with only 512KB of L2 cache, even though in reality their performance is fairly close (with the "Northwood" usually being slightly faster).
The third option would be to use a HUGE dataset, turning this entirely into memory bandwidth test. Fine for what it's testing, but hardly an accurate picture of overall performance.
There are good reasons why the rather smart guys over at Ace's Hardware make use of Linpack (basic Gigaflops test used by Top500.org) to show off the memory subsystem of platform. By varying the size of your dataset it does a good job of illustrated the effects of cache and memory. However it doesn't tell you much else about processor performance.
I think that gigaflops would be a slightly worse metric for processor performance than MHz because it's FAR easier to abuse that test. The best thing for consumers is if the model numbers are really NOT meaningful at all. For example, look at video cards, where our top-dogs today are the ATI Radeon 9800 and the nVidia GeForce 5900. Nobody looks at those and says "Ohh, 9800 is bigger than 5900, therefore the ATI MUST be better". Everyone KNOWS that the model numbers here are meaningless, so if they want to know which is faster they ask a friend (or at least the salesperson) or do some research on their own. That is what I would like to see for processors as well. AMD's already got this with their Athlon64 FX line and Opteron line of processors. Hopefully Intel will do the same.
"The problem is that AMD and Intel custom design their chips to perform better at different tasks/instructions. Then there is the problem of compilers. Was the SpecIntBase compiled with AMD and/or Intel specific instructions? Which versions?"
If they can get a better score on SPEC by using another compiler, go right ahead. You're meant to compile the spec suite for the computer you're testing anyway.
Wether or not specific applications take new features into account is really not relevant to baseline system benchmarking. That one is, as always, something the customer has to take up with his application provider and research himself.
Intel is quieter?
You claim to know more about hardware than "Tom's" and you comment on a CPU/platform being quieter?
Go buy yourself an Antec Sonata computer case with an AMD CPU in it and tell me how loud it is.
Loudness is a feature of fans, airflow and vibration, not a CPU.
If what you're trying to insinuate is that AMD chips run hotter and therefore need better cooling, that may or may not be true at a given performance-point, but say so or you come off being to inexact to be able to back up your statements.
While you're at it, pick up one of their copper-based CPU Coolers and maybe their heat-sensitive SmartCool case fans.
- Michael T. Babcock (Yes, I blog)
Not really. FLOPS is an attractive measure when comparing machines with different ISA's (Instruction Set Architectures), because they tend to be more constant than integer instructions. This is especially useful when comparing RISC and CISC processors, as RISC processors tend to execute more instructions per second but each instruction does less. But a floating-point add is likely to be the same on both machines, and for something like a linear-algebra problem, it is possible to compute the number of FP ops executed, and this will likely be the same for all machines. (It gets tricky when you start comparing machines with FP divide instructions against machines that require emulating FP divide with an inline routine that takes several FP multiplies, which is why such apps are generally not used for these comparisons.)
But this is not very useful when comparing different versions of the same ISA. And FP performance is just one component of overall system performance. A system with a slow bus is going to suck on anything that isn't lucky enough to fit in the CPU's caches.
Supercomputer users have been aware of this for years. The large US supercomputers build with thousands of multiprocessors would have impressive teraFLOPS ratings when they multiplied CPU's by peak FLOPS/CPU (what you could get if you could run every FP unit on every cycle), and get reasonably good ratings on their Top500 scores (because Linpack is relatively "friendly"), but on real apps, they'd call it a good day if they could get 10% of the peak rating.