AT News Update: Removing the MHz Myth, Again
by Kristopher Kubicki on March 12, 2004 10:03 PM EST- Posted in
- CPUs
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.
Further sources claim that Intel isn't dropping the Megahertz convention entirely. Rather, it is simply adding a model name to other specifications as part of a new approach to customer awareness. Front Side Bus (FSB) speed, cache size and core name will play a larger role in differentiating one product from another. Unfortunately, this will not be an easy task as the last 10 years of microprocessor marketing has been almost exclusively "MHz" oriented.
Even when AMD began to push its processor "ratings" on the Athlon XP and MP CPUs, the ratings were designed to correspond with the clock speed of an equivalent Pentium 4. Undoubtably, this new name change may force AMD to rethink their labelings as well. Although AMD was originally correct to shy away from MHz labeling, now it will be more difficult for them to inform consumers on which processors are equivalent in performance.
Hurting AMD is probably not one of the main concerns driving the new model name approach. Take the mobile market, for example. Even though Banias (Centrino) is a brilliant processor (we have covered many times), customers are more apt to buy higher clocked Pentium 4 notebooks rather than a lower clocked, "low power" Pentium M notebook. Unfortunately, this is a case of when the trained "MHz" rating system is confusing and poor for the consumer. This ultimately becomes poor for the notebook manufacturer, as consumers complain about low battery life.
Intel's decision to move towards the new MHz less model name coincides with several new technologies, including Dothan (Banias's replacement) and Socket T (Socket 775). With an even more capable Pentium M, a mobile Pentium 4 killer, Intel and the consumer have a lot to gain by picking up the new lingo. We will have more Dothan information later this week.
How will this affect us, the consumer? Well, whether or not you agree with Intel's strategy, there is a more mundane advantage to adding a model number (other than OEM SKU) to each new processor; simplicity. If you just saw a new Intel commercial and want to buy a new computer, isn't it easier to remember "Intel 550" rather than "Intel Pentium 4 3.2GHz, 1MB Cache, 800MHz Front Side Bus, Socket 775."
Facebook
Twitter
RSS
31 Comments
View All Comments
Pumpkinierre - Saturday, March 13, 2004 - link
Yes, Derek #11, you cant quantify everything with a single index. In oz, we say 'horses for courses' ie different strengths animals for different race lenghts and conditions. Cpus with different pipeline length and no., branch predictors, cache architecture and chipset interface cannot be lumped under one index. That's why I'm against the AMD PR rating. If you throw in chipsets, GPUs, RAM etc. the number of variables escalates expomemtially. I'm all for direct subjective testing (no demos/synthetics beyond scaling experiments ie exactly the same system at different MHz or memory settings etc.). However no one does this, because it requires extensive testing and grouping of results (like phyla in biology). Nevertheless if you are going to label a cpu, the Abraxas index (AI), MHzxIPC, would be a good one.MHz is liked because it is an absolute index. You wont get that with IPC but I'm sure a good assembler/machine code programmer would be able to set up a routine that would cover the instruction set in a similar fashion to expected application code, and be close to absolute. The branch prediction doubt (Kristopher #14) would be covered by the fact that branching statements are in the instruction set and different lenght loops and non branch sequences would be included in the routine. The size of the routine should be contained within the smallest L1 cache (L2 caches preferably switched off) and no reading/writing to main memory during execution.
The only other performance index would be the memory/cache subsystem which could be described by the memory/cache subsystem latency (MSL). This naturally would be dependent on chipset and RAM (run at default or standard -JEDEC specs.) So I propose the Pumpkin Index (PI) for systems:
PI (Pumpkin Index)= AI(Abraxas Index)xMSL = IPCxMHzxMSL
I'll get my 15 minutes in the sunshine yet!
Oxonium - Saturday, March 13, 2004 - link
Adding on to my previous comment: But since Intel is the most recognized brand name, it would likely be chosen over an equivalently marked AMD *IF PRICES WERE SIMILAR*.Oxonium - Saturday, March 13, 2004 - link
Wow, it took CRAMITPAL more than 17 hours to respond to a thread just begging for his usual dumbass mutterings. Must be a new record.I agree with johnsonx that using a third party wouldn't work. For one thing, neither AMD nor Intel would likely agree on a common metric and wouldn't be bound to it. To make it effective, there must be a baseline processor for comparison and that baseline must not change over the years so as to avoid consumer confusion.
A common nomenclature might cause more confusion. For example, if there was an Athlon64 570 and a Pentium 4 570, to the average consumer (and even some not-so-average ones) there won't be any reason to choose one over the other. Sure AMD could claim 64-bit support on the box and Intel could claim HT, but when it comes down to it, they'll just look at the number. Granted, by using a common nomenclature, they should be getting equal performance no matter which CPU they choose. But since Intel is the most recognized brand name, it would likely be chosen over an equivalently marked AMD.
CRAMITPAL - Saturday, March 13, 2004 - link
Expect BOGUS performance ratings from InHell to defraud consumers as they've done since the Pig$ was rushed to market like the PressedCrap, aka Pig 4 E (for Enema edition). InHell has fallen and can't fix their design/production/performance issues so they once again chose to misrepresent the facts to save face and bilk consumers. This has all been documented on the Pig 4 and that's why InHell has a multi BILLION dollar Class Action Suit against them in court right now for FRAUD !crimson117 - Saturday, March 13, 2004 - link
like others have said, IPC isn't sufficient...That'd be like saying one program was faster than another because it contained fewer lines of code. However, the larger program may have that code to more efficiently handle certain situations.
consider an sql statement...
"select * from mytable" is shorter, but "select * from mytable where mypk=10" will run much faster.
tsee - Saturday, March 13, 2004 - link
This just introduces a whole bunch of confusion. You can never tell the specs of the CPU anymore without a long index. I don't know what the author is thinking, sorry.johnsonx - Saturday, March 13, 2004 - link
A 3rd party processor rating scheme won't cut it either. Look what happened when 3dMark was taken as a de-facto rating system for 3D GPU's: vendors started writing driver optimizations just for 3dMark, and perhaps even designed hardware with 3dMark in mind (just a guess on my part). Then there was griping about how 3dMark might favor some hardware designs over others.Imagine how bad it would be for CPU's if, for example, Sandra were enshrined as the official rating system. Each architechtural change would have one CPU maker clamoring for Sandra to include their new instructions or optimizations, with the other CPU maker(s) crying foul. Each new release of Sandra would be hailed by the CPU vendor who's scores went up by a bit, and accused of bias by those who lost ground. Finally, internal cpu microcode would probably start getting optimized for Sandra instruction sequences, at the expense of real-life performance.
Like it or not, the rating system we are stuck with is the one that each CPU vendor implements for their own products, be it Mhz, performance ratings, or model numbers.
KristopherKubicki - Saturday, March 13, 2004 - link
The problem with IPC is you run into problems with branch prediction - IPC is only half of the solution.Not like I really have a better one, i just thought i would mention my two cents :)
Kristopher
Stlr22 - Saturday, March 13, 2004 - link
"Even though Banias (Centrino) is a brilliant processor"I dont mean to be picky but shouldn't the "centrino" part be *Pentium-M* instead?
Would help clear up some of the confusion as to what the "centrino" really is.
Abraxas - Saturday, March 13, 2004 - link
nice to see my name mentioned :). i see why intel would do this, as its been pretty obvious that a 1.7mhz pentium M is faster than any pentium 4-M on the market.i stand by my previous statement, while average IPCxmHz is not necessarily perfect, its the best solution anyone has to offer. AMD processors (aside from HT ability) are faster than intel CPUs at everything. Intel has an advantage in encoding apps, which are so far the only apps that benefit from HT.
if intel named a 3.6ghz P4 prescott the 575, AMD would just name a 2.4 ghz A64 the 580. It adds even more marketing lingo to the whole system. its time for a completely 3rd party organization to step in for processor naming, or for OEMs to start building their systems saying "a 2.2 ghz athlon 64 is about 10% faster than a P4 3.2ghz due to a more efficient design." let the manufacturers do the mudslinging and marketing, not the cpu makers.