Slashdot Mirror
AMD To Hide MHz Rating From Consumers
pezpunk writes: "Tom's Hardware is reporting here that AMD's next-generation Athlons will be identified by model number rather than Mhz rating. This means that an Athlon will be designated an "Athlon 1600" even though it's only a 1.4Ghz part. The true clock speed of the chip will NOT be shown either on the chip itself or even in the BIOS. Apparently, they're desperate to compete with higher-clocked Pentiums in the minds of consumers -- proof that even the underdog can pull dirty marketing tricks =("
I've been wondering about the Hz Myth issue for quite some time now. Mainly ever since Steve Jobs at MacExpo NY a few months ago where he demo'd some machines that smoked Intel chips despite being about half the MHz. Do note that this a long post, and had it pre-written waiting for the next oppurtunity to post it. I feel this is that oppurtunity.
I heard a rumor that Intel counts both the rising and falling edge of the cycle, while Motorola counts only the falling edge. This rumor was in respect to how Apple's chips were always considerably lower in Hz than Intel chips, primarily in how Apple's chips were half as fast as Intel's. (Motorola designs and manufactures Apple's CPUs)
A co-worker and I discussed this and did some analysis. The guy was an EE in college, and said that actions were performed only on the falling edge of the clock cycle. Hence, it was not possible to include both falling and rising edge in a clock cycle.
We took a mathematical algorithm, wrote an implementation in C, and added some timing code. I should note that this C program is a _typical_ complex math algorithm that could occur on any machine, and are not tailored to perform better on any given machine. We compiled the source using gcc 2.95 on a Red Hat 6.2 box with an Intel 1 GHz Pentium III and also on a Mac OS-X using the same version of gcc on a 500 MHz G4. Both compilations used full optimizations. We then ran them on their respective machines (using the same input) several times each and calculated the average amount of time it took for the algorithm to reach completion. The results were about the same for each machine.
After reflecting on this result, we think that Intel is using both the falling edge and rising edge in an attempt to better market their products. We arrived at this conclusion by going back to classical wave mechanics from our physics classes in college. Take a waveform, say the trignometeric sine wave for example, and notice how the wave rises for the first pi/4 on the X axis. It repeats this shape 3.14 later. Hence the wavelength of the sine wave is pi. By definition, the number of complete waves in a given second is the frequency of the wave.
Apply this now to the waveform of a circuit. Specifically, the clock. Notice how the frequency of the wave is composed of both a rising and falling edge. If actions can only take place on the falling edge of a clock cycle, then Intel has doubled the clock speed on paper only.
But wait! You might be asking why have AMD's chips also been comparable to Intel's in terms of clock speed? Because AMD is directly competing with Intel, and they need that edge in the market. Assume you're a regular Joe Q User, would you buy Intel's 2 GHz processor or AMD's 850 MHz? Our conclusion is AMD has also doubled their numbers in order to better compete with Intel.
But wait again! What if Intel has figured out how to get actions to occur on both the rising and falling edge? If this is true, Intel's chips would perform _largely_ better in benchmarks than they currently do. If this was true, the timing test my co-worker and I performed would not have resulted in similar numbers, it would have had Intel getting a timing roughly half that of Apple's. Therefore, if this is true, Intel's engineers have done a lousy job at exploiting this novel concept, which I highly doubt would happen. Our conclusion is that Intel hasn't figred out how to have actions perform on the riding and falling edge.
Final Conclusions: After doing some scientific analysis that includes benchmarks and revisiting concepts learned during a college physics course, we conclude that Intel is counting both the falling and rising edge of a clock cycle, despite facts that fail to support this idea when compared to a processor that is measured using only the falling edge.
Companies like Apple and AMD are doing the correct thing by wanting to find a better means of quantifying the performance of a processor. They are doing the correct thing by telling consumers there is more to a processor than it's clock speed.
The One Rule Of Chess You'll Ever Need: Don't play someone who carries a kit in their bookbag.
>>We do not buy cars based entirely on how many RPMs they are capable of.
No, but horsepowers do influence our decision. Much less, though, because the cars are not named 'Integra 180hp' and 'M3 340hp', while the CPUs *are* named 'Athlon 1.4GHz', 'P4 1.6GHz'.
So, it's a good marketing decision, to make up model names/numbers for different CPUs. As for hiding the actual clock frequency -- for the people who care to find out, it can't possibly be a big problem to figure it out.
Never underestimate the bandwidth of a 747 filled with CD-ROMs.
got look at some Tbird vs P4 benchmarks and tell me I'm wrong there
Please post a link to these "lies" that you so boldly claim. Even better, some proof that they are lies.
Unlike Intel, they're at least not lying about clock speed.
How is it better to attempt to intentionally mislead people? Cyrix tried this same trickery, and suffered the consequences.
This is no better than Apple's misleading claims that some bogus narrow benchmark or extremely optimized, specific operation (e.g., photoshop filters) is a measure of overall performance.
As evidence of my statements, I direct you to John Carmack's post regarding his performance tests of x86 versus PPC.
There is more to performance than what a lot of people want you to believe. This AMD move is simply about misleading consumers.
Sometimes it's best to just let stupid people be stupid.
It would show high-level results of independent testing of appropriately chosed AMD and Intel based systems (and not a whole raft full -- just one or two popular sets). With a couple of graphs and summary numbers, the consumer would see that an Athlon 1.4GHz performs comparably to or better than an Intel 1.8GHz, or whatever.
Don't overwhelm them. 1 page of reasonably large text and pictures should suffice.
The harder part (once control has been wrested from the brainless marketers) would be to get the sheet into salespersons, saleswebs, etc. hands for presentation.
Customers, even non-technical ones, are very adroit to smelling BS, and very adverse to it. I think an honest approach might work better than the proposed obfuscation.
Let's see numbers that really can be compared across different platforms, such as:
How many minutes to compile these particular 100,000 lines of C++ code?
How many gigaflops?
That that is is that that that that is not is not.
The AMD Athlon is faster than Intel's Pentium 4 at the same clockspeed. Therefore, a Pentium 4 must be clocked higher to match Athlon's performance. Most people who are buying computers do so largely on three considerations, brand name, MHz, and price.
/.er you'll notice that none of those is a form of performance measure. Therefore, unless Intel decides to start using a uniform performance measure of some kind AMD should make a conservative estimate as to how fast their parts are compared to a Pentium 4 as a service to customers who are out there buying based on a deceptive clockspeed number.
Being an observant
I'm not so sure that you could call this a "dirty trick". The MHz myth was expounded on here in /. recently. It'll be a nice change of pace to see chips benchmarked by true performance rather than artificial measurments.
But that's just my opinion.
--
If I actually could spell I'd have spelled it right in the first place.
Apple is doing this now -they face dwindling market holding because they are oft-seen as the easy-to-use-but lacking power solution. Not so. Truth be told, Apple could blow away any PC proc based on a purely MHz rating too - this from a die-hard PC user - but this is not the forum for that debate...
Apple has taken the high road. They have begun to educate users en-mass about the problems of relying on the MHz rating. Sure we know better: Intel cannot benchmark equal to AMD on a MHz rating because they cannot run the same number of operations per second. Thats simple math.
The analogy I like is who has more light: if everyone 100 light bulbs but all mine are 100 watt and everyone elses are 60 watt, everybody can see that the 100 watt bulbs are going to produce more light, but it still seems like everyone is comparing the number of bulbs - "Its got to be brighter becasue they have more bulbs!"
AMD has gone the opposite way in the analog, like saying we're giving you 60 bulbs but the amount of light will be the same as intel's 100 bulbs. And most people are still stuck saying "ya but you're ripping me off for 40 light bulbs!" AMD needs to take a better look at how the big picture will appear to the public, are they looking for more light, or more bulbs?
"Quando Omni Flunkus Moritati" -- Red Green
Yah, because risc chips take (or are suppost to, rember RISC originaly was toated as high clock rates low CPI, now they are low CPI and low clock rates) more instructions to get the same amount of work done. Sure, there are exceptions, especially since none of the processors you cited are accually RISC, they are load/store arch's though. No one accually uses MIPS (ha ha ha joke, think mips and in Rxxxx) anymore anyway, that was taken away from the 'RISC' camp when intel introduced the Ppro and it could add (rember that was one of the main components for calculating IPC) like mad. Thats when everyone started whining about SpecInt. Then the RISC people were selling their processors on MFLOPS which is the floating point version of MIPS. Until the PII/PIII (and especially now with the P4) had their clock rates bumped through the sky and there went SpecFP.
Then there was STREAM, which basically was a nice benchmark to measure memory bandwidth. This was good because memory bandwidth for the most part is bigger the more you spend on esoteric massivly parallel memory subsystems. Then the P4 came along with its rambus and prefetch logic. Whops!
Now we have people who would like to measure the speed of their general purpose processors on functions that should be in a DSP in your ___(fill in the blank, video card, sound card, etc).
Ah, the non x86 people will eventually face the light, Intel and AMD spend 10x as much on R&D as everyone else because they have 1000x as much volume. You may have better techology but you cant keep up with the people who have 100x as much money. So you have to come up with niche markets like watts per workload. Except that one is about to get clobbered by intel now that the notebook market has been sparked by Transmeta.
Not exactly what you're looking for, and not text, but here's a link to a series of charts I've found very useful. They cover pentium era through late P3, plus Athlons up to the 1GHz Slot A. They show handy stuff like clock speed, bus speed, multiplier, cache size, votages, etc.
Processor and Chipset Tables
Carpe Cerevisi - Seize the Beer
Intel can rename there pIV's as pIV 3200 and the consumer will false think the intel chip is twice as fast as the athlon 1600. Got to love marketing.
Hey, speaking of clever marketing, remember when NT 5 which was due in 1997 got renamed as windows2000? Hehe. It worked. I told my boss that microsoft took ages to make w2k and it was long overdue. He said quote "Its not late. Why do you think Microsoft named it Windows2000 ?". He fell for it.
Also go to your grocery store and look at Campbells Chicken soup. The can with a picture on it is $1.45 and the other can without a picture is $ .99. The 2 soups are identical ingredient by ingredient and quantity, yet the consumer pays more because one can has a nice pretty picture on it.
Sadly consumers are really suckers for things like this. Megahertz ratings included. Same is true for clever wording. Notice how microsoft's products are all verbs? Internet Explorer, Access,excel, etc. Marketing does really work and people subconsiously think of these actions each time they open the apps. Ask any Phsyc. major? Using verbs and positive adjectives does influence people. Anyway consumers just want something that looks visually appealing and is highly marketed. Perhaps AMD could rename the athlon to a verb. Hey Geforce256 is a great example. I admit a geforce is the fastest chip available but I am sure the name helped them greatly market it.
Expect intel to do something similiar like I mentioned above with names for its chips. Intel does have the extra hand in marketing due to brand name recognition. Also without a magehertz rating many consumers who are second time buyers know to look for a megahertz rating when buying computers. They may be nervous and wonder what AMD is hiding when no info is available. They will probably pick intel to be safe. Or pick the chip with the higher number in its name. :-)
http://saveie6.com/