CPU Clockwars - WARNING: GEEK ALERT
Fredrickville.com
Posted by FredFredrickson on 02/23/07 12:53am.

So, please excuse this random (but apparently increasingly common) geek news break from Fredrickville Senior Technician Roboto Fredrickson. CPUs, what the hell are they, and why can't I tell what the hell a 3200+ is compared to a 3.2 ghz p4?



A Marketing Train Wreck

Long gone are the days when it was typically easy to compare processors. Same-brand comparisons, cross-brand, what have you. I have an Intel 233 mhz that's comparable to the AMD 233 mhz machine. Life is good.

But after not too long, AMD decided to get more bang for your buck, and decided to get more done with each clock cycle. This confused the heck out of people. Why was my brand new computer running a 1.6ghz processor, while intel was well into the 3 ghz range? This makes no sense to the consumer, and it makes no sense for AMD, unless they change their marketing. Enter PR-Ratings. Or P-Ratings. Or PR. Public relations? Performance ratings? Pentium ratings? According to wikipedia, it's Performance rating. Ok, great. So now I know that my 1.6ghz Athlon XP is comparable to a Pentium 4 2Ghz machine, thanks to our new PR. PRating. PR Rating: 2000+. Right? Am I right?!

Well limited text on the subject actually exists. Was it coincidence that the 2000+ was out around the same time 2Ghz was popular? AMD knew what they were doing making the number appear to match in some way. But is there any official word of that? Actually, I couldn't find much. I found some varied reports from non-authorities on the subject mentioning that instead of an intel comparison, we're actually seeing how the processor compares to the thunderbird core. (from AMD). Of course, I couldn't find any official word one way or the other, so this seems to be a moot point.

But what do we know for sure? These PR Ratings can certainly be used to compare performance between different chips of the same brand. We know that the Athlon 64 3200 is faster than the Athlon 64 3000. Even if we don't know how fast either are in the real world, we know their relative speeds. But how useful is this seemingly arbitrary number?

Arbitrary Numbers

This ... great new system allows us to tell what we expect the performance to be, despite the many, and often cryptic, factors involved. L1 Cache? L2 Cache? Dual Core? Different Cores? Magic?

For example, the AMD Sempron 64 Manila 1.8GHz is labeled the 3400+ as well as the AMD Sempron 64 Palermo 2.0GHz (3400+). Both rated 3400+, despite the difference in clock speed. What makes the Manila faster than the Palermo? Sometimes the L2 cache is different. In this instance, it appears it's solely the core: Manila is more efficient than the Palermo. Neat. But does that matter to the consumer? No, because we have the advantage of this PRating, letting us know that the above combinations still give us the same performance. Thanks AMD, thanks for that!

So why does the Sempron even exist? VS the Athlon? What's the major advantage of buying an athlon 3200+ vs a sempron 3200+? Some would say the athlon has a much larger L2 cache (a great improvement, don't get me wrong.)

The athlon 3000+ has twice the L2 cache than the sempron has, despite the same clock speed. The L2 cache, in itself, would be a great advantage, except one major problem... the PR Rating. I don't see any reason to care about the L2 cache if they estimate the total performance with all factors taken into account give us the result of a 3000+. In other words, to weed out six-of-one, half-a-dozen-of-another issues confusing consumers, they've given us a PR rating. To the average consumer, and for most practical purposes, this means all those factors are irrelevant.

So why is the athlon more expensive? Why do people prefer the athlon over the sempron?

Let's turn to the benchmarks:

It appears that for both the 2800, and the 3000, the athlon and the sempron are very close in performance, with a rating difference less than 10 points in both cases; athlon wining both.

But here's the weird thing: both 2800's scored higher than the 3000.

And strangely, the 2800 scored higher than the Celeron 3.06GHz, but lower than the Pentium M 1.60GHz.

This means our PR-Ratings are counter-productive for AMD (not really boosting sales for more "advanced" chips), apparently completly irrelavant in comparison to other amd models and worst of all: they appear to be absolutely arbitrary to their intel counter-parts (what ever those might be).

To make matters worse, it appears intel stopped surviving the mhz/ghz game, and has resorted to pr numbers as well, with the Pentium D 820 Smithfield 2.8GHz at the same clock speed as the Pentium D 915, we're left with a mess.

No standard rating system

Now that Intel can't simply be stated in mhz/ghz, we suddenly have a huge missing gap in the industry for any kind of standard rating system that actually means anything.

With no standard system within the same company, it'd be even harder to find a standard system across different companies. This appears to be a giant marketing blunder, causing nothing but confusion for consumers, and not stressing the actual benefits of upgrading processors.

Without going into intel's idiot system, I will again show example of AMD's idiot naming scheme.

The AMD Turion. With model names such as the MK-36 and the MT-38, it appears to make no freaking sense. Well according to a few sources, the 'M' stands for mobile (as apposed to 'T' for.. dual core. That makes more sense than everything else so far, right?). The second letter stands for it's power consumption, or more over: it's usefulness in a laptop. The further down the alphabet, the less power, the longer the battery, the sexier the pron. MZ has a 400 year battery life.

Now what about the numbers? Wikipedia says "while the number represents a PR rating." Great, "A PR Rating." Making no attempt to make this PR rating relevant to the consumer.

If we took this number literally, I'd say the MK-36 is about 100 times slower than the sempron 3600, due to the loss of two digits. Most likely however, it's more likely the 36 is supposed to be comparable to the 3600. Unfortunately, the benchmarks from PassMark indicate that the ML-32 underperforms even the 2800 (let alone the 3600).

Are these benchmarks perfect? No. Are they biased? Probably. Has AMD made any effort to clear any myths or suspicions relating to their dumbass naming scheme? No.

Congratulations, AMD, you've added more confusion to an already confusing market. Intel you suck too.

Conclusion

My conclusion is that none of the above makes any sense. Because of this I plan to pretend that these numbers actually mean something. I'd love to pretend that the 3200 is a 3.2 p4 equivelent (for less money, yay!). But with the loss of the P4 constant clock additions, even this pretend world will not actually work. So I declare a new naming scheme that is horribly inaccurate, based on the only variable that seems to be a constant during this whole thing: Price.

AMD and Intel release their chips for a certain price based on what class they believe these parts to be. Now everybody knows AMDs are cheaper than Intels, therefore it's a given that AMD will win every battle within the same class - which is a great improvement, as opposed to the random battles that benchmark websites have been presenting since the beginning of time that take two completely arbitrary and random processors, fight them, and declare one the winner. Since the model numbers apparently don't reveal their other-brand counter-part, it doesn't make sense to fight processors based on their random model numbers. But their price does reveal what AMD believes their parts to match up with... kind of. Because AMD has more bang for your buck, this reasoning is inherently flawed. The cards are stacked against intel because AMD sells discount parts. But in this freaking stupid world of processors, all we have is price, so that will now determine the new classing.

I have written a quick program to take the benchmarks and pricing into account and show us exactly what processors really match up with eachother, I'll post it here when it's finished, so you can play with the results yourselves.

[EDIT: Program is finished. It uses the PassMark rating. I couldn't find enough prices to use those, sadly chips appear and dissapear almost as quickly. CPU Matcher]

Mostly, I'm just pissed right now.

StumbleUpon Toolbar Stumble this article.


ThatSam @ 02/24/07
"geek"



Login to comment...

Return to main...


 
Fredrickville.com

Who am I? That's a good question. I am the creator of Fredrickville.com, and the writer of the front page column. My actual name is Robbie, but for whatever reason I like Fred better. If you read my column, thanks. Feel free to leave a comment or two. Chances are, I've pissed you off anyway.

Search This Column


Home | Columns | Pics | Privacy Policy | Disclaimer / Terms of Use | Gripe
Everything Else ©2017 Fredrickville.com All Rights Reserved.

website tracker