Next-P4M (Banias) to use PR rating

nja469

Distinguished
Jun 23, 2002
632
0
18,980
Intel Corp.'s upcoming Banias processor will use a "performance rating" system similar to that used by rival Advanced Micro Devices, Intel president Paul Otellini said in a roundtable discussion with journalists Monday.

Intel faces a knotty marketing problem with Banias: since the upcoming mobile microprocessor runs at a slower clock rate than Intel's current Pentium 4, how to sell it to speed-obsessed consumers? The answer: a ratings system, although not the one AMD uses, Otellini said.

Instead of pure gigahertz, Intel will choose to highlight the Banias' battery life, which Intel officials have said runs "all day". Otellini highlighted Bapco's MobileMark benchmark as one the chip maker would highlight prominently. Intel belongs to the Bapco organization, as does AMD; however, AMD has also claimed certain Bapco tests unfairly favor Intel's chips.

"The difference will be that we will be up front on what the actual performance is, and what benchmarks we're using," Otellini said. "We'll always be up front on what the clock speed actually is. And we won't use products that are based on the competition."

Intel will not use AMD's "performance rating" scheme, Otellini said, which is based upon a formula that compares its cores against prior generations.

A year ago, Otellini began saying that "gigahertz didn't matter," Otellini said.

Banias will also be marketed as a "platform brand", Otellini said, probably similar in design to the Celeron brand Intel gave its processors for low-cost PCs, or the Xeon line designed for servers and workstations. Although a Xeon may be designed upon the same architecture as the Celeron, the two chips can be clocked at different speeds and feature different amounts of cache to ratchet performance up or down for the target market.

Banias will also include an integrated form of wireless networking. At this point, Intel favors hybrid 802.11a/802.11b chipsets and access points, Otellini said, who agreed that Intel was currently ambivalent to the 802.11g specification.

Intel may also develop special benchmarks to evaluate the effect of hyperthreading, which will first appear in Intel's desktop 3-GHz desktop processors during the fourth quarter of this year.

"I don't know whether SPECmarks cover hyperthreading or not," Otellini said.

Source: <A HREF="http://www.extremetech.com/article2/0,3973,524192,00.asp" target="_new">http://www.extremetech.com/article2/0,3973,524192,00.asp</A>
 

nja469

Distinguished
Jun 23, 2002
632
0
18,980
To summarize; Intel has planned a benchmarked Bapco PR scheme for the banias chips since they run at much lower speeds than the current p4-m, thus saying they need a way to appeal or equate performance to the speeds of the current line. This part is great, "to sell it to speed obsessed consumers". He won't outright say it, but why else would consumers be speed obsessed? We all know why, the years of brain-washing that MHz = overall performance. This is also why intel built the P4 around high-speed scaling to appeal to consumers and sell the sh!t out of these chips! Well the PC boom ended, and the plan wasn't as lucrative as it should have been, plus AMD offered a bit more competition than I bet intel expected, but this is another subject.

Anyhow though, the irony is great ppl. Intel is doing exactly what AMD did, exactly for the same reasons in which intel criticized them for. AMD needed a PR scheme for their athlon XP's, since they didn't scale as fast but offered comparable performance to faster clocked p4's, and now intel needs a scheme to compare to the faster clocked p4's :)
 

shallowbaby

Distinguished
Apr 5, 2002
204
0
18,680
most ppl are mad because amd is targeting intel's mhz rating with their PR rating.

ppl are pissed that athlon is really paired up with p4.

while i doubt intel is doing exactly what amd did (in terms of PR), i think it's pretty cool that intel had crushed amd so bad that has to compete with itself.. (bring it)

this line of discussion goes for itanium processors as well since its core freq is low.

<font color=green> there's more to life than increasing its speed -Ghandi</font color=green>
 

FUGGER

Distinguished
Dec 31, 2007
2,490
0
19,780
The new PR scheme is based on battery life according to the post. If they can get "all day" that is huge step in the right direction for mobil computing. The current "2 Hour" battery life we wish we could get is not cutting it anymore.

So the longer the battery life the slower the machine? brave new concept in PR rating mobile machines, "all day" is something that AMD cannot touch and Intel will exploit it as much as possible.

BTW, GHz = overall performance, screw Mhz we are way beyond that.

AMD still has no answer to the bandwidth desparity. Maybe when the Hammer rolls out with IMC it will close the gap some. Think I am wrong, in the best data compression tests AMD flatlines due to lack of bandwidth.

You are limited to what your mind can perceive.
 

nja469

Distinguished
Jun 23, 2002
632
0
18,980
Excuse me, GHz.

Also what does this have to do with AMD and them currently not being at the top? geez... ppl can't get through one post w/o skewing the orig topic! I only said intel is launching a PR scheme now since their new chip isn't a super, screaming GHz factory like the P4. Yet the chip performs well and uses less power so intel wants to this known since most people still assoiate the large number with performance. This is the same reason AMD intiated their PR scheme, and intel critized them for it.

The post does mention battery life, but if you read on and intepret it all accordingly, it says they will use benchmarks to relate to performance. They said, "they'll be up front with what benchmark used" (BAPCO) and what "the actual MHz frequency of the chip is."

As for battery life, this will be a "highlight" of Banias. Battery life is the main reason for ppl to actually look at Banias. It's supposed to perform well with very low power requirements.

The actual PR will be based on a benchmarked scheme, which hasn't be disclosed yet. Battery life is Banias' selling point, not what the PR will be based on. READ READ READ and analyze.

What sense would a P4-MII 8 HOURS or P4-MII 2 Hours make? LOL
 

eden

Champion
In my topic about how I found the 3.6GHZ to scale with less IPC as speed goes up, I did mention Banias' PR rating, but that's ok if a new topic is up.

I agree with you, I dunno what Fugger is saying, it makes no sense to mention battery life as the PR. PR means Performance Rating usually. It's true Intel is contradicting what they taught people to back away from, so some might feel betrayed actually.
But I guess it's for the good.

--
Where did your THGC username come from and why did you choose it? <A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=19957#19957" target="_new">Tell here!</A>
 

eden

Champion
Really? I didn't see it as a call for attention or flaming!

--
Where did your THGC username come from and why did you choose it? <A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=19957#19957" target="_new">Tell here!</A>
 

function9

Distinguished
Aug 17, 2002
657
0
18,980
no, i think what fugger was trying to point out is the fact that intel has something <i>legitimate</i> to go on with their own PR system. if the above statements are true, they are going about it the right way. putting up benchmarks and explanations of what it means. to this day amd has yet to publicly address their PR system. (by public i don't mean to people like us that have the sense to go to their site and read up on it). public meaning to joe shmoe that's walking into bestbuy or something. i see no reason why, for a mobile system, battery life shouldn't be a consideration of system performance. i believe thats the main concept behind a mobile system free of power cords, outlets etc.
 

shallowbaby

Distinguished
Apr 5, 2002
204
0
18,680
yes, if intel has the "performance" to back it up in their mobile p4s.... battery-life is the icing on the cake.

<font color=green> there's more to life than increasing its speed -Ghandi</font color=green>
 

zengeos

Distinguished
Jul 3, 2001
921
0
18,980
Uhm....the CPU is only a small percentage of total system battery use. The biggest power hog in most notebooks/laptops is the LCD I think. So, while Banias may be less power hungry than P4 or Athlon, unless the systems it is installed in have very low power screens (small smal?) the true battery performance may not be all that much better. It also is rather odd that Intel would use a bettery life PR scheme since the CPU is only one component of the system using power. If the notebook has a larger screen the battery won't last as long, etc...so using battery life as the main PR for a mobile CPU is a bit confusing IMO.

Mark-

<font color=blue>When all else fails, throw your computer out the window!!!</font color=blue>
 

imgod2u

Distinguished
Jul 1, 2002
890
0
18,980
The demo they showed at IDF suggested that the P4M will still be in the high-end performance laptops while Banias is suppose to replace the P3M in low-power, portable notebooks. Also, shaving off close to 30W of power drain is quite a lot not matter what notebook you're using it in.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
 

Dragon9137

Distinguished
Jun 20, 2001
42
0
18,530
Allow me to show my stupidity here for a second. I don't understand why there cannot be a universal measure of processing power. Everything boils down to math anyways. Look, the ultimate measure of a processor is information processed/time to process. So, if it takes 18 stages to process a byte of information and we have 2,000,000,000 cycles per second, then it seems 1,000,000,000/9 would be a universal measure of information processed per second. Now I can see the arguments against this: Well there are different numbers of stages for floating point numbers and binary numbers. Give me 1,000 randomly chosen operations and I can give a median and standard deviation pretty damn close to real life.
 

LancerEvolution7

Distinguished
Jul 17, 2002
571
0
18,980
Hehe, this post seems kind of funny to me. I'm not flaming intel, but I saw an advertising scheme, gimmik, whatever, by intel that completely contradicts the article. It went something like this:

"Pentium 4 2.0 ghz, a true 2.0ghz cpu"

Now that wasn't the exact statement, but the general meaning is there. Its nice to see that a universal "PR" might come out of these cpu wars.
 
G

Guest

Guest
I think your new to this world. If only it where that easy... those "randomly chose operations" are pretty meaningless. What you're suggesting here, is nothing more than the most primitive benchmark ever written ;-) There is no way such a thing would give any indication of real world performance. Already its not really possible to compare Athlons to Pentiums to Itaniums.. they all excell at (or have their weaknesses in) different programs / tasks. Besides How are you going to factor in things such as SSE, SSE2, Hypertrheading, memory subsystems, cache (big apps love huge/fast caches, for some it doesnt matter). Im only scratching the surface here, but benchmarking is somehting you could write pretty thick books about.

Either way.. SPEC probably comes close to what you suggest, but even SPEC doesnt help me decide on a platform to play games, edit video or do 3D rendering (to illustrate, using Lighwave or Cinema4D already paints a hugely different picture. Even within lightwave, choosing one 3D model instead of the other could make processor A look more then 20% faster than processor B or vice verca.. go figure).


= The views stated herein are my personal views, and not necessarily the views of my wife. =
 

eden

Champion
Hmm since when are there 18-stage CPUs in the first place, to process a byte?

Unfortunatly there had been long heated debates a long time ago here about universal processor performance measuring, but none were close to perfect and although some had creative ways, it's up to who can move and push it to be developped widely.

Give me 1,000 randomly chosen operations and I can give a median and standard deviation pretty damn close to real life.
I didn't get that!

--
Where did your THGC username come from and why did you choose it? <A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=19957#19957" target="_new">Tell here!</A>
 

imgod2u

Distinguished
Jul 1, 2002
890
0
18,980
Allow me to show my stupidity here for a second. I don't understand why there cannot be a universal measure of processing power. Everything boils down to math anyways.

That is far from the truth. If everything boiled down to the math that the CPU's do, there'd be no such thing as programming efficiency, algorithms and optimizations. If programs out there behaved like simple math calculation generators, it'd be nice, but they rarely do. They're a complex collection of logic and calculations, one following the other, depending on the other. If you look at any realistic code, you'll see that roughly only 40% of it is actually math. The majority of it is control flow, assignments, pointers, sorts, stacks, timers, etc.
And hence lies the problem. The measure of a CPU is not how good it is at math, but how good it is at handling all of these complex twists of logic and dependencies in code. If it were just math, CPU's would be nothing more than execution engines. There'd be no need for prediction, prefetch, decoders, tags, and all the things that we heavily put into to make it so we can get as many instructions as we can to run in parallel out of the code that we have.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
 

Dragon9137

Distinguished
Jun 20, 2001
42
0
18,530
Given any random set of input/output pairs that has > 30 members, I can define a statistically accurate bell shaped curve. Give me a set of 1,000 or 10,000 or 100,000 and the standard deviation becomes much smaller. In otherwords the bell gets more and more narrow. Regardless of whether or not a processor does better in one area or another, I should still be able to get a very accurate statistical graph that gives me a close to real life view of how fast the processor can process information.
 

Dragon9137

Distinguished
Jun 20, 2001
42
0
18,530
If only it where that easy... those "randomly chose operations" are pretty meaningless. What you're suggesting here, is nothing more than the most primitive benchmark ever written ;-) There is no way such a thing would give any indication of real world performance.
How so? I could create a frequency chart of the types of operations a computer performs. Creating a histogram of such with let's say a million pieces of data in my set allows me to model real life computing. I can use that information to create a processing time curve per operation. That curve would be so accurate that the standard deviation would be sqrt(sum(X-mu)^2)/1,000,000))). That value would be so small that any real varience would be null.
 

slvr_phoenix

Splendid
Dec 31, 2007
6,223
1
25,780
No offense Dragon9137, but this is exactly why math majors make absolutely horrible computer programmers. Edsger W. Dijkstra has an excellent anecdote for this:

<A HREF="http://www.cbi.umn.edu/collections/inv/burros/ewd594.htm" target="_new">http://www.cbi.umn.edu/collections/inv/burros/ewd594.htm</A>

It's a bit dry, but if you can get through it all, the last paragraph sums up the point succinctly.

<pre><A HREF="http://www.nuklearpower.com/comic/186.htm" target="_new"><font color=red>It's all relative...</font color=red></A></pre><p>
 

Dragon9137

Distinguished
Jun 20, 2001
42
0
18,530
well, by Dijkstra's account I must be a mathematician, though he never states how Philosophers react to that story, so maybe I'm not too inconsistent. In fact, what I am suggesting here is just like the guy in this story. You have one camp yelling "Measure cycles per second" and another yelling "Measure instructions per clock" and me saying "Statistics gives us a fairly undebatable method of measuring both together."
 
to some people a longer battery life is performance. because they can work longer which is better performance. Usually when someone is buying a mobile notebook they get it to work on the road and usually cares less how fast it is as long as it works.

But that's just what i think intel would say. What I say is that to rate performance is very difficult.

you need to have one benchmark and use the numbers from that benchmark is the pr rating for ALL chips as a global standard. which isn't likely. But doing a pr rating on battery life is a lot easier and thats usally what people want on a mobile notebook anyway.

<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>
 

eden

Champion
Yes but that would confuse people, because the higher the speed, the lower the PR, which means people will know if they are looking into that big number, they will get less MHZ, which is something all the average users look at for references.

I do agree battery performance is a must, but that should be put in a slash, / after the PR. Such as Banias (wonder what the name will be? P3.5?!) 2100/3H. That IMO would not be bad, may not be too accurate then even, since all OEMs would have to adjust the components to match 3 hours anytime. So it would be hard to put PR ratings on battery performance anyday.

--
Where did your THGC username come from and why did you choose it? <A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=19957#19957" target="_new">Tell here!</A>
 

slvr_phoenix

Splendid
Dec 31, 2007
6,223
1
25,780
well, by Dijkstra's account I must be a mathematician, though he never states how Philosophers react to that story, so maybe I'm not too inconsistent. In fact, what I am suggesting here is just like the guy in this story. You have one camp yelling "Measure cycles per second" and another yelling "Measure instructions per clock" and me saying "Statistics gives us a fairly undebatable method of measuring both together."
But the point is, computer programmers are creative problem solvers. As such, software is hardly ever a series of mathematical expressions, but instead a complicated nest of logic to choose the right formulas and perform the right tasks only when it is appropriate to do so.

As such, <i>most</i> software is a complicated mass of conditional statements, boolean logic, which derives from mathematical expressions. So in the end, you never know just what code is going to be executed next.

Processors have prediction algorithms to try and predict this though, as well as various caches to try and hold enough data to keep what you'll need next along with what you're currently executing to save time. (Because memory access takes a long time to a CPU.)

A serious of random floating point and integer calculations would <i>not</i> reflect the amount of work that goes into CPUs to predict what data is needed next or to cache a lot of data so that what you need next is already there. Nor would it even come close to identifying how individual programs run, because each program (and each programmer for that matter) uses different logic to determine how to solve problems and what to do. Different logic means different ratios and patterns of integer and float combinations in boolean logic determination.

And so in the end, the only <i>good</i> way to compare processor performance is the simplest in concept. Run commonly used software and compare those results.

We could come up with all sorts of idealistic ways to benchmark processors, but in the end, it'll only be meaningful for software that has logic patterns similar to those methods. And since problem solving is hardly ever idealistic logic, that means most software will never relate to such benchmarks. In the end, these idealistic methods (such as yours) would end up with a number virually as meaningless as MHz for comparing the actual processing capabilities of a CPU.

<pre><A HREF="http://www.nuklearpower.com/comic/186.htm" target="_new"><font color=red>It's all relative...</font color=red></A></pre><p>