Pentium M benches and a ****Shocker****

[dnuggett]

After reading several publications and seeing what was generally accepted in the industry I used to accept that the multiplier for conversion to P4 speed from the M was 1.6. It appears as though that is incorrect.

Measuring the Dhrystones and Whetstones in Sandra I have concluded that the multiplier should be approximately 1.3. The exact # according to Sandra is 1.26 but in testing I found a variance from 1.26 to 1.33. I feel safe saying the multiplier that should be used is 1.3. This comes as comes a shocker to me, but I don't believe it undermines the chip. It was designed to give great performance at an acceptable power consumption level and it does just that. It is just no longer the speed demon we used to think....

For example under our old way of thinking the M 1.7 would be equal to a P4 @ 2.72.
My conclusions show the M would be equal to a P4 @ 2.21.

There is a great bit of difference on the bench. But that difference would only be noticeable in CPU intensive tasks such as rendering, encoding compiling and or gaming. The majority of users buy a laptop for office use so the differences would be negligible.

All this being said it is still a Seti monster...

 

[RichieZ]

so you base your multiplier on a synthetic benchmark? I would think a more valid method would be to use real world applications.

 

[alexruiz]

I always thought something similar for a long time, and I am glad you have more data now to consider it a little more "official"....

These are my thoughts:

For office applications: 1.7 times a P4 or better, and a fierce competitor to the Athlon 64... even better than an Athlon XP at the same clock.

For multimedia: 1.1 or lower..... if a lot of data exchange is needed it will trail even further.

Gaming: 1.3-1.4.....

For an all-powerful laptop, Athlon 64 should be the best, but even the AMD fans (like me) concede that the P-M is the most balanced solution for laptops..... this doesn't mean I would get one, just admittting this is a good chip


Alex

 

[manko]

Interesting results. I suspected that the generally accepted 1.5/1.6 multiplier might be a bit overly optimistic. I'm most interested in video encoding and high definition decoding performance, but still haven't been able to find any good toe-to-toe P4 to P-M comparisons. I have always assumed the P-M would be generally weaker in these areas though.

 

[dnuggett]

Originally posted by: RichieZ
so you base your multiplier on a synthetic benchmark? I would think a more valid method would be to use real world applications.

Yes as is almost every processor. How do you think they come up with a Ghz rating? It's the FSB times the chips multiplier, not how fast it can open Word. The synthetic testing is the most objective way to come up with an accurate multiplier.

But since you have suggested it, go ahead and do some subjective testing and tell us all what what you come up with.

 

[Pandaren]

Deriving a general muliplier based on one synthetic benchmark will not produce accurate results.

The synthetic testing is the most objective way to come up with an accurate multiplier.

Without knowing the instructions mix and size of the program, it is impossible to say that synthetic testing is objective.

For example, one could design a very tiny benchmark that would fit in the Pentium 4's L1 cache, is easily SSE2 optimized, and has few branches. This benchmark would probably show the Pentium 4 to be much faster than any other x86-32 or 64 chip.

The only way to evaluate systems as a whole is to test that applications and tasks you want to run on different platforms.

 

[dnuggett]

Originally posted by: Pandaren
Deriving a general muliplier based on one synthetic benchmark will not produce accurate results.

The synthetic testing is the most objective way to come up with an accurate multiplier.

Without knowing the instructions mix and size of the program, it is impossible to say that synthetic testing is objective.

For example, one could design a very tiny benchmark that would fit in the Pentium 4's L1 cache, is easily SSE2 optimized, and has few branches. This benchmark would probably show the Pentium 4 to be much faster than any other x86-32 or 64 chip.

The only way to evaluate systems as a whole is to test that applications and tasks you want to run on different platforms.

As mentioned before the application is Sandra, one of the most accepted bechmarks in the industry. The program does sit in the L1 and does not utilize the L2. Only the core of the proc is tested. Two benchmarks are given as far as the Whetstones are concerned with one of them including SSE2 . It gets rid of the legacy FPU and takes advantage of SIMD operations.

BTW, bud incase you missed it I didn't evaluate any system as a whole just the processor. Your purpose and mine appear to be a little different....

 

[Pandaren]

Edit: Both you and the industry may be correct. The Industry is testing PM systems as a whole, while you have stated that you are testing only the CPU core.

Everyone should be happy now

is Sandra, one of the most accepted bechmarks in the industry

Do you have any proof of this? Aside from the fact that enthusiast websites tend to use it? From what I have seen over the years, I am extremely skeptical that SiSoft Sandra is a good indicator of general system performance in real world applications.

The program does sit in the L1 and does not utilize the L2

Which effectively negates any real world performance value.

bud incase you missed it I didn't evaluate any system as a whole just the processor. Your purpose and mine appear to be a little different....

You can't easily disconnect the two, esp if they are different platforms since the processor depends on the chipset, memory and other components to achieve its performance. If the supporting platforms are the same, a more dirrect comparison is attainable. AMD's rating system fell apart for the Athlon XP for precisely this reason (Intel improved supporting platform of P4).

At any rate, AnandTech's benches show roughly a 1.5x multiplier in Pentium M/855 against Pentium 4/845. Give the Pentium 4 a better chipset/HT enabled and the multiplier will vary depending on the application.

 

[dexvx]

Originally posted by: Pandaren

is Sandra, one of the most accepted bechmarks in the industry

Do you have any proof of this? Aside from the fact that enthusiast websites tend to use it? From what I have seen over the years, I am extremely skeptical that SiSoft Sandra is a good indicator of general system performance in real world applications.

Uh Sandra is not a very widely accepted benchmark.

The most wide accepted synthetic benchmark is http://www.spec.org

 

[dnuggett]

Originally posted by: dexvx

Originally posted by: Pandaren

is Sandra, one of the most accepted bechmarks in the industry

Do you have any proof of this? Aside from the fact that enthusiast websites tend to use it? From what I have seen over the years, I am extremely skeptical that SiSoft Sandra is a good indicator of general system performance in real world applications.

Uh Sandra is not a very widely accepted benchmark.

The most wide accepted synthetic benchmark is http://www.spec.org

I think you guys are off a little on your estimate of Sandra's acceptance. Here's a quote from PC World after handing it the 2003 Worldclass Award:

"SiSoftware (formerly known as "SiSoft") is known as a world leader in the PC Benchmarking / reporting arena. Over 8 years of development has gone into producing this independent diagnostics tool. It allows retailers, home enthusiasts and field professionals to gain additional insight into their computer?s internals and compare and contrast its components with others."

And a little background on the program's usage:

SiSoftware, founded in 1995, is one of the leading providers of computer analysis, diagnostic and benchmarking software known as "SANDRA" (used by over 350 review sites and magazines, featured in over 3,500 on-line reviews and with 10's of millions of copies downloaded).

To say that it is not widely accepted, well....... I'll just leave that alone.
If you disagree with it's testing methods or accuracy thats your bag. But it is a very commonly used tool by enthusiasts and professionals alike. I'm not saying it is the king of all benches or the even the most widely accepted. I really don't know if it is or isn't. But I do know it is a great tool and is very widely used.

 

[GnomeCop]

hey cool, 3dmark is widely used too

 

[RichieZ]

I can be just as arbitary as you and base the conclusion on one benchmark:

Multiplier should be > 1.5 since it acutally beats the 2.4 P4

You may say this tests overall system performane and is not just a CPU bench mark but I say the desktop system acutally has the advantage since it has a 7200RPM hd vs 5400RPM hd

Also may I ask which Centrino/Pentium M laptop you tested?

I'm not saying the multiplier should be 1.5, but that to base your results on one synthetic benchmark doesn't tell the whole story

 

[RichieZ]

Originally posted by: dnuggett

Originally posted by: RichieZ
so you base your multiplier on a synthetic benchmark? I would think a more valid method would be to use real world applications.

Yes as is almost every processor. How do you think they come up with a Ghz rating? It's the FSB times the chips multiplier, not how fast it can open Word. The synthetic testing is the most objective way to come up with an accurate multiplier.

But since you have suggested it, go ahead and do some subjective testing and tell us all what what you come up with.

I fail to see how FSB * multipler is a synthetic test. So in your thinking a Pentium 4 2ghz is the same as the Celeron 2ghz becasue their "Ghz rating" is the same?

 

[vegetation]

Dhrystones is a classic benchmarking system dating back to the VAX 11/780 = 1 mips. There may be some dubious results when comparing cpu's with completely different architectures but for x86 chips it's pretty much as valid as you can get to raw computing numbers.

 

[dnuggett]

DP

 

[dnuggett]

Originally posted by: dnuggett

fail to see how FSB * multipler is a synthetic test. So in your thinking a Pentium 4 2ghz is the same as the Celeron 2ghz becasue their "Ghz rating" is the same?

I fail to see it too??? What in the world are you talking about??? Yeah Richie that's my thinking.

If that were my thinking whay would we even be discussing a multiplier?

The process of attaining Mhz or Ghz is synthetic, it's not a benchmark.

You may say this tests overall system performane and is not just a CPU bench mark but I say the desktop system acutally has the advantage since it has a 7200RPM hd vs 5400RPM hd

That is correct you are testing overall system performance, I tested the core of the CPU. Go ahead and say the desktop has the advantage, we still aren't talking about anything near the same type of test. Apples and oranges....

And yes you could be just as abritrary (lol) as me, but when you do realize that my test was on the CPU core only not the entire system. So when you use that multiplier make sure you let everyone know it is valid only for those laptops you decided to test. Richie, read the whole post and get a feel for what it is about before you post.... I tested the CPU core only. You are obviously having problems understanding that since you asked me what laptop I used.


If you guys want to tell me I am incorrect or point out flaws in the testing, that's cool. All I ask is that you read and understand what is going on before you do. That being said I should have posted the findings and just let it go, or never posted any info at all. The point of the post as Alex pointed out is a little more data for decision making. I am not asking you to use my multiplier I could really care less what multiplier you use. It's not my laptop you are buying you could use 12 as a multiplier if you want.

BTW Richie I'm still waiting for you to test your suggestion and let us all know what you find out.

 

[Haps]

I would be interetsed in seeing some results using the L2 cache.

 

[UltraWide]

There is a whole different world of computers if you venture out of PC World, PC Magazine, etc. Those are just mainstream magazines and nothing more. The scientific community would never touch Sandra to do any significant tests. Nonetheless, Sandra is valuable for the mainstream and it does its job well enough. So always think about other options before settling for something and making it a "rule".

 

[dnuggett]

You guys are killin me. Seriously, I just got done posting that this info is to be used as more data in your decision making process, and somebody comes in and tells me to think about other options before making it a rule.

I never should have posted in this forum, it appears as though every critic with the exception of Pandaren did not read what I posted, or lacks the ability to read and comprehend. Like I said I have no problem with critics, as I stated before I love em'... that's what helps build knowledge. The problem is people who type before reading, or in the more extreme sense, people who just lack reading comprehension.

I really don't know about the scientific community using Sandra or not, but I do know that the majority of us are not scientists and do not need tests of lab quality to make a consumers decision. We aren't being specific and accurate enough to test the molecular make up of the die here folks. Nor are we engineering the chip or preapring to have our findings published. Maybe a few of you out there have a lab setup in your house and are above Sandra....

But I do agree that before someone should accept an idea as a "rule" it should be tried and true. Seeing as how you have 3 notebooks with the M chip, I could see why you don't like the tone of this test.

 

[GnomeCop]

congratulations, you proved the industry wrong.

 

[ElFenix]

Originally posted by: RichieZ

Originally posted by: dnuggett

Originally posted by: RichieZ
so you base your multiplier on a synthetic benchmark? I would think a more valid method would be to use real world applications.

Yes as is almost every processor. How do you think they come up with a Ghz rating? It's the FSB times the chips multiplier, not how fast it can open Word. The synthetic testing is the most objective way to come up with an accurate multiplier.

But since you have suggested it, go ahead and do some subjective testing and tell us all what what you come up with.

I fail to see how FSB * multipler is a synthetic test. So in your thinking a Pentium 4 2ghz is the same as the Celeron 2ghz becasue their "Ghz rating" is the same?

a lot of consumers sure think that way, regardless of if it accurately reflects the real world

 

[UltraWide]

I have 3 notebooks, one of each kind:
X24 - Pentium III-M
T30 - Pentium 4-M
T40p- Pentium M

I am not sure what you were trying to say about my notebooks, but I like them all just the same.

 

[Frightcrawler]

yes, can't we all just stop arguing and agree that the pentium-m is great?

 

[dnuggett]

Originally posted by: Frightcrawler
yes, can't we all just stop arguing and agree that the pentium-m is great?

It is absolutely great...

 

[dnuggett]

Originally posted by: GnomeCop
congratulations, you proved the industry wrong.

Prime example...