A Real Pr Rateing System

[DrSpanky]

As you all know there is a PR rating system that AMD use that is meant to be in relation to Intel’s CPU's clock speeds

Well I think its all crap cause AMD can just make up and tweak there PR RATEINGS lie to make it seem there better

As INTEL change there CPU's and they are made faster and work better will AMD change the pr rating so it more accurate?

I think probably so, they will pick what ever is the highest rating to big up there chips.

Well why not find a way to give ever CPU a PR rating...

it would work out what’s better more work per cycle or raw MHZ.

Just as u know we can test our whole system on 3d mark to get a how the system performs when doing 3d work. And get given a score, so why don’t some company make something that can test the CPU and give it a SCORE (A PERFORMANCE RATEING) then this can be used buy manufacturers of the cpu’s to market there CPU's

This way if they bring out a 1ghz CPU that dose 5x the amount of work a p5 4ghz u can have an official rating. of how it will per perform against each other in the form or a PR rating.

Any way this is just an idea I though at while I was at work board out of my mind any post your ideas or moan about how inaccurate I am or the spelling mistakes :p


edited to make it should better cause the last version was crap and so's this but what do u expect from a moron?

[Hedgehogforprez]

:confused: English please? All I heard was rambling....

[DrSpanky]

is that better now ?

[Hedgehogforprez]

Sort of I guess....my thoughts on it is what other companies would there be to run the program??? There are only 2 main companies people are buying from. AMD and Intel. If there was more than one, yeah that would be a great idea. Since there is only 2 right now I think the way things are is working out good for each company.

[ninthebin]

still a bit illegible, but giving EVERY cpu a PR rating is a bit dumb, as PR rating obviously co-incides with clock speed - the only reason you have the PR rating is the marketing department within AMD pointing out not everyone sees why there is a MHz gap.

I dont know about others, but when I look at AMDs latest PR ratings they mean squat didly to me, I have to go look at benchmarks, particulary those that compare the PR to various P4s at various clock speeds. So when I look at my XP2500 I dont think, thats equivilant to a P4 2.5GHz, I look at the various benchmarks and see hmm it seems to perform on part with the 2.66 there, and oo its only just scraping on the 2.4 in that department.

Now you have that whole "but its PR is based on performance vs thunderbirds" but its obviously just "get the next PR the same as the latest P4 Clockspeed" - without the P4 clockspeed, AMDs PR rating system would be negated.

Although that said I suppose you could just compare PR vs PR, but as PRs are just made up, easily changed to suit competition I think it would all be overly confusing and meaningless.

Dan

[FunkDaMonkMan]

The PR rating system isn't comparing AMD cpus to Intel cpu's. AMD is only comparing to themselves, to show how the "new" XP processors do more work per clockcycle than the origional t-bird. The t-bird stoped at 1.4ghz, so they had to continue off of that. a 1.4ghz XP processor did the work of a theoretical 1.6 t-bird to they named it the xp1600.

[gamefoo21]

ahh dis be true but still people look at 3200+ and think it must be faster than a 3.2ghz p4 when in reality if it was based on comparison with a p4 it would be a revised 2800+. so really amd is playing the same game with joe blow as intel is when they jacked up the mhz but dropped the ipc.

[breez]

gamefoo21: Blame the people then ;)

[NookieN]

I think you'd have a hard time coming up with a PR system that worked for all CPUs. Different CPU architectures can handle the same code differently. So essentially you would need a benchmark that works on all CPUs and is able to prey on both their strengths and weaknesses.

SPECint and SPECfp come to mind, but as Apple recently proved for us there are a lot of problems with SPEC benchmarks. They are very dependent on the compiler and the code optimizations used. Each company would have to take identical copies of SPEC, tweak it for their own architectures, run the benchmarks, and the submit it back to SPEC for approval.

This is a rather long and expensive process. I don't think either AMD or Intel would want to go to the trouble every time they release a new speed grade. Plus most end users probably wouldn't care.

[DrSpanky]

What I’m thinking of is a way to benchmark any CPU. I don’t know how it could be done, but what I’m trying to say is no matter what the clock speed or the amount of work dun per cycle on any CPU if we had some sort of PR system then no matter the design of the CPU we could have away of getting 1 simple number telling us what CPU is the best.

[aznchaos]

There is a program that gives you a pr rateing. Its called GCPUID. A memeber on this forum made it.

[aznchaos]

I found the link to the thread here it is
http://forum.oc-forums.com/vb/showthread.php?s=&threadid=161382

[Brunt]

Yes, that program is in my sig, It would be nice to have 1 program and be able to compare my AMD vs Intel in a real world way, I use Sandra atm to benchmark my CPU. My PR rating is 3000(from The Coolests Program) at my most current stable speed, but in Sandra it says I am faster then a pent4 3.2ghz.

I am so confused....:(

[David]

Okay, if you made a PR rating, what would you bench it on? Cache bandwidth? FSB bandwidth? FLOP performance? Integer Performance? Some CPUs would beat others at some of these, but others would excel in other areas.

If one CPU gets, say 78 marks in FLOP performance and 67 at Integer performance; another CPU gets 70 for FLOP performance and 74 for Integer performance.

Does that mean because CPU1 has an overall higher score that it excels at everything, better than CPU2?

No.

CPU1 is better at FLOP performance. CPU2 is better at Integer performance.

The only real PR system is real world performance. Gaming benchmarks, scientific computations.

Want a server to run fileserving? Compare which CPU is better at such tasks, not just by going "oh, it gets a higher PR score".

PR and benchmarks will never be perfect. It doesnt bother me: I can guestimate that my 1.73GHz "XP2100" CPUs are probably going to match or beat 2GHz P4s in most situations.

[aznchaos]

I think GCPUID is alot more accurate then sandra in pr ratings. Sandra makes our pr rating alot higher then what they are.

[Ffats]

I really think we should let apples be apples and stop comparing the two processors. I believe it is a futile task, as there are just too many variables involved.

Whatchu think?

[DaddyB]

Originally posted by aznchaos
I think GCPUID is alot more accurate then sandra in pr ratings. Sandra makes our pr rating alot higher then what they are.

Sandra is using a Pentium as the baseline (not Pentium 4, not a P3, not a P2... a pentium) so the numbers represent what a pentium would need to run at to equal your performance.

AMD says the PR rating compares the xp to a tbird but we know why it's really there - to give consumers an idea of how their chips perform compared to P4s. You can quote what AMD says (and no one will deny what AMD says) but the PR rating is not a comparison to the tbird, most anyone who has looked into it knows that. Most people see the numbers, they don't know what the number means except that higher is better.

The problem I see with the AMD PR rating is that it is old. It compares the xp to a P4 willamette, when you compare equal clock speed/PR rating willy to XP the XP will come out on top most of the time. AMD has stuck with that rating but since then the P4 has doubled it's cache, shrunk the die and increased FSB speed BUT their CPUs are still just rated with MHz/GHz. AMD incorporates the increased cache and FSB into the PR rating so if you compare an XP3200 to a willy running at 3.2GHz (and good luck with that) the XP will win in the majority of benches, when you compare an XP3200 to a P4 3.2C the P4 wins the vast majority of the benches.

I think AMD should have lowered the PR and or raised the clockspeed (at a given PR rating) to balance against the new P4s, the gap between an XP3200 and a P4 3.2C is larger then it should be IMO.

[Elephanthead]

Can someone give me a link to GCPUID that I'm not blocked from viewing cause I don't post 100 times. Thanks in advance.

[aznchaos]

http://www.thecoolest.zerobrains.com/apps/GCPUID16.6.zip

[modenaf1]

for PR rating, IMO, they should be based of off benchmark scores, divided by a certain number to resemble the actual Ghz number.

ex. intel 3.2 ghz gets a bench of 13,000. 13,000 divided by 4000 = 3.4 pr rating

AMD at 1.5Ghz gets a bench also of 13,000. it gets a pr rating of 3,4 as well.

what do you all think of this system?

[NookieN]

Originally posted by modenaf1
ex. intel 3.2 ghz gets a bench of 13,000. 13,000 divided by 4000 = 3.4 pr rating

AMD at 1.5Ghz gets a bench also of 13,000. it gets a pr rating of 3,4 as well.

what do you all think of this system?

I think it doesn't work if it tells people that a 1.5Ghz Athlon is equivalent to a 3.2Ghz P4...

[modenaf1]

Originally posted by NookieN


I think it doesn't work if it tells people that a 1.5Ghz Athlon is equivalent to a 3.2Ghz P4...

well, that was an example, i just made it all up. but if it was under real benchmark testing, then it would work. I think.

[texp4]

Originally posted by gamefoo21
ahh dis be true but still people look at 3200+ and think it must be faster than a 3.2ghz p4 when in reality if it was based on comparison with a p4 it would be a revised 2800+. so really amd is playing the same game with joe blow as intel is when they jacked up the mhz but dropped the ipc.

Good point. I am wondering this for a long time to

IS Athlon XP3200+ equivalent and as fast ( or faster ) than P4 3.2Ghz given both runs at stock speed. ????

My guess is NO but never know the truth

[gamefoo21]

no it isn't most of the bench's i have seen a 3200+ can just give a 2.8 a run for its money with the 2.8 still winning in some spots.

[NookieN]

Originally posted by modenaf1
well, that was an example, i just made it all up. but if it was under real benchmark testing, then it would work. I think.

In theory it should work just fine. But in practice, I think it would be nearly impossible to find a benchmark (or suite of benchmarks) that fairly compares chips of different architectures and gives you one absolute metric of performance. Some benchmarks are AMD-biased, some are Intel-biased, and some are just crappy benchmarks.

Plus, everytime a new benchmark revision comes out, all test on all chips must be re-run, or some scaling factor must be determined. Finding such a scaling factor would be no easy task. It would probably be somewhat arbitrary and would inevitably lead to giving certain chip architectures a handicap.

[ninthebin]

modenaf1, as nookien says, you can make what your going to do with the numbers work, its getting the numbers thats the problem.

I think one company running clockspeed, and the other using a marketing ploy is just fine.

[allen337]

i think toms hardware says it best about amd vs. intel. ~~~ http://www6.tomshardware.com/cpu/20030623/index.html

[dropadrop]

I find it hard to believe that tom would say it best... :D

[imgod2u]

As software is constantly changing, there is no real way to make a "definitive benchmark". The "ultimate benchmark" from 5 years ago is vastly different from the "ultimate benchmark" of today, if such a thing even exists. The question isn't whether a benchmark is "fair" or not, merely that it's relevent. That is, a lot of people actually depend on performance in such application. Does it matter whether Sandra gives you a better score? Are you going to spend most of your time running Sandra? Does it matter more than, say, your UT2003 scores are better? Your video encoding times are better? Definitely.
That's really the only way to determine "which is the superior processor" in terms of performance. Take commonly used software, test it on both platforms. That's what most sites do and that's where we draw conclusions. Are all consumers going to look at them? No. Is there a way to dumb down all that information to something that would capture the consumer's limited attention span without loosing information? Absolutely not.
Computers are complex things and performance is a complex thing. You can't make it simple without making hasty generalizations. Hence, this "educating the average Joe" thing is a lost cause.

[Lithan]

Bear in mind it's an intel user saying tom's says it best. ;)

[Gautam]

Comparing AMD's to Intel's is like comparing apples to oranges. You aren't going to be able to come up with a strict PR rating system that could work even 50% of the time; it simply isn't possible to generalize overall system performance from each. The PR rating system is based on the original Athlon, and this is about the best that can be done. You could go further by taking into account that AMD's perfom 1.28 more int calculations per clock cycle than Intel, but then you also have to keep in mind that Intels have earth-shattering memory bandwidth in comparison to AMD, not to mention SSE2. It just inevitably falls apart. AMD's PR rating system does what its meant to do; give the average Joe a rough idea of where AMD processors stand in relationship to their Intel counterparts.