Why is the doesn't the P4 compare to other CPU's clock for clock?

[Quickfingerz]

We've always seen CPU's clock for clock efficiency improve. Remember when the P5 first came out and how a 200 Pentium Pro would easily beat out a 233 MHz Pentium MMX or how a Athlon 500 MHz could easily beat out a K6-3 at 550 MHz. How is it that some P3 at 1 MHz benchmarks can beat out a P4 at 1.4 MHz which has a clock speed 40% faster? A K6-3 running at 500 MHz would never beat an Athlon running at 700 MHz at any clock speed.

Why didn't Intel improve clock speed as well as clock efficiency?

<edit> I'm not bashing Intel. I have complete respect for Intel. There are serious architectural issues that the P4 has. L2 latency (clock for clock, the P3's L2 is faster), small L1 (why have an L1 that is so small that most processes have to result to L2 anyway), L1 to L2 speed (yet again slower than the P3). This all bothers me. Yes, it probably will be likely that P4 clock speeds will increase dramatically (It might not be enough). I just hope that AMD will use a 256-bit to cache L2 in a new release of the K7 or the like to take the performance lead. That alone will compete with anything that Intel has to offer in their P4.

I know a lot of you hate RAMBUS but if AMD chips currently used dual channel RAMBUS they also would gain a nice boost of performance. Right now, AMD chips can't use the bandwith of SDRAM because the chips aren't fast enough. RAMBUS solves this issue. (Just a thought)

 

[NeonFlak]

Cause at this point in time the p4 sucks @ss. There are no programs right now that are optimized for the p4. Until that happens it will just be an overpriced piece of $hit.

 

[office boy]

Because that is Intel's dirty little trick. Do less work per cycle, so you can do more cycles and the same or less work. Then people think &quot;Oh look more cycles (Mhz) it must be faster&quot;.

 

[fkloster]

You freakin' idiots! Stop your &quot;clock for clock&quot; bashing of the P4! This chip doesn't perform clock for clock with AMD procs because IT WAS NEVER INTENDED TO!! Longer pipeline and superior branch prediction allow P4's to climb easily to speeds over 1600mhz while maintaining integrity. It doesn't have to perform on a clock for clock basis if the stinkin' chip runs @ 1800 mhz. This chip performs badly @ NOTHING!! Understand? If you think P4 architecture is $hit then you don't understand why it was desinged...FOR HIGHER CLOCK SPEEDS!!!!!!!!! It wasn't meant to run in 2nd gear. P4 1.3's &amp; P4 1.4's are under performers because they run to slow to take full advantage of their 20 stage pipeline. Anyone want to go head to head with a wide open retail P4 1.5 runnin' @ 1.8 let me know o.k.?

Office boy, my frustration is not directed toward you, its just seems like eveyones opinion around here is starting to irritate me...



<< Do less work per cycle, so you can do more cycles and the same or less work. >>



Actually, Intel's plan of less work per cycle, then doing alot more cycles figures out as MORE work done, currently...

 

[office boy]

OK (sorry I called you a whore)

But a big part of the P4 was intels need to retake the Mhz battle with AMD.
Now they have an easy way of fooling the average consumer into thinking that a 1.3Ghz P4 is faster then a 1.2Mhz Tbird.

 

[TravisBickle]

&quot;fkloster the RAMBUS impostor!&quot;

 

[JimboG]

Check out www.emulators.com/pentium4.htm for a very detailed and technical article on why the P4 is a POS.

-Jim

 

[MrGrim]

OMG, don't be so touchy. Some don't like the P4 architecture, including me. You like it, you bought it, good for you. Calling them idiots is not going to change their opinion. You are definately not going to get your point across by calling people names.

 

[fkloster]

<< ...Until that happens it (P4) will just be an overpriced piece of $hit. >>




<< ...Cause at this point in time the p4 sucks @ss >>



...anyone who makes the above remarks is an idiot. Sorry for calling him that but he is.

I'm not an Intel whore. I am sick of people runnin' their mouth how Intel's P4 is a piece of crap. I use one that runs every freakin Sandra benchmark to the hilt and any app I throw at it (mind you without SSE2 optimizations) very quickly. Clock for Clock comparisons are terrible. A freakin' John Dear tractor might nose ahead of a ferrari in the first 12&quot; of a quarter mile but big deal. Its only 12&quot;

<< ...but the tractor has a lot more foot pounds of torque under 2 miles an hour than the ferrari... >>

give me a break! (corny comparison but hopefully you git my drift...)

 

[Modus]

fkloster,

<< Actually, Intel's plan of less work per cycle, then doing alot more cycles figures out as MORE work done, currently... >>

Totally false: Currently, the vast bulk of real world benchmarks (even those designed to be biased in Intel's favor) show a P4/1.5 accomplishing less than an A/1200/DDR. And seeing as how the P4 system will cost almost twice as much money, where does that leave Intel?

Unfortunately, Intel's behemoth marketing division has the perfect solution: market clock speed above true performance. That way, even though a P4/2.0 has no chance in hell of beating an A/1600, ignorant people will buy based on the bigger number and the bigger name.

Modus

 

[TravisBickle]

the P4 has been done to death. there are plenty of interesting articles on it. I think the Emulators, inc. article on it demonstrates a programmer and chip designer's view with quite a lot of weight. sure, the p4 has strengths, but also comes with drawbacks, like the fpu. why rehash, when those reviews can talk more cogently?

 

[Modus]

fkloster,

<< I'm not an Intel whore. I am sick of people runnin' their mouth how Intel's P4 is a piece of crap. I use one that runs every freakin Sandra benchmark to the hilt and any app I throw at it (mind you without SSE2 optimizations) very quickly. >>

And how, exactly, is this different from any other x86 CPU in the history of the universe? Oh, forgot, the P4 does it &quot;very quickly&quot;. Right. That'll look great on AnandTech's next CPU review:

&quot;In conclusion, the Acme CPU's record breaking Winstone score of &quot;very quick&quot; brings performance to new hights. Not only that, but we threw every freakin Sandra benchmark at it and ran it to the hilt, and it actually worked as it was supposed to. Stunning!&quot;

<< Clock for Clock comparisons are terrible. >>

There are only two valid kinds of comparisons: clock for clock, and dollar for dollar. The P4 looses both. End of discussion.

Modus

 

[fkloster]

<< ...And seeing as how the P4 system will cost almost twice as much money (as a DDR setup).... >>

-Modus

Where for the love of God did you dream that one up Modus? One word for you: DELL



<< There are only two valid kinds of comparisons:... >>

-Modus

Nice opinion...keep runnin' your mouth

Int ALU/RAM bandwidth/s 1582
Float FPU/RAM &quot;&quot;&quot;&quot;&quot;&quot;&quot;&quot;&quot; 1618

CPU Drystone ALU MIPPS 3376
CPU Wetstone FPU/SSE2 MFLOPS 1077/2190

Yup, Sandra scores from my P4. Funny how the Doctors results &amp; mine vary... What a piece of crap this chip and its platform are. I'm sure a High Flyin DDR/1.2 setup Modus is refering too would absolutely double these scores

 

[JimboG]

Dell heh. You actually think those Dell P4 systems for $1500 are of any quality? What are you smoking. Tom's Hardware sorta blew that out of the water with his latest review.

-Jim

 

[office boy]

YEAH Hmmmmm TNT2 or GTS???

 

[DaddyG]

Hmm, Dells 'low end' P4's are pretty crappy, see Dr Toms review. However, I agree that bashing the P4 is stupid. When we start to see the data streaming apps of the future, the P4 will shine. Believe me, there will be data streaming apps, Intel and Microsoft always work together. When compilers get optimized for P4 and SSE2 we will start to see the P4 shine. But, for today, AMD has the best price performance, and considering that technology move so fast, and prices drop like stones, I wouldn't invest in P4 yet.

 

[TravisBickle]

? have you actually looked at the fpu? sse2 is not an excuse for castrating standard application performance. it's a step backwards and it hurts.

 

[Modus]

fkloster,

Those numbers must really make you happy when you play Sandra all day long. . .

Those of us who work and play in the Real World, however, will be content that our lowly AMD processors will best Intel's latest and greatest in the vast majority of real world benchmarks.

Don't make me link to myriad reviews that show the P4 getting its chain yanked by the Athlon. Please, give up now. Make it easy on yourself.

Modus

 

[AmazonRasta]

Here are my SiSoft Sandra scores for my Thunderbird 1.0GHz @ 1.224GHz

Dhrystone ALU 3397 MIPS
Whetstone FPU 1661 MFLOPS

For being 600MHz faster than my chip, you're not beating my CPU scores by a whole hell of a lot.

 

[kombatmud]

<< Remember when the P5 first came out and how a 200 Pentium Pro would easily beat out a 233 MHz Pentium MMX >>



That's simply not true. The P6 (Pentium Pro) beat the pentium only in 32bit software, the software it was optimized for. At the time, the only operating system (other than unix/linux) available that was entirely 32 bit was Windows NT (3.51 I believe). Windows 95 was not completely 32 bit (it's largely 16 bit as well), and the pentium MMX outperformed the pentium pro consistantly on that environment. This is extremely similar to the Pentium 4 vs. Pentium III where the Pentium 4 outperforms everything in software that is optimized for it, and is consistantly outperformed on everything else. The P6 architecture ended up edging out the P5 because of it's ability to scale to much higher clock speeds, and the industry moving more to 32bit. When Pentium 4s get to 2GHz+, they will almost definately outperform the 1GHz PIII, with or without SSE2 optimization. That's why intel is making the move to Pentium 4, because they have to in order to get higher clock speeds. The only difference between this situation and the P6 situation is that the Pentium pro was never intended for the desktop market, it was designed for servers that ran 32 bit O/Ses and had big enough budgets to pay for them (the Pentium pro was significantly more expensive than the Pentium MMX, much like the Pentium 4 vs. Pentium III).

 

[AmazonRasta]

TTT for more P4 bashing!!!

 

[fkloster]

<< vast majority of real world benchmarks >>



O.k. Modus. Define please...then I'll get whatever realworld benchmark your runnin' your mouth about; take it home run it on my machine and post what I get. Deal? (Hint: Anands and a few other reviewers that your thinking about that jumped the gun on early i850 Bios revisions [i.e. 1001 &amp; 1002] benchmarks show innacurate current results. You wouldn't care about this because you are less concerned about accuracy as you are in slamming a company. You are in for a big surprise...) P4's are not crap. People who say they are, are idiots.

 

[MikeyP]

I find it funny that in the Drystone and Wetstone benches posted, the 1.224Ghz Tbird beat the 1.8Ghz P4

P4:
CPU Drystone ALU MIPPS 3376
CPU Wetstone FPU/SSE2 MFLOPS 1077/2190
The score of 2190 is irrelevant at this time, there is no software out there taking advantage of SSE2 yet. So, P4's FPU should be 1077, not the 2190.

Tbird:
Dhrystone ALU 3397 MIPS
Whetstone FPU 1661 MFLOPS

Amazon, what platform are you running?

On to bigger and better things,

<< You freakin' idiots! Stop your &quot;clock for clock&quot; bashing of the P4! This chip doesn't perform clock for clock with AMD procs because IT WAS NEVER INTENDED TO!! Longer pipeline and superior branch prediction allow P4's to climb easily to speeds over 1600mhz while maintaining integrity. It doesn't have to perform on a clock for clock basis if the stinkin' chip runs @ 1800 mhz. This chip performs badly @ NOTHING!! Understand? If you think P4 architecture is $hit then you don't understand why it was desinged...FOR HIGHER CLOCK SPEEDS!!!!!!!!! It wasn't meant to run in 2nd gear. P4 1.3's &amp; P4 1.4's are under performers because they run to slow to take full advantage of their 20 stage pipeline. Anyone want to go head to head with a wide open retail P4 1.5 runnin' @ 1.8 let me know o.k.? >>

fkloster, you need to chill a little bit. I do agree with the clock for clock thing, we need to compare the performance--not clock speeds--of available chips. That means 1.2Ghz Tbird vs 1.5Ghz P4. Who cares if it can run at 1800mhz? Right here in my possesion, I have a 10000 petahertz processor capable of 0, that's right 0 operations per clock. Why do you keep saying that the branch prediction has something to do with clock speed? Unless I am gravely mistaking, branch prediction has little to do with it. Yes, the P4 was designed for higher clockspeeds, but that does very little in today's applications! By the way fkloster, will you please start using real world benches, not synthetic ones Have a nice day!

 

[fkloster]

<< fkloster, you need to chill a little bit. I do agree with the clock for clock thing, we need to compare the performance--not clock speeds--of available chips. That means 1.2Ghz Tbird vs 1.5Ghz P4. Who cares if it can run at 1800mhz? Right here in my possesion, I have a 10000 petahertz processor capable of 0, that's right 0 operations per clock. Why do you keep saying that the branch prediction has something to do with clock speed? Unless I am gravely mistaking, branch prediction has little to do with it. Yes, the P4 was designed for higher clockspeeds, but that does very little in today's applications! By the way fkloster, will you please start using real world benches, not synthetic ones Have a nice day! >>



1) Branch prediction becomes very important when the pipeline gets longer. Reason? The longer the pipeline, the oportunities for incorrect prediction &amp; error. Therefore, P4's are given a totally new &amp; superior branch prediction mechanism that enables longer pipelines with little to no errors. Why must we have a longer pipeline? To increase mhz. Understand?

2) P4's higher clock speed does very little in today's apps? Incorrect. Give me one example of a current &quot;Real World&quot; application that runs slow on my P4 please...

3) I suppose if P4 performed badly @ synthetic apps and soared in real world apps, then you would all be crucifying Intel for their poor Sandra scores

4) Why is it no one brings up the issue of memory bandwidth and efficiancy? Not an issue any more? No one cares? Or is it that the P4 so blatently dominates and crushes all competition so completely that eveyone just aviods the subject in its entirety? Is it possible that a chip that maniacly bludgeons all competition with such absoluteness could be crap and/or a piece of $hit? Perhaps; but my opinion is that it is not.

5) SSE2 doesn't matter? WTF?

 

[DaddyG]

Travis, about 10% of todays apps make extensive use of the traditional x87 FPU. SSE and SIMD use FPU functions but in a totally different mode. x87 apps will simply fade away, Intel made this very clear quite along time ago.