Is it fair to benchmark P4+RDRAM vs AthlonXP+DDR?

[SteelCityFan]

Originally posted by: majewski9
Is it fair to compare a paper launched P4 2.8ghz to a paper launched AthlonXP 2600+ ?

Umm, going to www.dell.com shows me I can very easily buy a P4 2.8Ghz system.

 

[SteelCityFan]

I don't care what is Officially supported. I want to see AMD's fastest combo vs Intel's fastest Combo. To determine a performance leader you can't cripple a P4 with DDR266 just because the AMD is stuck at a 266 FSB. It is not Intel's fault that the highest speed memory technologies are not available to AMD users.

I have not been keeping up too much with the processor market and development over the past few months, but I do believe that the Hammer is supposed to have an on board memory controller. This will give it an advantage... does that mean we should ask AMD for a non Memory controller hammer so we can get the true processor power comparisons? No.

If I want the fastest computer possible, I want the FASTEST setups tested. Give me the fasted AMD chipset/RAM vs the fastest P4 chipset/RAM. I don't care if it is DDR vs Dual Channel DDR vs PC 800 vs whatever. If I want the fastest setup possible, and if AMD can't give it to me because they might not have the RAM capability, then they need to get to work on making that possible.

Even if you could get PC1066 on an AMD chipset, it would do no good because AMD is still playing with a 266FSB.

This thread is basically an AMD fanboy mad because Intel has the performance crown.

 

[SteelCityFan]

Also, why are the use of SSE2 enabled benchmarks unfair? Majewski9, are you trying to tell me that no one out there uses Lightwave, 3ds Max, or other SSE2 enabled programs? Do you think that Newtek (the makers of Lightwave) or Discreet (the makers of 3ds Max) put SSE2 code into their programs just for the sake of making it look better on the P4 architecture? Give me a break...they use SSE2 code to help consumers get the most out of their programs.

It will only be unfair until AMD copies the SSE2 support in their next line of chips, then, SSE2 will be the best thing since sliced bread to him.

 

[majewski9]

AMD is incorparating SSE2 into the Hammer series. That was annouced last october. AMD can add SSE2 to the AthlonXP which I have heard many rumors that they would. As for all the people critizing the paper launch of the AthlonXP 2600+ well Do you all remember the P3 1ghz launch? Yeah Intel perfected the paper launch to remain competitive! AMD is following suit!

I would also like to see cost specific benchmarks! I mean its like comparing a dodge viper to a neon when people compare P4 and the Athlon!

A lot of you people are missing the point of this post. Shouldnt two CPU architectures be compared fairly? they arent when one company deliberately manipulates benchmarks so they favor their new CPU. How can people reasonably trust cpu benchmarks ever again?

Oh and does anyone hear that Intel is being sued for the quote poor performance of the p4? At least AMD never had a problem improving their architecture performance.

 

[Macro2]

Isn't DDR II due in a few months? I'd like to see Scambus put to rest once and for all.

 

[SteelCityFan]

Originally posted by: majewski9
AMD is incorparating SSE2 into the Hammer series. That was annouced last october. AMD can add SSE2 to the AthlonXP which I have heard many rumors that they would.

That is what I said. I would be willing to bet that once they do finally copy Intel's work in this area, that none of the AMD fanboys will be complaining about the added performance it gices to their chips...

As for all the people critizing the paper launch of the AthlonXP 2600+ well Do you all remember the P3 1ghz launch? Yeah Intel perfected the paper launch to remain competitive! AMD is following suit!

The reference to this was in regards to many sites declaring that with the 2600, the AMD chip was now right up there with the best P4 (2533). They failed to mention that the 2800 P4 would actually be available a month before the 2600+ they reviewed.

I would also like to see cost specific benchmarks! I mean its like comparing a dodge viper to a neon when people compare P4 and the Athlon!

Cost is easy to figure out on your own. You mean to tell me you can't determine if the price is worth the performance on your own? You get what you pay for in terms of the cars and performance. The people who can afford a Viper are getting it because it is one of the (if not The) fastest street legal stock cars. The people paying for a P4 2.8Ghz and PC1066 are buying it because it is fastest. That is what they want to know. I would hope you are intelligent enough to do your own cost / Performance analysis based on the published performance and price numbers. Get the fastest system for the money you have to spend. People don't need the ever changing prices hard written into an article.

A lot of you people are missing the point of this post. Shouldnt two CPU architectures be compared fairly? they arent when one company deliberately manipulates benchmarks so they favor their new CPU. How can people reasonably trust cpu benchmarks ever again?

This is the AMD fanboy in you talking. How has Intel manipulated the benchmarks? FPS, seconds, etc don't lie. I fal to see how they are manipulating anything by designing fast chipsets and designing them to take advantage of the fastest memory on the market. AMD's FSB is a slow 266Mhz. They can't even fully utilize DDR333. That is no-one's fault but their own. There is nothing wrong with benchmarking the fastest P4 setup with the fastest AMD setup and declaring a speed king. Like someone mentioned a few posts up... not one AMD fanboy was crying when the Williamette used RDRAM... because it was losing.

Oh and does anyone hear that Intel is being sued for the quote poor performance of the p4? At least AMD never had a problem improving their architecture performance.

Did you also hear that McDonalds and other fast food places are being sued for get this... Making someone Fat. What is the basis of this court case when the P4 is the fastest CPU available.

 

[majewski9]

SteelCityFan
1st off AMD will have 333fsb next month and 512kb L2 cache. 2nd the suit brought up against Intel is worth mentioing since it so unprecedented and they have a good chance of winning. 3rd AMD is suing bapco and they will win. I mean using SYSmark 2002 you would see the p4 1.4 ghz willmette trounce a tualatin p3 I mean gimme a break. Yeah and when the willmette was out the only Intel chipsets out supported SDRAM and RDRAM.

Yeah I wish AMD would continue to make its chipsets since relying on VIA is a bad choice in my opinion. Although Nvidia is showing some good potential.

Yeah gaming benchmarks dont mean anything to me since the difference is negligible. No person can tell the difference between a couple frames. Quake3 loves the p4 Ill give you that. Unreal loves 3dfx does that mean the Voodoo 5 was a better graphics card than any Nvidia GF3 card? hmm youd be crazy to say that. Well lets see anything that benches actual CPU performance and not memory bandwidth will show an Athlon lead.

 

[andreasl]

Yeah gaming benchmarks dont mean anything to me since the difference is negligible. No person can tell the difference between a couple frames. Quake3 loves the p4 Ill give you that. Unreal loves 3dfx does that mean the Voodoo 5 was a better graphics card than any Nvidia GF3 card? hmm youd be crazy to say that. Well lets see anything that benches actual CPU performance and not memory bandwidth will show an Athlon lead.

Why don't we just bench real applications (and games) instead of picking those "that benches actual CPU performance" to try and justify a platform of choice? If the Athlon performs worse in a game because of lack of bandwidth, does that mean we should discard this game then? What if you wanted to play this game at the best performance? Forget Quake3, there are lots of other games where the P4 beats the Athlon in currently. One of these days you will realise that CPU performance is also dependant on the memory subsystem. Think of that as the first step to a higher enlightment...

 

[jbond04]

GuYz!!!1!! Teh P4 has 512K cache! This is so unfair!!!! They should disable 256 of teh cahce and make it's path only 64 bits wide, since the ATHLON XP is like that. Then it would be even more fair!!!!!!1! A true test of CPU power!

I wouldn't be surprised if that was the next argument around here.

Majewski9, the P4 was designed to take advantage of a high bandwidth memory subsystem. Do you understand what that means? It means that it was meant to run with RDRAM of dual channel DDR. To run it with anything less is crippling the P4. You obviously understand that, since that seems to be what you are aiming for. If for some ungodly reason Via made a chipset that supported RDRAM for the Athlon, and we benched an AthlonXP 2600+ w/ PC1066 RDRAM and a 2.8GHz P4 w/ PC1066, the P4 would still win. Do you understand why this is? :disgust:

EDIT:

Andreasl wrote the second half of my post for me. 😛

 

[SteelCityFan]

So AMD will finally be able to take advantage of DDR333. Good for them... The FSB is still too slow to take advantage of RDRAM, and chipset makers know this which is why you don't see any PC1066 AMD chipsets. That is AMD's own fault.


I don't see the merit for a lawsuit that sues the maker of the best performing CPU because of poor performance? I guess AMD is next in line to be sued followed by via and whoever else makes CPUs. But I guess anything is possible when someone wins 2 Million because they spilled hot coffee on their stupid selves.

You seem to be really mad that an Intel based system can beat an AMD based system. You really need to get over it. The bottom line is that if you want the fastest system available, you are going to have to buy Intel and PC1066. If you want to save a few bucks and have less performance, buy AMD's best.

 

[majewski9]

Hmm I was just hoping that people would be smart to realize that I was talking about actual cpu performance. Sandra shows how much stronger the Athlon is in terms of raw calculations. Something the p4 can never catch the Athlon even if they cheated like in SYSmark 2002. Sandra may be synthetic but its a very good benchmark.

Yeah like I said before gaming performance depends much more on what graphics card you are using than CPU ( once your past 1 ghz ).
The couple frames the P4 Northwood has over the AthlonXP is simply unimportant when measuring CPU performance. I see gaming benches as just article filler especially I play very little 3d games anyways.

 

[Askheart]

Why don't we just bench real applications (and games) instead of picking those "that benches actual CPU performance" to try and justify a platform of choice? If the Athlon performs worse in a game because of lack of bandwidth, does that mean we should discard this game then? What if you wanted to play this game at the best performance? Forget Quake3, there are lots of other games where the P4 beats the Athlon in currently. One of these days you will realise that CPU performance is also dependant on the memory subsystem. Think of that as the first step to a higher enlightment...

There are lots of games where the Athlons fare better than the P4's too.

The question is, there are bandwidth-bound games (the old Quake 3 is the foremost known example) and cpu intensive engines. Athlon does not have issues running the majority of the newer games faster than P4 in a per clock cycle basis, even going beyond its performance ratings.

Just to highlight, an Athlon XP 2200+ processes the UT2003 test faster than a P4 2.4 GHz/PC1066. And UT2003 is way more recent than Quake3 (UT2003 isn't even out!).

Newer games tends to be more and more cpu intensive (the evolutionary steps in gaming always were to stress the cpu up, because of the huge cpu/cache bandwidth that lies past the front side bus), and this fact adds value to the Athlon platform, hence Athlon cannot dispose from great DDR bandwidth, but its raw processing/computational power simply rocks 😎

 

[andreasl]

We know what you mean, but we are telling you that you are wrong. You cannot isolate CPU performance without taking the memory subsystem into account. There is not a single application that runs without accessing the memory. It's not possible! And I'm only aware of a single REAL application that runs within the CPU caches (this is not the same as not accessing memory) and it's the distributed.net client. You can quote toy benchmarks like Sandra ALU/FPU all day long if you want, but those have no bearing on real world performance.

Let's put it like this.. If AMD would put the Athlon XP on the 486 bus (32 bit wide @ 33MHz ~ 256MB/s bandwidth) this would not affect the Sandra ALU/FPU benchmarks much (less that 5% probably) but how much do you think it would affect REAL applications? As I said, this is the first step to higher enlighment.

 

[majewski9]

Well look way over there jbond04 there goes the point!

okay ill get redicolous. Lets take two AthlonXP 2600+ and through them in a dual AMD board. Lets bench them against the p4 2.8ghz! Also all the benchmarks will be CAD and database apps.

Okay you people are obviously got away from the point. Look PC1066 isnt supported by any chipset but reviewers use it. This helps the p4 beat up on the Athlon XP. People not knowing anything about PC's will get an unrealistic portrayal of the real performance of their P4 systems. I dont think this is fair to AMD! Maybe not as quite as unfair as SYSmark 2002, but it is still an advantage to the p4.

 

[andreasl]

Askheart,

You are doing the common mistake of crediting the Athlon to higher CPU performance where it wins and P4 to higher memory performance where it wins. Let me ask you this. Do you know the first indicator for when a benchmark is memory bandwidth limited? I'll give you a hint, it's not automatically when the P4 wins...

 

[Askheart]

Hmm I was just hoping that people would be smart to realize that I was talking about actual cpu performance. Sandra shows how much stronger the Athlon is in terms of raw calculations. Something the p4 can never catch the Athlon even if they cheated like in SYSmark 2002. Sandra may be synthetic but its a very good benchmark.

This is synthetic, and andreasl beat the question to a pulp.

For example, Sandra reports Hyperthreading boosts P4 performance in ~50%. Intel itself says that Hyperthreading could provide a 20% boost overall, in a turndown, Intel might advocate Sandra which provides best results, but is known that the real-world performance of Hyperthreading doesn't match the synthetic results.

 

[Askheart]

You are doing the common mistake of crediting the Athlon to higher CPU performance where it wins and P4 to higher memory performance where it wins. Let me ask you this. Do you know the first indicator for when a benchmark is memory bandwidth limited? I'll give you a hint, it's not automatically when the P4 wins...

For the issue, everything is memory bandwidth limited. What happens is that the newer games of course demand more DDR or rdram bandwidth, but the step of getting most of the cpus using the caches' bandwidth always was the logical thing to do, thus the most recent games were designed for it. No DDR or rdram matches the L1 or L2 bandwidth.

There is no way, even using benchmarks, to acknowledge the exact factors why a game runs better in platform 'x' than in platform 'y'. Could it be the memory bandwidth, memory latency, the way the game was programmed, optimizations for a specific cpu, etc.

The only factors we can denote is that Athlons have consistently showed higher raw computational power (which does play a role, indeed) and that most apps were more optimized for the P4 than for the Athlon. And Athlons still beat P4's in a per clock cycle comparison under most situations.

Of course DDR or rdram bandwidth are factors income.

But as you said the Athlon on a 486 bus is rediculess, so as a 486 on a qdr bus and rdram is, in the same manner.

 

[SteelCityFan]

Originally posted by: majewski9


Okay you people are obviously got away from the point. Look PC1066 isnt supported by any chipset but reviewers use it. This helps the p4 beat up on the Athlon XP. People not knowing anything about PC's will get an unrealistic portrayal of the real performance of their P4 systems. I dont think this is fair to AMD! Maybe not as quite as unfair as SYSmark 2002, but it is still an advantage to the p4.

The motherboard makers support it. That is all that is needed. Give an AMD PC1066, and they still would not be able to beat the P4 because the chip could never take full advantage of it.

 

[andreasl]

Actually there are hardly any applications/benchmarks that are memory bandwidth limited. The best way to describe that is by looking at the benchmark results. A memory bandwidth limited app would gain no performance by simply raising the CPU clockspeed. But it would gain 100% from adding memory bandwidth. If you look carefully you'll see that most apps gain about 3-8% in performance for every 10% increase in clockspeed. That is the normal range. A bandwidth limited application would gain 0%. The best one for that is STREAM because it was designed this way to be completely bottlenecked by the memory bandwidth.

You are right when you say it's very hard to determine whether an application gains performance from reduced memory latency, memory bandwidth. This is why synthetic benchmarks exist. These are made to stress a single component in the system only. Which is also why they cannot be used to gauge real-world performance because real applications don't behave that way. And quoting them to prove a point about CPU performance is useless.

 

[Askheart]

To rephrase my statement, I said everything is bandwidth limited in the sense of the cpu internal registers and caches, which clock at the same speed of the cpu 😀

 

[dexvx]

Sandra gives the best CPU performance possible. A P4 + SDR SDRam with 100FSB will be the same as a P4 + PC1333 RDRam running on a 166FSB. In real world there will be differences, but not in Sandra. I do not think it is unfair to pair it with PC1066 RDRam seeing these are enthusiast sites and people who look at them clearly know what they're getting [test setup info on each reviewer]. If you want totally in spec benchies, then go look at PC Magazine where they bench a Pre-Made Compaq against Pre-made Dells and gateways, etc.

The only factors we can denote is that Athlons have consistently showed higher raw computational power (which does play a role, indeed) and that most apps were more optimized for the P4 than for the Athlon. And Athlons still beat P4's in a per clock cycle comparison under most situations.

Really? Go look at LightWave. Thats P4 optimized and no non-SSE2 CPU can stand up to that. A 1600 Northwood trounces a 1677Mhz XP (2100+ !). There are HARDLY any SSE2 optimized programs out there as of now. If most apps were optimized for the P4, AMD would be literally out of business. Then again we might be talking about different kinds of "optimizations" (and compiling a program through the Intel compiler isnt optimization IMO).

Hmm I was just hoping that people would be smart to realize that I was talking about actual cpu performance. Sandra shows how much stronger the Athlon is in terms of raw calculations. Something the p4 can never catch the Athlon even if they cheated like in SYSmark 2002. Sandra may be synthetic but its a very good benchmark.

Really? You ever see Single Xeon (same thing as a P4) Sandra with hyperthreading? Phenomenal increase. I guess by your rationale, a P4 with hyperthreading will waste an XP in terms of raw CPU performance. In my rationale says it wont because Sandra gives scores for the most optimal of conditions and is sometimes useles (hint MIPS = meaningless indication of processor speed).

At a recent LAN party, there was a guy with a 2000 XP system that was similarily configured [ti4200/512MB/7200 rpm HDDs/WinXP Pro] as my 2133 / PC1066 system. He had the balls to actually bet Hardware that his 2000 XP system could own my system in 3dMark2001SE (his proclaimed god of all benchmarks ... ). He even said RDRam hindered my system's performance (he must've lived in a cave since the i820/RDRam days) and would give me a 5% leeway because of the rambus. Well guess who won that benchmark. Now he claims 3dMark2001SE is a biased benchmark.

Yea XP's are good CPUs, but as you scale them higher, they're showing their age. Their unbeatable performance in the Willamette days are long over. Now we'll see how 512KB L2 cache and a 166FSB will do.

 

[MadRat]

Just ban advertisements that link P4 w/RDRAM performance to P4 systems using disimilar memory, notably DDR. It would save alot of hassle for consumers to weed through the truth.

 

[Askheart]

Sandra gives the best CPU performance possible. A P4 + SDR SDRam with 100FSB will be the same as a P4 + PC1333 RDRam running on a 166FSB. In real world there will be differences, but not in Sandra. I do not think it is unfair to pair it with PC1066 RDRam seeing these are enthusiast sites and people who look at them clearly know what they're getting [test setup info on each reviewer]. If you want totally in spec benchies, then go look at PC Magazine where they bench a Pre-Made Compaq against Pre-made Dells and gateways, etc.

Negative, Sandra cpu benchmark, especially the x87 FPU whetstone is very limited by "FXCH's" and it has no optimizations to deal with the FPU stack. Athlon and P4 are capable to deliver much higher x87 FPU performance than what is shown by Sandra.

Really? Go look at LightWave. Thats P4 optimized and no non-SSE2 CPU can stand up to that. A 1600 Northwood trounces a 1677Mhz XP (2100+ !). There are HARDLY any SSE2 optimized programs out there as of now. If most apps were optimized for the P4, AMD would be literally out of business. Then again we might be talking about different kinds of "optimizations" (and compiling a program through the Intel compiler isnt optimization IMO).

Negative either, save for a specific application footprint of another. ScienceMark is heavily optimized for SSE2, but the Athlons still best P4's in this benchmark. The old x87 FPU covers a wide range of floating point instructions that SSE2 doesn't. Also, Athlon's FPU is out-of-order 3-way superscalar and P4 SSE2 is vector-dependant 2-way superscalar (vector dependant means that SSE2 simply cannot get most of its performance under various situations) and the loading and storing of operands is faster (under Athlon, again). P4 is weaker than Athlon, even with SSE2... the lack of a very specific compiler for Athlon makes its true figure of performance absent. About Lightwave, I'm a little confused.

Athlon XP destroying P4 in a Lightwave scene

Remind that the Athlon XP's are fighting P4's in the worst possible case scenario. Applications are optimized big time for P4, but in terms of software optimization Athlon is lagging behind. Athlon, just like the P4, might benefit greatly from very specific software optimizations. Hardly it's "showing its age" an architecture which doesn't count on software optimizations yet.

Really? You ever see Single Xeon (same thing as a P4) Sandra with hyperthreading? Phenomenal increase. I guess by your rationale, a P4 with hyperthreading will waste an XP in terms of raw CPU performance. In my rationale says it wont because Sandra gives scores for the most optimal of conditions and is sometimes useles (hint MIPS = meaningless indication of processor speed).

Look at my first statement above and you'll see that Sandra is Hyperthreading-optimized but poorly optimized for the Athlon FPU, thus the bad Athlon whetstone results (within ~5% of a P3 per clock cycle).

 

[Ilmater]

Originally posted by: majewski9Okay you people are obviously got away from the point. Look PC1066 isnt supported by any chipset but reviewers use it. This helps the p4 beat up on the Athlon XP.

Y'know, I'm an outspoken Athlon and AMD supporter, but, with all due respect, you're making us look like idiots. Do you think that it's OK to run the Athlon with DDR400 RAM? Answer that to yourself before you read on. If you think that it's ok, then let me be the one to break the news to you: the KT400, no matter what its name is, does not support DDR400 RAM. Motherboard Manufactuers have decided to validate their boards for it, but VIA did not.

Everyone else is exactly right. The P4 was designed for high bandwidth memory, while the Athlon was designed for low latency RAM. It's that simple. The Athlon wouldn't be able to use RDRAM, not only because of its laggin FSB, but also because RDRAM has TERRIBLE CAS latency. The Athlon doesn't have the number of clocks to handle that. The P4, as I've said before, takes a good hit in performance if it miepredicts a string and has to start all over, effectively losing every clock that it took to get the data in there in the first place. However, it makes up for that loss with that ridiculously high clock speed.

Originally posted by: majewski92nd the suit brought up against Intel is worth mentioing since it so unprecedented and they have a good chance of winning.

A good chance of winning?? Now you're on crack. They're going to fail and there's no two ways about it. Judges don't have the intelligence to understand the techno-babble that it would take to prove a case like that. Neither do the lawyers that are trying to explain it. However, they are clearly correct. Their suit claims (correctly) that while the OEMs and Intel itself were claiming huge performance increases from the P4 compared to the PIII, on a clock by clock basis, this simply isn't the case. In fact, even Intel would admit this. They've admitted before that their processors were designed to ramp up clock speeds. Hell, back when the P4 2.2 came out, I saw a Compaq commercial claiming that they had the top of the line processor in their "new" machine, whhile in actuality it was packaged with a 1.4GHz Proc.

*WARNING* This next paragraph will be in all caps. The frustration has become too much and this is the only way I have left to express it.

WHAT THE HELL ARE YOU TALKING ABOUT??? OF COURSE THE PROCESSORS ARE BEING COMPARED ON ACTUAL APPLICATION PERFORMACE!! WHO THE F*** CARES HOW THE ATHLON DOES IN A SYNTHETIC BENCHMARK? IT DOESN'T MATTER! I DON'T CARE IF AN AXP 1600+ DESTROYS A P4 2800 IN SISOFT, BECAUSE MY ACTUAL COMPUTER PROGRAMS WON'T RUN ANY FASTER!!!

GODDA@#$!!! That's the stupidest thing I've ever heard! Why are you even arguing about it? Hey, if Warcraft III was entirely optimised for the P4 and the AXP could barely run it, I'd first be really pissed off at Blizzard, and then I'd go buy a P4. Why? BECAUSE THAT'S WHAT I DO WITH MY F****** COMPUTER!! JESUS CHRIST! If you only run CAD and database programs, then you go get a duallie AMP system and use it for what it was intended for. I, for one, will look at the UT2003 and the Quake III benchmarks because I PLAY LOTS OF GAMES! If something runs Jedi Knight II better and I played that more (haven't really gotten into that one), then that would use whatever it was.

I'm done. I've got heart problems and I'm getting worked up by a F****** forum thread.

I haven't had enough sleep and I'm on medication AND I just drank a feew beers, so please forgive me if I've offended anyone. So damned ignorant...

 

[Askheart]

Okay you people are obviously got away from the point. Look PC1066 isnt supported by any chipset but reviewers use it. This helps the p4 beat up on the Athlon XP.

While the hardware reviews put the KT333 against i850, was DDR333 ever supported by JEDEC? 😉