Sign in with
Sign up | Sign in
Your question

Comparing Northwood to AthlonXPs, not accurate?

Tags:
Last response: in CPUs
Share
November 9, 2001 9:37:49 PM

I have been thinking about this: If AMD's cycle of new AthlonXPs are 60mhz increments or so, while Northwood is released it means AXP is still not going to be at 2.2ghz. And this means that it is still not the helping hand here if we're to compare a new 2.2ghz processor to one with 1.8ghz. The only time it would count, is if AXP 1.8ghz is still beating the NW. If not, then the benchmark is useless, as we are comparing and going back to MHZ debates where the lower one loses!
How will we compare them to know which really is working at its MHZ assigned? Will we have to underclock NW just to see results? If not, then it is normal that it beats an AXP because it's more MHZ and pumped more per clock cycle. Hopefully we will straighten this out.

<P ID="edit"><FONT SIZE=-1><EM>Edited by Eden on 11/09/01 06:43 PM.</EM></FONT></P>
November 9, 2001 9:45:27 PM

The cache will not fix everything. Even with 512KB cache and a better FPU, the Northwood will be slower than an Athlon XP (per clock, in some cases) because it has a longer pipeline.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor <P ID="edit"><FONT SIZE=-1><EM>Edited by AMD_Man on 11/09/01 08:29 PM.</EM></FONT></P>
November 9, 2001 10:34:06 PM

I will compare dollar for dollar anything available at the time. Not clock for clock.

"Ignorance is bliss, but I tend to get screwed over."
Related resources
November 9, 2001 10:40:52 PM

Clock per clock dont not exciste.Cease that like AMD is king beacause it have a higher IPC.

Wisdom dont come with time
Meilleur chance la prochaine fois
November 9, 2001 10:44:16 PM

"the Northwood will be slower than an Athlon XP (per clock) because it has a longer pipeline." - false statement.

Will have to bench 1.6Ghz Northwood against PR1900 XP (PR1600 wouldnt be fair at all), your guessing/speculating is a waste of bandwidth.

Do you have a northwood w/ benchmarks or link to back that up? I didnt think so... P4 nw smokes the XP, I dont even consider the XP as a contender anymore with its poor stability, god forbid you put XP's into production/mission critical machines.
November 9, 2001 11:09:15 PM

"what I'm going to show you today is the version of our Northwood 0.13-micron technology, and this is the first version of the processor anywhere outside the fab. So why don't we put up to the big screens the output here, and we're running at 3.5 GHz. So we've not only met, but exceeded your expectations and have delivered 3.5 GHz." -August 2001

Basicly, AMD is getting left in the dust and not soon enough. AMD will not have the .13micron fab changes done till 4Q 2002 (yay)

I love our samples that OC far beyond anything in .18 micron. kinda sick getting 30% overclock from early samples. As we are already finding northwoods runing over 3Ghz overseas now (those who dont give a rats ass about NDA's). whine all you want about per clock but its not gonna be a fair fight anymore at comperable clock speeds. PR rating hurts AMD in that reguards, they should have stuck with pure Mhz rating and they would have held some gound in the low end (sub 2Ghz) arena in 2002.
November 9, 2001 11:22:12 PM

Quote:
As we are already finding northwoods runing over 3Ghz overseas now (those who dont give a rats ass about NDA's). whine all you want about per clock but its not gonna be a fair fight anymore at comperable clock speeds.

1) It currently isn't a fair fight at equivalent clockspeeds. It isn't even a fair fight at PR ratings vs Intel clockspeeds. The Athlon XP wins almost every time, and you've never been able to admit that. :lol: 

Unless the Northwood 2.2GHz performs like a Willamette 2.6GHz, the Athlon XP will still be in the lead, even with a 200 "PR point" handicap.

2) Show benchmark linkage? The closest thing we have to a Northwood benchmark shows an October engineering sample with its FPU and ALU performing like nothing more than a clock-speed-increased Willamette with twice the cache. It still doesn't cut it against the Athlon XP.

3) You still can't provide any proof of AMD instability. You gave us one incompatibility that may very well be crappy software. You have what you say are personal experiences, but so do we. A lot of us have personally experienced that AMDs are rock solid.

Quote:
AMD will not have the .13micron fab changes done till 4Q 2002

That must be why Intel says parts of the AthlonXP are already 0.13micron.

Oops. :lol: 

Kelledin
<A HREF="http://www.linuxfromscratch.org/" target="_new">LFS</A>: "You don't eat or sleep or mow the lawn; you just hack your distro all day long."
November 9, 2001 11:31:21 PM

The cache cannot fix everything Fugger. It is a FACT! If cache was key then why is the G4 faster than the Athlon and P4 per clock when it has less cache than the Athlon (correct me if I'm wrong, not including L3 cache).

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor
November 10, 2001 1:34:36 AM

AMD wont have its dresden fab 100% .13 micron till q402 yes, but the thoroughbred is coming early q202, right about the same time northwood is launching. More FFF from fugger.

~Matisaro~
"The Cash Left In My Pocket,The BEST Benchmark"
~Tbird1.3@1.5~
November 10, 2001 1:48:59 AM

That thry try get realistic Q3 is


Non plausible fait la traduction si tu veux

Wisdom dont come with time
Meilleur chance la prochaine fois
November 10, 2001 1:51:18 AM

Quote:
Do you have a northwood w/ benchmarks or link to back that up? I didnt think so... P4 nw smokes the XP

Do you have benchmarks and links to back up your statement?

We'll just have to wonder till it comes out.

Chesnuts roasting on an open CPU
Bill Gates nipping at your wallet
November 10, 2001 1:52:38 AM

I dont understand what you meant juin, but if you meant that q3 is more realistic for the release of the thorooughbred, AMD has said that the begining of q2 is when they will release. Dresden has .13 tools working in their fab as we speak, intel is even accusing them of already having .13 micron technology in their chips already.

~Matisaro~
"The Cash Left In My Pocket,The BEST Benchmark"
~Tbird1.3@1.5~
November 10, 2001 10:00:27 AM

Hay guys If he had the Bench Marks. Dont You think he do the same thing. He did on his own computer? Marked them so it would make them Look good. Remember he a Fourm Lier.
November 10, 2001 10:07:01 AM

::wouldnt trust benchmarks from fugger ever::


It would be amusing to see what he comes up with however.

~Matisaro~
"The Cash Left In My Pocket,The BEST Benchmark"
~Tbird1.3@1.5~
November 11, 2001 10:21:26 PM

So in conclusion what people? So far I haven't really got a concrete answer.
November 11, 2001 10:36:02 PM

question: does fugger work for intel now?
the phrases he uses suggests so.

tizz all moot point anyway.
very difficult to compare them chip to chip

they use different motherboards, archetectures and types of ram.

im not even sure how comparable the KT-266A and the P4X-266 is... is the P4X derived from KT266 or KT266A tech?

OEMs selling "High End"PCs with integrated video will be forced into Q3tournaments using a TNT2M64!
November 12, 2001 11:36:18 AM

<font color=blue>"::wouldnt trust benchmarks from fugger ever::
It would be amusing to see what he comes up with however."</font color=blue>

All Fugger will come up with are his

<b>Fabricated Fugger Facts (FFF for short)</b>

Then he will throw a tantrum, and start personal attacks.


<font color=blue>This is a Forum, not a playground. Treat it with Respect.</font color=blue>
Anonymous
a b à CPUs
November 12, 2001 1:06:33 PM

>Basicly, AMD is getting left in the dust

Funny.. it has been over a year that intel fan boys keep saying AMD will be left in the dust when P4/Northwood/SSE2 optimized apps /Itanium / blablabla come out. Still, when ever I check any reputable website, they all show AMD products to outperform anything intel has to offer.

Not that any of the fuggers/meltdowns/etc acknowledges the fact that AMD is the performance leader today.. as well as the value leader.

>they should have stuck with pure Mhz rating and they would
>have held some gound in the low end (sub 2Ghz) arena in
>2002.

This sub 2 Ghz "low end" is so low end, Intel currently has *nothing* that can outperform it. So, a current 2 Ghz P4 is lower than low end. Pretty expensive low end if you ask me...

= The views stated herein are my personal views, and not necessarily the views of my wife. =
November 12, 2001 1:25:59 PM

Complete bollocks.

Itanium is the worlds fastest individual processor in FPU operations, roundly troucing the Althon and I'm confident the Hammer will also leave with its tail betwixt its legs after a show down with itanium in 64 bit apps.

Next you will say that pitting 64 chip against a 32 bit one is an unfair test, however you did say and I quote "AMD products out perfrom anything Intel has offer". Clearly this is not the case.
So the moral of this cautionary tail is don;t make sweeping statements. BTW if you call me a troll, you will bethe worst hippocrit in the history of the world. So there.

Did you know there never was there ever a cat so clever as magical Mr. Mistoffeles? <P ID="edit"><FONT SIZE=-1><EM>Edited by charliec2uk on 11/12/01 03:29 PM.</EM></FONT></P>
Anonymous
a b à CPUs
November 12, 2001 1:55:58 PM

>Itanium is the worlds fastest individual processor in FPU operations

You'll be hard pressed to find any usefull applications that can demonstrate that.. Of course, I know its a very young platform, and there is very little OS /Apps support, but so far, Itanium performance has not really been impressive, other than in one or two very specific becnhmarks (like the SSL thing).. perhaps you can provide me some links to convince me of the opposite ? (dont gimme Spec.. thats more of a compiler benchmark than anything else. And if you insist on spec, make sure you include the IBM Power4).

>Next you will say that pitting 64 chip against a 32 bit
>one is an unfair test, however you did say and I
>quote "AMD products out perfrom anything Intel has offer".
>Clearly this is not the case.

Its perfectly fine by to compare a 64 bit cpu to a 32 bit performance wise. Rather, its the word performance that should be refined. By perfomance I mean how fast actual programs that are sold and used run on available platforms. For me that includes general office apps, databases, 3D rendering, games, video encoding, .. that sort of stuff. I dont run Spec.. do you ?

That being said, I think we all agree Itanium is not really a competitor to the AXP or vice versa. Considering its price, the effort to develop on it, the power requirements, etc.. God forbit it would *not* be faster than a cheapo $200 cpu.

btw, Compaq recently stopped all Itanium deliveries because of a bug in the cpu (cf the inquierer). Though I suppose that is to be expected with a completely new platform, it doesnt look good..

Regardless. I dont think Itanium is currently a cpu considered by anyone on this forum. I see it more as a "beta" chip for developpers.

>I'm confident the Hammer will also leave with its tail
>betwixt its legs after a show down with itanium in 64 bit
>apps.

We'll just have to wait and see. If have no idea.. I just think it will be hard to compare both platforms, as they wont run the same software. Given the very different nature of the VLIW (oops.. EPIC) core of the Itanium, Im sure there will be some specific applications that run (much) faster on Itanium than on anything else (probably including Power4 and others). But I really dont see it as a general purpose computing architecture that runs everything reasonably well. I think Hammer will do this. But thats all speculation, we'll see.

>BTW if you call me a troll..

Nah.. I'll keep that to the Fuggers and Meltdowns of this forum for now.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
a b à CPUs
November 12, 2001 2:37:35 PM

It's gang up on fugger day! Ok, my turn my turn!

What about branch mispredictions caused by the 20-stage deep pipeline? Don't they become even MORE of a hinderance as clock speed increases?

Back to you Tom...
November 12, 2001 3:44:39 PM

The point!

Not just branch mispredictions, but routine code switches as well. On a GUI OS like windows where everything is timed, can the processor predict a task switch? and that happens too often. okay may not be too often when you consider the speed, but within the code there are too many branches and loops that are mispredicted the first time and the last time. there are data contentions only larger cache could help.

All this reflects in the poor performance of the P4! Its for real and everybody can see it on the charts. 20 stage, boy its too deep!

girish

<font color=red>No system is fool-proof. Fools are Ingenious!</font color=red>
a b à CPUs
November 12, 2001 4:32:58 PM

But that's the responsibility of L1, does the Northwood have more L1?

Back to you Tom...
November 12, 2001 4:33:30 PM

Bbaeyens: There is tons of usefull 64bit apps now. you obviously are clueless, dont know where to look, and have no use for. most of you never seen 64bit let alone built one (wtg on brilliant observation).

Girish, you are much smarter than Intel engineers, Microsoft engineers, and like god. Why didnt you make a procesor yet with all your insight and knowledge? Ooops I guess I mispredicted that one huh?

Go speculate more on the performace of the hammer, and brag how well it will zero extend 32 bit code by the end of next year.

After northwood release, the PR rating will have less value. clock/clock northwood hurts the PR rating by 15~25%.

Granted northwood will be unaccesable by most of you cheap ass losers. but hey, what else can you bitch about?

Crashman, no, faster clocks are reuired for 20 stage pipeling.
November 12, 2001 4:43:31 PM

Why? Surely a higher pipeline frequency will just mean that bubbles get flushed out more quickly. At the end of the day, all these problems as more real estate is given over to cache etc. and the clock speed scales, making these little glitches less important.

Thats what I think anyway. Oh, I can't think of a good sig, inspiration welcome.

Charlie

Did you know there never was there ever a cat so clever as magical Mr. Mistoffeles?
November 12, 2001 4:50:09 PM

Touche my friend, but try to take the venom out of your posts, it makes us feel a tad uncomfortable.

But a good point any way.

Charlie

Did you know there never was there ever a cat so clever as magical Mr. Mistoffeles?
a b à CPUs
November 12, 2001 4:51:06 PM

But if the rate of errors stays the same, twice the frequency means twice the errors. But I have a feeling that the rate will increase as the frequency increases.

What's the frequency, Kenneth?
November 12, 2001 4:59:16 PM

No, if we assume the number of errors to stay constant, it should mean that each error takes up less absolute processor time as each error spends less time in the pipeline. It seems to me that no what operation is in what pipeline stage, as a preportion of processor time, each operation is less significant simply becuase there are more operations passing through the pipeline. But now I'm tripping over my own words. Bleeeeeeehhhhhh! Brain hurts. Lets just wait until *mumble mutter mumble* [-bad words and assorted curses-] Intel releases northwood. Why the bloody hell did they move the release date back, I mean it is just giving the initiative to AMD.

Charlie

Did you know there never was there ever a cat so clever as magical Mr. Mistoffeles?
a b à CPUs
November 12, 2001 5:17:04 PM

I didn't say number, I said RATE. We cannot assume that the NUMBER of errors will stay constant. At best we can hope the RATE stays constant. A rate is a proportional number, such as 1 error for every three executions, etc.
So if the RATE stays the same, the holdback will be the same. But I propose that the rate may increase with clock!

What's the frequency, Kenneth?
November 13, 2001 2:19:54 AM

According to the professionals at realworldtech.com, the itanium sucks.

~Matisaro~
"The Cash Left In My Pocket,The BEST Benchmark"
~Tbird1.3@1.5~
November 13, 2001 2:47:35 AM

it makes a really good room heater though...
(if you have a large room, use a 4 way cluster)

Why do i feel like the lone sane voice in the mental assylum?
November 13, 2001 2:52:47 AM

ummm... u do realise that as the price of a processor increases the number of sales drops off quickly dont u?

intel will want to sell the northwood u know. gonna be hard with high prices and a depressed economy

Why do i feel like the lone sane voice in the mental assylum?
Anonymous
a b à CPUs
November 13, 2001 5:15:12 AM

I seldom check these forums anymore but it's refreshing to see FUDGEPACKER still doing that same ol' thing; talkin' smack about AMD. FUGGER, for what it's worth, I ditched my dual PIII 866 server for a dual Athlon MP 1.2 setup with the Tyan Thunder. This is my main production server. It's been running for weeks and not a single problem. I've never seen SQL Server run so fast!!

AMD Rules!!
Anonymous
a b à CPUs
November 13, 2001 7:20:44 AM

Guys, why don't you ignore the FUGGER troll?

He is sick because hes 450$ P4 is beaten by a 230$ XP. And he can't live with it. Let him scream and cry. This is the only way he can relax.

Show some mercy to the guy.

<font color=blue>Get a T-Bird...
Impressive CPU + House heater in one package. What do you need more?</font color=blue>
November 13, 2001 1:53:34 PM

yep, Northwoods are supposed to have 512k of it!

but there is simply no cure for deep pipeline instruction restarts.

I had suggested one sometime earlier in Matisaro's thread, I have tested it with my own simulator, and it does work! Complicated to get it on silicon probably, its hard to believe Intel and AMD engineers dint thought of it.

girish

<font color=red>No system is fool-proof. Fools are Ingenious!</font color=red>
November 13, 2001 2:12:26 PM

There goes FUGGER again! dunno why he is called fugger!!

BTW pass on a link to half a ton of 64 bit useful apps pls.

Maybe Intel engineers are smart enough to make 20 stage hyperdeep pipeline, run the processor at dizzy speeds, and mind you the engine works twice as fast! the RAE (Rapid Execution Engine) works at twice the core clock, that is 4 GHz for a 2 GHz processor!!! and yet, the processor perfoms rot.

Fugger has probabely made by heart the Netburst presentation (guess how is it connected to the *net* anyway except it was downloaded from there ;-)) and refuse to see and accept the realworld performance of his beloved processor. I bet he still hasnt got a P4!

BTW the Hammer is supposed to be primarily a 32 bit processor, with 64 bit extensions, and why does it have to zero extend all the integers? In fact even the Itanium will have to zero extend them. And it isnt going to be a zero extending factory making 1000G zero extended 32 bit integers an hour. Neither will Itanium be.

And you say PR for Northwood will be down 15~25% - so you accept it still isnt 100%!

And yes, most of us on THG have some idea how a processor works and how it is usually made. I havent studied any semiconductor theory since I passed out of the engineering colege and that wasnt enough too. But I am pretty good at instrumentation and keep making gadgets and utilities and programs for testing various things.

Here I am working on a simulator, architecture description language for last four years. Hence I do claim to know <i>something</i> about microprocessor design. wait for end of next year and you will see the product.

I know you work at SGI, but man, you seem to have gained nothing working at such a good company. Maybe next time SGI lay people off, FUGGER will be one of them. Usually companies cut staff those are good for nothing.

Northwood, if it works out a good performance/dollar ratio then we will buy it. else we will have the Thoroughly bred ones by then! And there wont be any better performance/dollar ratios until intel makes a string of price cuts like 26%, 54%, 51% and 24% within a year of its release. man, P4 1.5 GHz cost $968 when it was released last october. See how much it costs today! Its anybody's guess why intel had to cut it by so much!

girish

<font color=red>No system is fool-proof. Fools are Ingenious!</font color=red>
November 13, 2001 2:27:39 PM

exactly, its not a count by number, or any absolute parameter.

if you need to go by fugures sorry, figures :smile: say in terms of cycles. Code switch in a processor happens all the time, like the way you breathe, the processor switches code paths. And as much as they claim they are good at branch prediction, they arent better than 98%, and its still a percentage. In absolute terms, <i>any</i> isolated branch is always mispredicted, and in a loop or a function calls it is missed just at the begining and the end of the loop iterations. The BTB or the Branch Target Buffer is supposed to hold the code that would be possibly branched to. That is a sort of L0 cache, and in event of a branch hit the code switches to the BTB and still loses a cycle. Then the BTB code needs to pass through all the stages that adds to the missed cycles, but technically, the a number of instructions are being <i>handled</i> but not executed. What do you say to that?

Well, this happens in all cases the Athlon and the P4 and even the P3, but the fact that P3 and Athlon give comparable results while the P4 lags far behind is that P3 and Athlon have 10 stage pipeline while the P4 has a 20 stage pipeline!

We dont need any predictor to predict branches, we need a smart code handler that could turn all the branches into inline code whether they branch or fall through. I will post a link to my earlier post on this sort of model, I have simulated it and had better results. The problem is to get it on the silicon and it is upto Intel and AMD guys.

girish

<font color=red>No system is fool-proof. Fools are Ingenious!</font color=red>
November 13, 2001 2:38:04 PM

Poobaa lets get this into a different thread. We will talk about the by-products of a Itanium server as we already had somewhere else.

<font color=red>No system is fool-proof. Fools are Ingenious!</font color=red>
November 13, 2001 2:41:20 PM

Yes, its easy to ignore the Fugger troll, but we have a saying that goes "Mhatari melyache dukhha nahi, pan kaal sokawto!"

Meaning we dont care about what he trolls about, but he might get too adamant about it if we dont. maybe this is not the best translation, but hope you get the idea. :smile:

girish

<font color=red>No system is fool-proof. Fools are Ingenious!</font color=red>
November 13, 2001 3:04:05 PM

Benchmark results on the Athlon XP's seems to indicate that AMD has taken the upcoming P4 Northwood into account when setting their PR ratings. A XP 1900+ is performing quite a lot better than a P4 1900 MHz in almost every common benchmark.

So don't expect P4 Northwood to work out magic. Only increasingly faster clock-speeds will eventually allow the P4 to out-perform Athlon XP.

By the way, I wonder why Intel has delayed the introduction of Northwood? I get the feeling that they are surprised how well AMD is doing, and now they see that the superiority in clock-speeds doesn't seem to be enough to beat AMD. Please note, that AMD is able to be faster than Intel using the current 0.18um process, which makes me conclude that the same relationship will be present at 0.13um. Why not ?

Intel seems to be wanting to slow down the current development pace, by not provoking/challenging AMD and hoping they will slow down too. But it doesn't seem to work. AMD is just turning up the heat, at the moment.



--------------------- P4 1700 MHz ---------------------

Overcloked 1700 MHz P4 Willamette Socket 478 with stock cooling benchmarked using SiSoft Sandra.

CPU Arithmetic.
Dhrystone ALU: 4039
Whetstone FPU/SSE2: 1097/2571 MFLOPS

Multi-Media:
Integer iSSE2: 8304 it/s
Floating-Point iSSE2: 10109 it/s

Memory Bandwidth:
RAM Int MMX: 1950 MB/s
RAM Float FPU: 1957 MB/s
November 13, 2001 9:44:26 PM

well the 0.13 micron northie will arrive earlier than the 0.13 thoroughbred, even with the delays. it is said that northie will have a better FPU as well as more cache and higher Mhz, but initial benchmarks dont look that fantastic actually.

so yes... i can see the same sort of thing that happened this year to happen next year too
the p4 streaking away with Mhz, but the athlon keeping pace with more IP per clock cycle.

Why do i feel like the lone sane voice in the mental assylum?
November 14, 2001 1:56:25 AM

"Benchmark results on the Athlon XP's seems to indicate that AMD has taken the upcoming P4 Northwood into account when setting their PR ratings. A XP 1900+ is performing quite a lot better than a P4 1900 MHz in almost every common benchmark."
While AMD is competing against Intel with Athlon XP, I suspect that AMD hasn't even really considered Northie at this point. The XP ratings seem to be pretty accurate performance indicators to show the difference between T-Bird and Pally. So AMD will likely deal with Northie when it needs to.

"So don't expect P4 Northwood to work out magic. Only increasingly faster clock-speeds will eventually allow the P4 to out-perform Athlon XP.
By the way, I wonder why Intel has delayed the introduction of Northwood? I get the feeling that they are surprised how well AMD is doing, and now they see that the superiority in clock-speeds doesn't seem to be enough to beat AMD. Please note, that AMD is able to be faster than Intel using the current 0.18um process, which makes me conclude that the same relationship will be present at 0.13um. Why not ?"

AMD isn't faster than P4 presently. Athlon outperforms P4, true...but it's not faster. Just a little splitting of hairs here, I know.

But I DO agree with you in general. Howerver, AMD has come clean that Palomino does indeed have some circuitry that is already at .13. Intel claims this might minimize the performance increases of the full die shrink as compared to Intel's P4 die shrink.

Mark-

When all else fails, throw your computer out the window!!!
!