Slashdot Mirror
Intel "Northwood" vs. Athlon XP 2000+
Augustus writes: "LinuxHardware.org has just published their results in the Pentium 4 verses Athlon XP war. In this review, the new Pentium 4 'Northwood' 2.2GHz is pitted against the Athlon XP 2000+. To level the playing field, both platforms use DDR memory which make for some interesting results."
We need better testing. You can not (really)
expect the same test to give you a good judgement, and then on the other hand, having different test for each CPU would obviously not give you a good judgement... Hmm, perhaps some speed testing on regular apps. I know they do that on quake and the like, but then that was written for a specific architecture also. Just a thought.
P4 was designed with Rambus ram in mind.
.13 Athlons, hopefully it will work out ok.
They should really use Intel's i850 motherboard to pit against the Athlon.
The p4 platform is simply not designed for DDR in mind, adding DDR in the i84x boards are afterthoughts and IMO I would much rather use Intel boards with Intel processors.
Athlon is doing quite well right now, seems like there might be a delay for the
kawai
Wouldn't it be more of a level playing field if both processors were allowed to use their optimum RAM types? Sure, keep the rest of the system honest, but don't handicap one processor by forcing it to use a RAM type it wasn't initially designed for.
The reason DDR was used is because there have been COUNTLESS tests done with RAMBUS.
The whole goal was to see how well it'd do with DDR now that it supports it.
In Soviet Russia, the television watches YOU!
Take a look at
d ex .html
http://www6.tomshardware.com/cpu/02q1/020107/in
They posted the results of their showdown 2weeks ago.
Quidquid latine dictum sit altum viditur
If RDRAM can get its prices down to closer to DDR, it might actually compete properly. Until then, AMDs lower prices and the lower price of DDR ram is going to wipe Intel's ass on value for money.
I would check the website so I know what I'm talking about when I comment. But since I can't access the site due to the (I assume) brutal slashdotting, I feel almost compelled to comment without any supporting information to base my wildly inaccurate opinions on. If at least the article summary had summarized (probably incorrectly) the article content beyond saying it was "interesting" then at least we could get the debate rolling, at least until the page became accessible again. At which time, everyone else would join the fray complaining that nobody reads the articles.
-Restil
Play with my webcams and lights here
The last /.ing of LinuxHardware was due to provider bandwidth rather than server constraint. I'm guessing this is the same issue.
"It's tough to be bilingual when you get hit in the head."
Well, maybe they ditched all the load balancing and extra hardware and placed only these machines as the webserver. It's a new benchmark: how long does a system last a nice /.'ing?
related?
Thanks to file sharing, I purchase more CDs
Thanks to the RIAA, I buy them used...
... they're not really levelling the playing field because DDR memory is a mature option for AMD whereas it's brand new on the Intel boards, and apparently has some problems.
If you're going to compare just CPU power then use synthetic benchmarks that test just that, otherwise if it's system performance you're going after why not compare AMD DDR to Pentium 4 RDRAM, at least those are two mature configurations.
Perhaps /. can warn sites that they're gonna link to them?
/. effect on 09-11-01 with no problem. Yet a fairly major web site can't handle it. Hmm...
The funny part is that the "Augustus" that submitted the article is from linuxhardware.org . Which leads me to believe that they knew what was coming...
It's amazing that my little K62-350 stood up to the
Ender
Nothing to see here
I don't see why hardware sites insist on seeing which chip "is fastest." I'd be more interested in an acceptable price/performance ratio. The Athlon XP 2000+ (which can still hold its own fairly well against a P4 2200) costs LESS THAN HALF of a P4 2200. Why anyone would spend the extra $350 on a P4 for the minimal performance gains (relative to the cost) is beyond me. And for those who want absolute, unforgiving, raw performance.. For the same price as a P4 2200 with a decent motherboard, you can buy a Tyan Tiger MP with a pair of Athlon XP 2000s and a bunch of DDR memory (AMD reccomends you use Athlon MPs but there's no reason the XPs won't work.) Sure, graphs and kernel compile times are pretty and all, but eventually you have to think about what is practical..
Given that the P4 costs more than twice as much as the Athlon ($548 versus $263 on PriceWatch), why would they bother with only DDR? Just by including the P4 they've pretty much thrown price/performance ratios out the window anyway.
A better question to ask of the P4 might be whether it could beat the Athlon with any kind of memory, and if so, by how much?
uh, the AlthonMP is _one_ speed grade (67Mhz) behind the top of the line XP
that's hardly 'ultra slow'
Just killing time while my program compiles and the site becomes available again.
Actually, you are right. However, Tyan now manufactures a motherboard that is "approved" (not necessariyl by AMD) to work with Athlon XP Processors (not just XPs masquerading as MPs).
The board is the Tyan Tiger MPX (S2466)
~.Evanrude
Read the damn faq once and shut the hell up. I see this quesion so many damn times. This is EXACTLY what the fact says: Slashdot should cache pages to prevent the Slashdot Effect!
Sure, it's a great idea, but it has a lot of implications. For example, commercial sites rely on their banner ads to generate revenue. If I cache one of their pages, this will mess with their statistics, and mess with their banner ads. In other words, this will piss them off.
Of course, most of the time, the commercial sites that actually have income from banner ads easily withstand the Slashdot Effect. So perhaps we could draw the line at sites that don't have ads. They are, after all, much more likely to buckle under the pressure of all those unexpected hits. But what happens if I cache the site, and they update themselves? Once again, I'm transmitting data that I shouldn't be, only this time my cache is out of date!
I could try asking permission, but do you want to wait 6 hours for a cool breaking story while we wait for permission to link someone?
So the quick answer is: "Sure, caching would be neat." It would make things a lot easier when servers go down, but it's a complicated issue that would need to be thought through in great detail before being implemented.
With processing power up at the level it is today, how much of a difference in these two chips' performances is really notable? Even if the Athlon "XP" outperformed the intel (as I am told is often), it probably wouldn't be much, at least not enough worth talking about, and it sure wouldn't negate the fact that there have been several AMD "bugs" which notably inflict Linux users. They are, however, much cheaper. So I guess for me, the comparison isn't "price/performance", it's "price/functionality". Just my .02
http://www.anandtech.com/cpu/showdoc.html?i=1574
It's nice to see a review that is NOT by tom's hardware on slashdot... nice to see a little variation in the works
...the Pentium 4 verses Athlon XP war...
'Pentium 4 verses'? Are they anything like Spam Poetry?
It's 'versus', Mr. Editor Sir.
the P4 wasn't designed for any particular memory
Not quite true. When you design a processor you do consider your memory type. For example, RAMBUS gives better potential bandwidth and slightly better latency. This will impact the expected memory read time. As modern computer architecture can be summarized as "finding useful things to do while waiting for memory accesses to finish" the memory speed certainly does impact the design of the processor.
That said, the slight differences between the two memory types probably is in the noise margins for this issue. But I promise you that someone on the processor design team was thinking about memory types when designing the P4.
Mark
As Nietsche famously said, "If you stare too long into the Abyss, 1d4 Tanar'ri of random type will attack you."
Wrong. The P4 was designed for high memory bandwidth. In fact, that may even be why it performs better on RDRAM chipsets than DDR chipsets. Who wudda thunk it?
Goddamn slashdot moderators. My orignal post gets modded down as a troll for pointing out a valid hardware issue, and this piece of cluelessness gets modded up.
Really? It doesn't say so on the page you linked to, not that I could see anyway. It did list the XP as being supported for the board, but also said "Single only!" right next to it. Too bad, I almost got my hopes up there. ;^)
main(O){10<putchar(4^--O?77-(15&5128 >>4*O):10)&&main(2+O);}
That's why I build AMD machines, not Intel. But they were doing a hardware comparison.
As this highily informative post points out, AMD has already fixed the problem in the latest revision of the Athlon XP.
Except that RDRAM hasn't been accepted well in the marketplace, and systems based on such RAM are more expensive. Most people buying a P4 are getting either PC-133 or DDR.
I'll grant you that it's silly, given that (according to Pricewatch) RDRAM is commanding only a 15% premium these days. I guess Rambus Inc. really did annoy just about everyone in the industry...RDRAM seems to be anathema.
It's pretty amazing that the Athlon so regularly beats higher-clocked P4s using RDRAM, given DDR's inferior bandwidth. Intel is ripe for the plucking! (Anyone happen to catch Jerry Sander's statement at the last AMD earnings conference call? "Intel is over", quoth Jerry. Now that's what I call confidence!)
299,792,458 m/s...not just a good idea, its the law!
Galileo: "The Earth revolves around the Sun!"
Score: -1 100% Flamebait
How do the next few months look in terms of the ability of either Intel or AMD to improve upon these products?
While I'm a fan of AMD's price/performance ratios, it looks as if they will be hard pressed to keep increasing the clock on the Athlon, while the Pentium 4 seems to have a lot more potential for higher clock rates.
Then, too, I'm wondering about the news reports that suggest that Athlons won't be paired up with the new DDR 333 MHz memory.
It may mean that the highest performance x86 architecture this summer will be from Intel and will be able to command more of a premium in price than if AMD were breathing down their necks, which has been the case over the past year and a half.
"Provided by the management for your protection."
Did you do it? How well does it work? I was thinking of doing the same thing, but thought better of it since it would be a major fn() hassle if it didn't work out.
Well, to a first order you are right, except that those decisions involving hardware (eg DDR being cheaper than RDRAM, so the AMD system gets more ram, or a faster HD) will always be open to dispute, and then the results invalid. The question of which platform is better becomes obfuscated.
Another point is that while most everyone has a budget, those budgets are not entirely fixed. As in, you might want to only spend $1200, but if by spending $1300 you get something noticeable for your money, you might do it. Similarly, if you can get pretty much what you want for $1100, you might save the $100 for the beer fund.
I think the best thing to do is to compare a wide range of systems, and include the system price as one of the "benchmarks". I do -not- think that major changes to the hardware configuration should be made to account for more available cash. Things like the disks, the video cards, etc should remain constant or otherwise the relevant comparisons start to become meaningless. However, this would still give you a way to see how the motherboard, chipset, ram, and processor (the things that substantially vary between an Intel vs AMD setup) affects the price/performance, and let you make a reasonable decision as to what you want to do with your computer/beer money.
But at the same time, I still want to see who is the fastest, and that means throwing more or less equivalent systems together, but not throttling either based on price.
The enemies of Democracy are
A comparison of the two top products from AMD and Intel reveals the astonishing: although the processors are as different from one another as apples and oranges, the difference is much less obvious in the benchmark results, when taken from an absolute standpoint.
In any case, one thing is visible: in the majority of performance tests, the new Pentium 4/2200 is ahead. After all, the top AMD processor has to make do with 1666 MHz, while its archenemy steps in with 2200 MHz. A closer look at the comprehensive benchmarks reveals that in Office performance as well as Linux Kernel compiling, the Athlon XP still takes the lead, despite its 32% clock speed disadvantage!
Hello, the fastest MP processor is 4% slower (1600 compared to 1667MHz), probably due to the extra stability they want there as the server marked will drop anything unstable faster than lightning. Stick with the facts (they're expensive and don't add any real value add-on beyond certification) and don't FUD.
Kjella
Live today, because you never know what tomorrow brings
While reading through the results an idea came to me. Is it possible that the reason the P4 generally does better on 'one algorithm type of tasks is because its long pipeline wouldn't get busted as much, meaning that the branch prediction worked, which is based of past branch statistics (right?).
This makes since to me actually, in speaking with my friend about this last night I was asked 'well, what do you need a fast CPU for, when does it matter?', I replied 'Well, games, anything to do with multimedia, like Photoshop effects, ray tracing, mpeg encoding, but ya for general use, the CPU doesn't as much'.
but wait, lets look at that list in how it relates to the pipeline idea:
games: probably a good deal is going on here, AI, 3d pipelines, IO, networking probably not something a branch predictor would excel at
Photoshop effects ray tracing, mpeg encoding: all relatively contained algorithm that (if I'm right) would work well with the brand prediction.
So actually maybe having such long pipelines isn't that bad of a thing, because the majority of your day to day would doesn't care that much anyway, and most of the time when you need something as fast as possible its a small repetitive algorithm that could be predicted.
no?
this is my sig.
Are those $1500 Gb cards quite a bit different from the $35 Gb cards (copper) or the $175 cards (fiber) listed on Pricewatch?
Bleh!
Gee, you get one less node in your P4 cluster and still spend $900 dollars more... I'd say that shows savings right there.
Ten nodes with $300 Athlon 2000+ and $1500 Gb NIC = $18,000.
Ten nodes with $600 P4 2.2 GHz and $1500 Gb NIC = $21,000.
It has relatively high latency, but it has much higher bandwidth which is what the PIV wants.
Considering that the original poster was refering to myrinet and not
gigabit ethernet, I'd hazard a "yes" there.
Peace,
(jfb)
To spur "enterprise Linux," Big Bang, the distributed two-phase commit.
Not to trusting of Tom's Hardware? Have another set of benchmarks.
Stop spreading FUD about DDR chipsets and do a bit of research first. Any set of benchmarks I've seen has shown Intel's i850 w/ RDRAM and SiS 645 chipset in a dead heat - and most of the time SiS comes out on top.
------ 24.5% slashdot pure
It's a stupid attempt to remove a variable in order to isolate the CPU performance. The problem is that the benchmark community seems to think that CPU comparisons are relevant to normal buyers. They're not. The only thing relevant to 99% of buyers is overall system performace at a given price. Yet I've never seen one attempt to compare systems that way.
I've not read the article because it's still /.'ed, but...
I was wondering about the stability of P4 vs. Athlon platforms. The awsome overclock-ability of the P4's has gotten me very interested in them, so I've been reading up on the recent reviews. Reading them carefully, I've caught a few interesting lines about the Athlon systems locking up some when running benchmarks, but the P4 system running like a rock. Makes me wonder if a P4 wouldn't be a better choice when you want performance and stability, with the Athlon being a better choice when you want performance and low cost. While this doesn't sound like a big deal, it is if you're like me and like to leave your computer running 24/7.
Just wondering if anyone had any comments on this as it's not something that I've seen anybody mention anyplace and it seems very relevent.
overclockers, don't forget that the Athlon MP is "unlocked"...
That was classic intercourse!
Well, RDRAM is crap in single-bank configurations. High-latency, and low bandwidth (relative to DDR). True, RDRAM runs at effective rates of 800 MHz, but the data path is reduced to 16 bits over the SDRAM 64 bit typical bus. Granted, RDRAM can allegedly be more easily configured in configurations where modules can be run in parallell, providing effective 32-bit bus, but I haven't seen this implemented in any consumer grade PC motherboards, and even then it wouldn't be clear cut that RDRAM would be a winner (especially with it's high latency). The only reason why RDRAM could even win is because RDRAM is much more efficient in a way, it can come much closer to its theoretical bandwidth than SDRAM can come to its theoretical limit.
In essence, Rambus memory is a complete bust, and Intel is *finally* giving up at least in part.
XML is like violence. If it doesn't solve the problem, use more.
A 2-channel 266MHz DDR memory controller has the bandwidth of 2x266x64/8 = 4.2GB/s. P4 front-side bus has the bandwidth of 4x100x64/8 = 3.2GB/s.
Even a single-channel of the new 333MHz DDR has the bandwidth of 2.7GB/s, almost enough to keep up with P4. Once the 400MHz DDR shows up, a single channel will have the same bandwidth as the P4 bus.
So, in a well designed DDR system, P4 will not have any bandwidth problems -- if anything, RDRAM systems will be slower because of RDRAM's lower latency.
Never underestimate the bandwidth of a 747 filled with CD-ROMs.
I don't know, I still would not even consider buying a Rambus based system - but I guess this whole righteous outrage thing lasted longer with me than most others.
sic transit gloria mundi
Hmm... it was a tie at 1.5 vs 2.0 GHz with both processors winning in the areas where they always win, what do you think it's gonna be at 1.66 vs 2.2?
sic transit gloria mundi
It's not just that DDR is a mature tech for AMD but not so much for Intel. It's that the two memory types work very differently, and how the processor expects data to arrive is going to be affected by the memory. P4's are designed to expect the higher bandwidth and latency of Rambus -- they expect to wait longer, and to get more in one chunk.
So no, I think that this would skew results in AMD's favor. I actually think that a fair comparison would require Rambus on the P4 system, and DDR on the AMD system.
Was hoping to upgrade this summer...
Oh and so we are on topic: it sure is interesting about all them Athlon vs. P4 comparisons!
sic transit gloria mundi
That article always left a funny taste in my mouth.
Why was he comparing next-gen DDR (DDR333), which isn't officially out yet, to the OLD PC800 RDRAM? Wouldn't it make more sense to compare PC1066 RDRAM (see the AcesHardware benchmarks)?
PC1066 RDRAM and DDR333 will both come out officially around the same time in official chipset support.
In other words, next-gen DDR performance for the P4 is about 1.5 years behind the RDRAM performance. Tom didn't mention that part...
In other news, Samsung is sampling PC1200 RDRAM now, too. 4.8GB/s in a dual channel config.
Unfortunately most of the magazines I've seen (though I don't really read them) tend to only give star ratings or whatever rather than actual benchmarks. What I'd like to see is a comprehensive benchmark comparison between different systems with different CPU, RAM, HD, graphics card, etc, at the same price point. That would be useful to me.
Another person who replied to your was right - stop spreading FUD and KNOW before you speak (yeah, I've been guilty of it too - but learn!).
SIG: HUP
The problem with this analogy is that it looks at the wrong thing. The performance of a CPU isn't a matter of how quickly it can do one thing (ie, how fast it can finish a race), it's a matter of how much it can get done in a given period.
/big/ assumption, but for the purposes of this argument it's reasonable)), that means the P4 is retiring about 1.33 instructions per clock cycle. (Note that the 1.6ipc figure for the K7 is from an old RealWorldTech article by Paul DeMone that I can't find right now - it's somewhere in the Silicon Insider archives . . . )
A much better analogy is water flowing through a pipe. You want to get some volume of water out the far end, so you have two choices: you can pump the water through a thin pipe really quickly (the P4), or you can pump the water at a slower speed, but through a much fatter pipe (the K7).
The K7 core can retire, on average, about 1.6 instructions per clock cycle. At 1.667GHz, that means that your XP1900+ can complete about 2.667 billion instructions per second.
Now, assuming the equivalent performance on benchmarks and the like indicates that the P4 2000 can complete about the same number of instructions (since it gets the same results at that clock speed (this is a
/That/ is the fundamental difference between the two chips: the K7 completes more instructions every time it's clock ticks. That's what people talk about when they talk about "brainiac" versus "speed demon" processors: the P4 gets it's performance because it completes lots and lots of clock cycles in a given period; the K7 gets it's performance because it does a lot in every clock cycle, even though it completes fewer cycles in the same period.
CPUs aren't a horse race - they're a production line, where what matters isn't how fast an individual thing is done, but how many things get done in a given time period.
himi
My very own DeCSS mirror.
And mobo, and RAM, and blah blah blah. Take into consideration something more relevent: Athlon's have better floating point operations - something that would be more important in most cluster-utilizing apps. Getting that for less cost makes it worth it. So yes, you're right to a point - but Gigabit nics can come for much less than $1500 and you purchase more than just that and a processor...
SIG: HUP
Hey, thanks for the compliment, that's all we're hoping for with this thing - people who don't want to bother setting up insane mod_mp3 modules or shoutcast and who want a web interface.
Thanks for the tip on versiontracker too, we just got listed on Download.com and our downloads are through the roof, so maybe a versiontracker listing will really put us over the top.
Standard animation programs are quite expensive on a per processor basis. Once you get past blender, things get expensive FAST.
Actually, any specialized field has expensive options that real professionals need. Except programming. And that's because of the FSF. GNU/FSF tools allow programmers to get away from the hyper-expensive tools. But animators, circuit board designers, secretaries, architects, etc. can't write their own applications. KDE, OpenOffice, etc. are working on the problem for secretaries (there's a lot of them, and middle management generally uses the same toolset, plus a few). Blender exists for animators, but it isn't (doesn't appear to be?) up to professional use. Gimp is challenging Photoshop, but only Kontour is challenging Illustrator, and it needs a LOT of development. etc.
So per processor licenses effect a lot of folk. (Well, few word processors need multiple processors, but you know what I mean.)
.
I think we've pushed this "anyone can grow up to be president" thing too far.
Sure, under extremely difficult circumstances (many great companies lost money over the last quarter), predatory pricing from Intel, and taking into account some large infrastructure investments.
I'm sure they'll handily beat estimates this quarter, just as they did last quarter. And yes, AMD will turn a profit this quarter (barring unforeseen disasters etc.).
I'm not sure what you mean by Intel being ripe for the plucking - AMD is selling 100% of fab capacity right now.
Haven't you heard about the 0.13 micron fab coming online this quarter?!? Also, AMD can farm out as much production as it wants to other fabs. This will first happen in the Far East as a way to get more Durons into the marketplace. We'll see later on...IBM has some very advanced fabs these days, and there has been cooperation with AMD on things like copper technology.
For the near term, they can only raise prices, not gain marketshare. Which considering their poor profitability, is probably something they should consider.
Not only are they considering it, they've already gotten a good start on it! Average selling prices were up 50% over the previous quarter, from $60 to $90.
Regardless, the real reason to buy AMD stock is to bet that Hammer will sample this year and ship in volume next year. If that happens the whole competitive landscape will be dramatically altered, and AMD stock will go through the ceiling. Shorting Intel will be the way to go then. =)
299,792,458 m/s...not just a good idea, its the law!
Galileo: "The Earth revolves around the Sun!"
Score: -1 100% Flamebait