Slashdot Mirror
Tom's Hardware benchmarks K6-3 and PIII
Christopher Thomas writes "Tom's Hardware Page has
finally posted comprehensive benchmark results for the Intel's
Pentium III and AMD's K6-3." The quick summary is
that Intel must be experiencing deep fear right now- the most
significant change from the P2 is the cost. Fortunately
the rumors of the 1ghz P3s are flying fast and furious. If you can't
beat 'em, up the clockspeed.
If you can't beat 'em, up the clockspeed.
And then you can say "we've upped our clockspeed, so up yours."
-- acb [sorry, couldn't resist...]
Why are people comparing the G3 and G4 and P3 to the Merced and Elbrus chips? There's no comparison to be made. When/if they arrive, they will be in an enterprise/supercomputing league, not the desktop league. This means, they will kill the desktop style processors.
He said, "You'll be able to tell your grandchildren that you helped assemble the first NT supercomputer," and I cringed.
go to www.amd.com Now go to www.intel.com (please don't use lynx!)
AMD just doesn't market well-it's as if the K6 series is an afterthought.
Yes! The next issue of c't (February 1) will have a big K6-III vs. P-III test. They usually do kernel compilation as one of their tests.
They also have a SPEC license, but they don't seem to use it :-(.
-- Erik Corry without his cookies
To some degree, the "eyeing up Windows 2000" is a moot point. By the time the "desktop" version of Windows 2000 is released (1 to 3 years from now?) the CPU market is going to look a little different.
Merced got preempted by years. If you want a high performance, relatively low cost RISC desk box or tower box, get a computer that uses the G3. Or wait around for the G4 to get released. The only thing that makes the Merced superior in any way to the G3 or maybe the G4 is that the Merced has 64 bit addressing, which PowerPC chips currently only have 32 bit addressing.
DES Khaddafi KGB genetic jihad Uzi Rule Psix Qaddafi cryptographic Peking Mossad Legion of Doom Albanian Serbian Saddam
Celeron 400 costs $150 and AMD K6-3 costs more than $300 on pricewatch. Now, according to tom's tests Celeron Smoked AMD on most tests... you choose.
I don't understand how one would conclude from the article that Intel is experiencing "deep fear". I think it's time for those middle school reading comprehension exercises again!
jds
This is a completely valid point. But for those who need a more cost-effective solution, a K6-3 might better suite their needs. Besides, if you need power, why are you using an x86 chip anyway?
Tom says (in the conclusion):
"AMD will definitely have a serious problem to place K6-3. Its Winstone performance would make it eligible to be promoted as a high-end processor, but this won't really work out as long as it performs worse than Celeron in most 3D-games."
and
"... get a Celeron if you care about 3D-games or other floating point intensive software."
Michael
Has anyone EVER succeeded in getting AMD to
send them a chip manual?
They say: "Oh yes, right away, can't imagine why
you never got it..." And nothing comes.
Good reason why so few people write code optimized
for their chips.
Their own damn fault.
Phil
What about the Motherboard ?
the CPU is not really an isolated system -
a PIII in a wonderful mobo is bound
to outperform K6-3 in a cruddy one..
What's the chances of them having used one
without any L3 cache for the K6-3,
by some strange argument that the external
L2 cache of the other processors is equal
to the internal L2 cache of the K6-3,
when in fact, it would be fairer to give
them equal amounts of _external_ cache,
regardless of whether the processor then
thinks it is L2 or L3 ?
Use your head a bit. www5.tomshardware.com ... hmm... what about the 4 before it, or just www.tomshardware.com?
For Ye Guys Who Post Articles, you might want to check for things like that, so one computer (assuming 1 computer/www#.host) isn't overloaded.
don't buy Intel chips
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
A picture is better than a thousand ?'s. I couldn't stand reading the text since it is infected with the ? disease. But basically it looks like the Celeron's are the thing to get for per/$ ratio.
I sent there webmaster a short note to please turn off Dumb^H^H^H^HSmartQuotes and refrerenced them to the Slashdot article that was posted recently.
Yes I have to say that it will be a while before something like the celery comes out again. I got one of those 300A's a couple of weeks ago and it runs at 450mhz with 2.3 volts on the CPU. Linux is great on this baby. Best of all the CPU was $80 dollars. Can anyone say cheap power to go???
Frank Hale
frankhale@worldnet.att.net
Of course they compare your computer to a VAX... but hey, they're freely available (in a public-domain sort of way... I don't see any license)
Glückwünsche, haben Sie Slashdot ermordet, indem Sie zum korporativen Druck beugten und Subskriptionen einlei
Like kernel compile times or something. I'm not particularly interested in "Winbench" scores. I also wish they had thrown in some K6-2 scores so I could see how much better the K6-3 is over what I have now.
Even if the K6-3 isn't as fast as a celeron for 3D, if AMD gets the price down quick enough, it still might be a good buy. I beleive the PIII is going to be too expensive for most.
Perhaps not but I don't see why... Even if it's as stable as a non-overclocked cpu it's bound to crash sometime and when that happens it doesn't look good to have run it above specs.
the Celeron doesn't support SSE, yet it still did better than the K6-3 (at same clock rates) on most tests...so it's not a result of benchmark bias toward SSE.
And from a realistic standpoint, you have to work with the software available. So it doesn't matter if you have these great 3DNow! instructions if your software doesn't make use of them.
why would I do that?... cool G3, G4....but then U put them in a Mac?!?!...
Why shouldn't he? Stability is as good as with the 300 Mhz setting. But obviously you don't need performance that much when coding but it's still nice and it comes for free so why not use it?
...because it still sucks at fp. A good FPU is what I want from a processor anyway. Don't care about how fast my word processor runs when my type rate sucks anyway... :-) I have a Celery 450 myself and that's the kind of performance I like. *Don't* flame me saying that the Alpha, MIPS or SGI will smoke my little Celery any day because I don't care. I want my box to run Q2 and other 3d-games natively. Alpha's don't do that, right? Maybe Q2 has been ported but that's pretty much it in any case.
AMD's goose is cooked. The K6-3 is costly to produce and slower than the Celeron that Intel charges very little for, forcing AMD to charge less than they would like. Intel is bludgeoning them to death... They're going to twist people's arms to optimize for the PIII instructions, something AMD was unable to do for 3dNow, and Intel retains its FPU and graphics advantadges. Super high end users with too mucchmoney will buy the PIII and the budget buyers will buy Celerons. AMD better do something to shake em up,.
Almost all the tests done were with apps that supported SSE, but not 3DNow!. Of course the K6-3 will look bad when the playing field isn't level.
First of all, the CPU bus speed doesn't change the PCI and AGP bus speeds, so peripheral performance won't change. If they do come out with a 200MHz bus, it will only change the bandwidth to memory and off-board cache.
Second, assuming they will use a 200MHz bus might not be realistic since it will require exotic memory that probably won't be readily available when the K7 debuts.
Also, did you think that maybe they took a sample of the available software out there? There just isn't as much written for 3dNow!
Juiced? Or Not?
looks like they /. ed already. it's only been posted for 5 minutes and I can't get there.
DAMN! It's 4:09 am in San Diego. I got to go to bed.
-dave
Neither chip does anything for me. Maybe Intel and AMD will get into another price war and drive chip costs down a bit more, that's about the only good I see either of these two chips doing.
Ignoring KNI, the PIII is equal to or just marginally better than the PII.
The K6-3 is improvemed over the K6-2 in most benchmarks. If it was cheaper, I might get one.
All in all, I'm pretty disappointed with both of these new chips. I think I'll wait for the next round before I upgrade.
Any conspiracy theories regarding the late 3DNOW support and the early KNI adoption, or is that all above board?
--
Fuck the system? Nah, you might catch something.
Anand has a nice little comparison of SSE and 3DNow! Biggest surprise to me (stupid me) was that you can execute 2 3DNow! instructions per cycle, but only one SSE--so the smaller registers don't make that much of a difference.
Why not have someone who has Linux benchamark them. Linux seems to not care very much about which x86 cpu you have because it doesn't have optimizations for most of their speical features anyway. This would tell us the best chip for Linux. I have used OS/2, Linux, Windows 9x, Windows NT, and many others and it seems the operating system does impact the performance that you can expect. Just becuase something is the best cpu for windows does not mean that it is the best Linux CPU.
Computer modeling for biotech drug manufacturing is HARD!
The main point is that the K6-3 keeps socket 7 nicely alive: for instance I sold my previous motherboard and still bought a K6-300 rather than a Celeron (in part because my Asustech SCSI was working on socket 7). If the K6 had stalled at 300 Mhz, socket 7 would be dead ; but it is not the case and as a bonus, the K6-3 solve the cache issue L2.
"The only thing that will stop AMD from stealing the show is
either high prices or buggy chips."
Or, the same things that always stop AMD from stealing the show. . .
late delivery and tight supply.
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
why would you 'buy' linux?! its free
Well it means it's easier for devs to make their kni games work with 3dnow! with little fuss
I got tired of the obvious PIII enhanced benchmarks so I didn't get to the end of the report.
But what I really wanted to see was a second set of bars for all the tests was a price/performance evaluation.
AMD's goose is cooked. The K6-3 is costly to produce and slower than the Celeron that Intel charges very little for, forcing AMD to charge less than they would like. Intel is bludgeoning them to death... They're going to twist people's arms to optimize for the PIII instructions, something AMD was unable to do for 3dNow, and Intel retains its FPU and graphics advantadges. Super high end users with too mucchmoney will buy the PIII and the budget buyers will buy Celerons. AMD better do something to shake em up,.
The benchmarks suggest that the PIII and K6-III perform equally on business applications, but that the PIII is better for games and other programs that use a lot of floating point operations.
I think this just goes to show, celerons kick ass!
I'm sure you could pick up a Celeron 400, o/c it to 450mhz(or 125mhzx4, 500mhz!!!), and you'd have something that is faster, and cheaper then a K6-3
It is not wise to trust Tom for a lot of things... For example, for the longest time (is it still there?) he had an S3 texture compression example. His images looked bad. However, S3 said that when they ran the same test, their images were fine.
The problem? Somebody from Matrox (I think, or some other rival company) gave Tom the images!
Dragon NaturallySpeaking? Microsoft Netshow Encoder? Adobe Photoshop?
/.er to flame him...
Please. Not only were these were all Intel-supplied benchmarks, used for the sole purpose of making the K6-3 look bad, they're nowhere near being real-world bench marks. Who uses Netshow? Isn't Real G2 the standard? Why not use THAT encoder? What about MP3? It's becoming a major standard in on-line music distribution. Let's see how fast Fraunhofer or Xing codecs convert a George Clinton tune from WAV to MP3. We can call it "The Atomic Dog test."
Maybe I'm biased because he's going to high school at my alma mater, but Anand seemed to be a little more thorough in his benchmarks and explanations thereof than Tom was. Either way, SOMEONE needs to come up with some more useful benchmarks for PC multimedia. Relying on ZDNet thingies and Intel propaganda just doesn't work like it used to.
/me inserts 2 cents and waits for some lunatic
Did you bother to read the text of the article at all? The point was not that 3D Now! was bad, the point was what kinds of benefits you could expect from programs with KNI optimization over a standard 686 compiled version. I'm sure a 3D Now! optimized version would be somewhere up near the top of all of those benchmarks. The reason such "non-real world" benchmark tools were used was because there are NO good standard benchmarks for either of the new instruction sets. I'll admit Tom ran a pretty one sided review with lots of programs that had KNI optimizations and few with 3D Now!, but I don't think he tried to misrepresent that at all. Wait a minute, where are the 3D Now! optimized programs to do benchmarking? I guess AMD forgot to tell people to make them. Maybe that's why I would want to buy an AMD chip, crappy floating point and no developer support. Or is it the moderately better Win98 performance? Could someone please tell me the obvious advantage other then price that I am missing here?
Look, WC is a company of computer geeks for computer geeks. In a company run by, say, bankers, if your computers fail and someone finds out that you were running them at 150% of their specified clock speed and at the same time invalidating the warranty, some banker is going to find out about it and fire your ass.
The tests do seem to be weighted somewhat against optimal 3Dnow performance - although the text below them explains why the 3DNow performance may be less than optimal, I bet there will be loads of fsckwits who don't know any better ( and don't tend to read things properly) going around saying "K6-3 sux!!!" now. Pity. I hate the whole x86 architecture anyway on general principles, but AMD seems the best of a bad bunch, overall.
Remember kids! Never trust benchmarks!
Hmm, the 300a (66x4.5) is typically overclocked to 450 MHz (100x4.5). the 100 MHz bus seems to be friendlier to some things than the 83 MHz one (if you have PC100 RAM), so why not do the same thing with the Celeron 400 (66x6), and up it to 600 MHz (100x6)?
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
The address bus in RS/6K POWER boxes is 64-bit. Unfortunately, those aren't the chips being put in Macs.
it's going to be awhile before intel goes up to 133 mhz let alone 200
If anyone thinks Intel is feeling Deep Fear they're kidding themselves. Try looking at Tom's review with a business hat on and it looks a whole lot different.
Tom recommends using a K6-3 for office apps, but it only gives a 1% performance advantage under Win98, but run the same tests under NT Workstation and the advantage changes hands. Most corporates are moving to NT, or eyeing up Windows 2000 for desktops anyway.
Compare the K6-3's 1% office app. advantage to a 30% performance advantage to Intel at high end NT apps. This is the market P3 is targeted at. For Intel, home users and 3D games are a market for Celerons, not P3s. Just try using a K6-3 or Celeron on an Oracle server and you'll see what I mean.
Overall I think AMD are holding their own for the home market, but need to get better driver support for their enhanced instructions. As for intel, I think they'd be very happy with Tom's performance figures. P3 does exactly what it was designed to do.
Uh, it only took them 2 weeks to send me the 4 manuals I asked for. And I work for a "competitor", Motorola. Are you outside the US?
We don't have to speculate about which motherboards Tom used. See http://www.tomshardware.com/releases/99q1/990223/c pu-news-03.html
The K6-III was tested on an Asus P5A, and the PIII was tested on an Asus P2B. Tom didn't mention the cache sizes, so I checked the Asus web site. The P5A has 512 KB, and the P2B has none.
Well, that's why I trust fps in games and other real life situations that Tom brings us. The K6 is just kept alive while AMD tries to clear things up with the K7. The fact is still that AMD lags Intel in the top of the line segment. Let's wait for the K7 now... I just hope that AMD can ship that baby before it's too late...
At a tech luncheon at uni recently an engineer for AMD (who had worked on the k7 - I got to hold a waffer!) said that KNI and 3DNow were so similar that it was possible to change KNI (oops, i mean SSE) assembly code into 3dnow assembly code with a perl script.
Unfortunately I don't know if AMD has told any of the software developers that this is the case, or even better provided them with a script to do it. It was hoped that adoption of SSE pushed by Intel's muscle would bring greater adoption of 3dnow. Still, if AMD doesn't get the word out, it won't happen.
The enemies of Democracy are
Is there any real reason (except for inferior coolness) against it?
I have production machines on P133 and PII/233. No problems yet.
*price. i DO think the k6-3 will be cheaper than the p3. right now the p2 450 costs about 3.5 times as much as the k6-2 400 (in norway). i hardly think its worth that much more. and comparing p3 to a chip that probably will be half the price isnt really fair.
*While the celerons now are good and cheap, remember intel are putting a stop to overclocking in the new chips, or so i've heard all over.
after reading the benchmarks i too wondered why intel would be afraid of anything. the benchmarks really is in intels favour. the one thing is though.. the things the p3 is good at is what most people doesn't use anyways. i used to work for the biggest computer company in norway and did alot of support for businesses and private customers. of the thousands of pcs i looked at/fixed/etc just about none of them would need fast 3d graphics or voice rec. most people and businesses just don't _need_ it. as tom said, you dont have to wait for office apps anymore anyways.
unixuae@online.no
Oops, I meant to say that it *is* faster, alot as well... :-)
That's actually very important: the pII/pIII are PentiumPro (weasel) cores. If we all think back to when PPro appeared, it was a bit scandalous -- it was slower at 16bit code than its predecessor, the Pentium.
:) and in fact manages to run _faster than 95_ on the pII/pIII (ppro core) systems.
Having used Intel's vtune, I can tell you that win95/98 has a _lot_ of 16 bit code still in there, and I can believe that win95 runs faster on the K6. Yet, for all its pathetic dogged slowness, NT is truly 32 bit (even if it's 32 bit crap
If NT is faster than 95, Linux the HolyOS will really fly (which it does).
However, my Celery300a is in the mail. You can still find 'em if you look carefully. Ohbaby.
The 21264 600mhz is already as fast as the pre-release specs on the Merced will be. When you look what is comming out within the next couple of months, the Alpha takes the lead. Compaq is contemplating the 128-bit version of the Alpha already. Of course, the Alpha is the chip used by the Cray so it is both the knig of the supercomputers as well as the king of the desktop/worstation market when it comes to raw CPU performance. They are very price competitive with Intel as well (comparing top-of-the-line and run-of-the mill versions, they are about the same price on the street).
Not so simple.
If you might want to appreciate the blessing of multiple CPUs systems (at affordable price) you have to buy Intel CPUs.
The Deep fear that Intel will be feeling is related to the cost/benefit analysis of the P3. For corporations who want the most bang for their buck (leaving the Celeron out of the debate for now), AMD makes a comparable chip at a far more affordable price. If the P3 costs 3 times as much and only runs 10% faster, which would you choose?
Oops, forgot to log in. Did you bother to read the text of the article at all? The point was not that 3D Now! was bad, the point was what kinds of benefits you could expect from programs with KNI optimization over a standard 686 compiled version. I'm sure a 3D Now! optimized version would be somewhere up near the top of all of those benchmarks. The reason such "non-real world" benchmark tools were used was because there are NO good standard benchmarks for either of the new instruction sets. I'll admit Tom ran a pretty one sided review with lots of programs that had KNI optimizations and few with 3D Now!, but I don't think he tried to misrepresent that at all. Wait a minute, where are the 3D Now! optimized programs to do benchmarking? I guess AMD forgot to tell people to make them. Maybe that's why I would want to buy an AMD chip, crappy floating point and no developer support. Or is it the moderately better Win98 performance? Could someone please tell me the obvious advantage other then price that I am missing here?
I agree.. looks like AMD is going to have problems until the K7 due to the cost of the
K3.
Each time AMD releases something new I get worked up, hoping that finally they've got something that can compete with Intel. Each time however they always are shown to be six months behind at least. A few friends who are close to the K7 development have said it's a killer processor, but considering what Intel will have coming out at the same time, I have serious doubts about it being AMD's savior.
It seems they can't learn from their own mistakes. The biggest knock against AMD for a long time has been terrible floating point performance, yet they've never done anything about it. I wish they would just shake things up and break away from their apparent mindset of, "Intel is doing this in 6 months, we have to do it in 12 then," and do something completely different. If AMD ever wants to truly be taken seriously and end the constant talk of how this is the end, I don't see what else they can do.
Should we really be getting excited about a 1 GHz P3? The article I read said it was probably being cooled by liquid nitrogen, which _probably_ means it won't be shipping anytime soon. In fact, I think the article said AMD was likely to have a 1 GHz chip before Intel, plus the hype (which may be unreliable, granted) is that the K7 will be faster than the P3 at similar clock speeds. Intel watch out!
You idiot, didnt you know walnut creek overclocks
their pentiums? My celeron is o/c and rock stable.
Go and buy a clue.
-kojak
Both Intel and AMD keep making progress...
Neither will roll out a chip that blows away their previous offerings until they've sold as many as they can.
These new chips are just a preview of what will be REASONABLY priced a year from now. There's no reason to buy any new design right away unless you're extremely rich or a moron.
Hmm, I guess it's time to look into that liquid cooling :P
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
At work, I have a P133 and its performance is perfectly adequate for 95% of the tasks I do. There are probably 10 PCs used for general office work for every one PC used as workstation for CAD/CAE, 3D, etc. Few people in an office environment need a PII, let alone a PIII.
But at home, I play 3D games, compile stuff, edit large images, encode mp3s, etc. Every bit of performance counts, and many of the things I use the computer for are very FP intensive.
This is why the K6-3 sucks and the Alpha rules. Who cares about improved integer performance when it is already good enough not to be a factor anymore. And who cares about 3DNow! if AMD can't get anybody besides 3dfx to really use it.
Since the release of the K6, AMD has known that FP performance is their achilles heel. But they have released two new chips since then without addressing it. I bought a K6 when it was brand new, but they have since blown their chance to get my repeat business. One can only hope they will pay attention to FP performance in the K7.
Dave K
daking at infinet dot com
That site must have been /.ed Ive tried for a while to get in now, why wouldnt Tom expect to get hit from us once in a while so be prepared?
*sigh* Guess I gotta wait a while for the effect to dissipate.
> I'm not sure whether to laugh or cry at this.
How about neither. How about posting a worthwhile response to the guy's comment (even though his tone was probably a little out of line). The first guy said using an overclocked CPU in a production machine is asking for trouble. The second guy responded that Walnut Creek uses overclocked CPUs in thier production machines (I have no idea if this is true or not. Can anybody verify this?). Are we to assume that the people at Walnut Creek are out of thier minds for doing this?
- TicK