Slashdot Mirror
AMD K6-III released
Several folks wrote in to announce that AMD has officially
announced the
K6-III.
Thats a link to the official product release if you're interesting
in reading it direct from the horse's mouth.
← Back to Stories (view on slashdot.org)
Yay AMD! :)
but does anyone know if they are any good 2mb boards
available? The FIC PA-2013 is unstable, so don't mention it.
Let's see 'em actually try to maintain high volume shipments.
And then let's watch them lose MORE money next quarter.
Sounds great, but I think that there is going to be just as much difficulty getting my hands on one of these as there was for the K6-2's. AMD makes some cool chips. I just wish that they would be available instead of sold out. When I bought my K6-2, there wasn't a single one anywhere. I had to settle for one of the slower (300Mhz) chips that nobody wanted.
all purpose server,with 128 MB of ram,it should serve X fine to a bunch of low cost PC's configured as X terminal and it should also do well as web,mail,nntp,imap,ldap,etc.... server,also,it could serve well as development platform though i would put in at least 256 MB of ram if i want to do all this on the same machine.
Canadian AC
They're not in full production of the 450's yet so to depress demand they have an unusually high price. It should drop as the production ramps up.
Seems fairly stable to me. Of course I havent tried overclocking it.
K6-* have really poor FPU's in general. Try using one to render some intense stuff, compared with a equivilant Pentium. Now I don't especially like Intel, but is the K6-3 really going to perform?
-AC
Just noticed that they are writing it as a K6-III instead of a K6-3 (like K6-2s). This couldn't possibly be a clever marketing attempt to identify these chips with the Pentium III could it? nah... ;)
This AC is eagerly awaiting the Quake benchmarks and overclocking stats on these chips.
Anyone know if the FIC 503+ will do the dance with this little demon?
Hey - my K6-200 chugs along nicely at 225, and
a 12.5% speed increase for free is always welcome.
Once I get PC100 RAM, I'll try taking it up to 248...
AMD's press release claims that the K6-III is shipping in volume now, but I haven't seen any place that actually claims to be selling chips. Pricewatch doesn't list any prices for it, etc., etc. I'd really like to get one (to replace a P166!), but am I going to have to wait until April?
It does matter when your CPU mfgr is turning $1 bil a year--it gives them more room to drop prices. Not to mention better mfging and more dev resources. AMD has already backed off the "We'll undercut Intel by x%" (was it 15 or 25? I don't remember) promise. They couldn't afford it. The only reason they've turned a profit is because they're charging higher prices now (and IBM bailed them out last year). Take a look at Celeron-400 vs K6-2 400 prices sometime. Then compare their performance at Tom's hardware guide. Then try and actually find a K6-2 400. Think AMD's mfging has improved since K6-2's? I don't.
According to FIC's press release, the VA-503+ and the PA-2013 will support the k6-III.
Cheers,
--Pete
But the bus on the slot-A boards will be 200mhz and be upgradable to the Alpha (which already makes the vaporware Merced look like a slug). I haven't bought into anything AMD since the 386 (when they kicked the bits out of anything Intel had to offer). Looks like they might be doing some serious kicking if they can ship the K7 this summer and live up to their specs on it.
I'll mention it if I like!
Think of it really as a K6-2+. The internal core (including FPU) has not changed. The differences are the cache, and I believe they moved from a 0.35 to 0.25 micron process (which will help the clock speed out a little).
First you would be able to easily speed up things like Povray simply by recompiling it, possibly tweaking it to use as much single precision arithmetic of course. Especially the divisions could benefit by a huge amount.
But apart from that the prefetch instruction could be of use, especially with the K6-2 wich suffers far more cache misses. This could speed things up even for non floating point code.
Not too many capable people seem interested in doing such though, but maybe KNI will get some interest. Once compiler support for SIMD and data prefetch is in it shouldnt be too difficult to port over to 3DNow!.
KNI and 3DNow! have the same potential peak throughput of 2 additions and 2 multiplies per cycle, KNI has the advantage of greater number of registers and more advanced cache management 3DNow! has a slight advantage in its greater granularity.
FPU is still the same with the rather lame throughput of only one instruction once every 2 cycles. Most high level language benchmark are also very dependent on cache&FSB-speed, so K6-3 should do better as K6-2.
My 1MB PA2013 screams compared to other K6-2 300 systems. 85FPS Quake2 at 640x480 with single Voodoo2. 33 FPS in Unreal. Never had any problems with it at all.
It doesn't RUN x86 binaries...
;-)
...It straps a jet engine to the back of the x86 binaries and clears them for takeoff!
Floating point means dick to me
It's not every day that you have an excuse to say that, but it's fun to say.
What is wrong with using the K6-III with more than 128MB of RAM? Is there a problem with the cache handling more than that?
In just a few short sentences, you have managed to spew out an incredible amount of information that, while commonly regarded as true, is in fact false. Very false indeed. First of all, intel is only putting the celeron in their socket (Which is named socket 370, where did you get 'A' from), not their PIIs, PIIIs, and certainly not the Xeon, which is the K7's closest thing to competition from intel. Yah, there's williamette, but that's not going to be until late 2000 by intel's own roadmap, and roadmaps have a tendency to slip. AMD is more than covered on the lower end by the K6-3, so don't expect to see intel cheaper any time soon. AMD has also mentioned plans to make a cheap K7 probably sometime in the first half of 2000 so that they can retire the K6 line completely, although details at this point are almost nonexistant.
Finally, there is the 3DNow! vs. KNI issue. The fact that KNI has more instructions in it does not make it better. They both have the exact same theoretical performance. KNI has more instructions because its seperate processor mode means it is forced to duplicate existing instructions rather than simply using the existing ones. The only result of this is that it is more difficult to support KNI. It does NOT mean KNI offers better performance. It doesn't.
the reason i said it was unstable is because the
2mb version uses much slower cache chips that
have both higher access times and higher latency,
which causes instability at anything above 100Mhz.
With slow (PC66) ram even at 100
It just burns my britches that the most SMP a layman can expect to get is a dual Pentium Pro 200. I'm as excited as the next guy that the K7 will be SMP, but I also read that the K6 *could* SMP, but there is no chipset for it. That's nutzoid. Assumably a "low-end" K7 debuted shortly after the dawn of the last year of the millenium will bring SMP into reach of a lot more users, but in the meantime, is there any chance of SMP'ing K6s? Has anyone heard anything about this?
--Jogar
The Socket 370 chips are around. People are claiming that the Slot 1 variety are hard to find.
Intel says the 300A will be made till year's end (probably meaning the S370 variety). Use an adapter card...
There are SMP Alphas, Sparcs, Mips, etc. Since, K7 will use EV6 bus it will use the same SMP protocol that Alpha uses.
AMD: welfare for the working class.
Hopefully AMD won't ride SS7 all the way to the poorhouse.
BN
Some marketing goon must have came up with this ploy.
K6-III indeed. I'd have respected it more if they'd have called it K6-3. O'course, I thought K6-3D was a better name than K6-2.
Roman numerals suck...they're pretentious. Romans werent much on scholarship, and their numbers and calendar show it. The Roman calendar is just about the worst caledar in existance, and their numbering system is just one step up from scratching out hash-marks. Doesnt even have a zero.
Selling below cost is not illegal (in the US - antitrust and FTC concerns excepted).
Dumping is a concept existing in international trade law. It is not selling below cost, it is selling your product in a foreign country for less than you sell it in your own country. The remedies for this range from import duties, retaliatory price sanctions, and trade restrictions.
But illegal? More like a violation of international agreements.
That's strange, I bought a k6-2 2 weeks ago for $165 (retail box). That seems pretty reasonable to me.
Buy a cheap K6-2, Celeron or K6-3, and spend the money you save, later on a better CPU, or now on additional/better hardware (19'' monitor, 3Dfx, ...). Unless you want to do real-time CPU intensive tasks most of the day. For instance, if you're just compiling it, doesn't matter whether your CPU goes 30% faster: compiling big packages will still be too long to wait for, and compilation of small programs will be fast enough.
Oh man, I remember my friend having an Nx586 w/o FPU. Ran Win95 really nice, but that was about the time the Quake 1 test came out. Nexgen supplied him with FPU emulation software. Not fun. As I recall Nexgen did replace his !FPU chip either for free or very little.
But if you can't afford a Jet Engine you may as well run Jet fuel through a 57 Chevy.
The K6 FPU did fall short of... well everything, the reason this carried over to the K6-2 (at least as I understand it) is because the K6-2 was almost exactly the same core, the only difference being the added 3D-Now instructions, the K6-III and the K7 however are new chipsets however and (ideally) this problem is now nonexistent.
>1) I don't game. So, paying for a good FPU is a >waste for me.
:P
And you probably dont do CPU intensive computing
tasks, with the possible exception of compiling,
so a K6-II/300 would be fine anyhow.
>2) AMD doesn't limit overclocking in any shape or >form.
They don't need to -- their chips cant take the
kind of overclocking that the Intel ones do.
>3) AMD chips are cheaper than intel and now >perform just as well, if not better mhz for mhz.
Over clocking excepted of course (try running a
K6-2/300 at 450Mhz without supercooling)
>I'm an AMD-Linux fan... K6-2 333 @ 375mhz with >nothing but a cheapo MB/Heatsink/Fan. (what's not >to like?)
I'm not -- I have a K6-200, and I'm sure it has
more problems than the P100 sitting next to it.
>I'm quite interested to see if anyone can >overclock the K6-3's past that 500mhz barrier >people keep talking about for AMD chips.
Doubt it
>$300 for 450mhz?? and it beats the P-III 500mhz? >Sounds great to me.
Only for integer -- remember that Intel can't afford to lose the high end customers that need
the FPU performace -- AMD's good fortune is due
to the hole in the market created by that.
Intel is quite profitable, so I doubt they are selling at a loss.
I hope to review the K6-3 and P-3 at my site RSN.
For now, I've just posted a K6-2 400 review.
The K6-3 looks like a winner, I believe it is the first non-Alpha based micro with a three level cache architecture.
--------- Webmaster, http://www.cpureview.com and
I rather doubt he meant x86 compatible, seeing as he mentioned alpha.. grin
Posted by Skip Franklin:
Anyone seen any 3rd-party K6-III reviews? I know of some that disliked the PIII - I'm interested to see what people have to say about the K6-III.
Just curious ... I always thought that selling at a loss to drive a competitor out of the market was a monopolistic practice and, in the context of antitrust law, illegal. Isn't it?
... I know that Intel is the target of an antitrust suit but do these issues facter into it?
I've always wondered how AMD or Intel (or any company that loses money quarter after quarter because they were selling at a loss in order to eliminate competition) could escape antitrust action
I doubt very much that anything but the most synthetic of benchmarks will get anything like the 12.5% gain implied by the clock speed. On the other hand, a 75% increase in CPU cost does not translate to a 75% increase in system cost, so the effects partially cancel each other out.
But it's really silly, anyhow; just get a Celeron 300A and clock it at 450, and you're probably within 10% of any of these puppies, unless you have an application that specifically takes advantage of the larger cache (i. e. blows the Celeron cache but fits in the K6-III cache). That saves another $200, so you can buy another Celeron or two in case the first one goes bad.
Of course, in a server context things are a bit different...but then again, most "servers" these days need I/O more than cycles.
The AMD-K6-III/450 processor is priced at $476, and the AMD-K6-III/400 processor is priced at $284, each in 1,000-unit quantities.
$476 / 450MHz = $1.06/MHz
$284 / 400MHz = $0.71/MHz
That seems a pretty big gap, for a ~10% speed increase.
As long as it's not the MS ODBC "Jet engine"!
OpenPIC, a SMP technology developed by Cyrix and AMD, supports up to 32 processors (theoretically), and all scheduling hardware is off-chip, so any x86 CPU can be used. If you were a masochist, it would bentirely possible to build a 32-processor SMP 8086 box!!! It looked farily easy to port to other CPU architectures, so other 'non-SMP' CPUs may have worked with OpenPIC (i.e. StrongARM).
It needed specific SMP support in the OS, and although several chipsets included full OpenPIC support, no board manufacturers included multiple sockets on their boards (can you say Intel and monopolistic pressures??). Since there wasn't any hardware support, it was kindof difficult for anyone to put the support into any OS.
*sigh* Sometimes I just hate the computer industry.
asinus sum et eo superbio
asinus sum et eo superbio
in omnibus veritas
Yep, AMD chips, Cyrix chips, and IDT chips (Winchips) are all x86 chips.
I was browsing the processor listings over at http://infopad.eecs.berkeley.edu/CIC/summary/local /. My k6-200 gets about a 6.5 int95 and a 3.5 fp95. Pretty poor, compared to a Pmmx-200. So here's what i'm wondering, and I mean raw FP, not just the 3DNow stuff:
How much better is the k6-2 in FP? I was under the impression the FPU was just a k6 clocked up appropriately. If the k6-3 is still the same FPU, with all due respect I'm not interested. Here's how I make the numbers
k6-200: int95 6.5 fp95 3.5
pII-400: int95 18 fp95 18
(some people say your milage may vary...perhaps by as much as 50% less)
pIII-400: same
k6-2? k6-3?
I'd appreciate if someone could fill this in
Isn't the best bang/buck ratio right now still an overclocked Celery^Hon?
-- Rick
Now if they had only made it (and the K6-2) able
;-)
to do SMP... Imagine if you will a quad CPU system
running K6-III 400Mhz chips... (400 because the
450s are still a little steep in price)...
Ah well, I can dream while I save up for a K-7
system when they come out
As far as I've been able to tell, they are going to officially release the chip itself next monday, to coincide with the big PentiumIII release the same day. That's probably why you (and I) aren't seeing any prices at online reseller pages. Today's announcement was just the "official" announcement of the chip, like Intel's announcement of the PIII a couple weeks ago.
And the intel PII 400's going for ~$300 (US) vs the PII 450's going for ~$480 (US) isn't almost a $200 price gap (source: 3rd lowest price on pricewatch.com). Also looking at most places on a singular basis (memory-man.com and paragoncomp.com among others) most vendors have the PII-450 about $210 over their price for the PII-400... CPU pricing is usualy exponential with clock rate, not linear.. the top few MHz of a particular CPU line often doubles the price of the CPU, while increasing speed by 10%
-Matt
There's a review on Anandtech, and there was another over on SharkyExtreme. They were both pretty positive.
Tom's Hardware Guide loves the K6-3!
Go to Tom's Hardware Guide
Anyone know what the status is with regards to Linux kernel support for this beastie? Is it even an issue?
What about support for 3DNow in XFree86 or one of the commercial Xservers?
Macka
$200 difference between a 400 and a 450? Surely a typo...
I'm also curious to see how a K6-3 FP benchmark optimized for 3D-Now stacks up against a P-III (3) FP benchmark optimized for KNI. KNI has higher potential calculation ability, but it's open to question how effectively it can be used. However, these benchmarks will probably have to wait until Intel and AMD perform their own tests and submit them to spec.org.
It's only illegal if they are already a monopoly.
Pricing below a competitor and below cost if you're not a monopoloy is not a problem.
Jeff
Does anyone out there have a pure statistcal (Read: no "AMD RULES" or "INTEL SUCKS") review of the K-6? I'm building a new box soon, and I want to know whether to go with a K6-2, K6-3, Celery, or wait for the K7. Being a student, I just can't do the PIII. Any factual input would be greatly appreciated.
-Dave
Juiced? Or Not?
1) I don't game. So, paying for a good FPU is a waste for me.
:P
2) AMD doesn't limit overclocking in any shape or form.
3) AMD chips are cheaper than intel and now perform just as well, if not better mhz for mhz.
I'm an AMD-Linux fan... K6-2 333 @ 375mhz with nothing but a cheapo MB/Heatsink/Fan. (what's not to like?)
I'm quite interested to see if anyone can overclock the K6-3's past that 500mhz barrier people keep talking about for AMD chips.
$300 for 450mhz?? and it beats the P-III 500mhz? Sounds great to me.
just ramblin
Friends don't let friends buy Compaq's. (Dell/Gateway... same same) You want a good computer? Build it yourself.
SMP willl be in the K7 to be released later this year,June or July I think. The K6-3 only has had the cache improved which help it greatly but the FP unit is the same as the K6-2. For a new FP unit you'll have to wait for the K7. I beleive the K7 will be AMD's first intel killer(if there is such a thing!!