Slashdot Mirror
K6-3 on Monday
This Monday, AMD will unveil the K6-3 at
400 and 450 MHz. The chip will have an on-chip 256k L2 cache, and
will go in existing Super 7 motherboard. The chip will probably top
out at 500 MHz shortly, but the chip should be produced into Y2K.
The chip reputedly performs like Pentium *3* of equivalent
clock, or even outperforms it with enough cache. People
like me have a decision to make - I have a K6-2 333 with 1MB of
L2 cache, and need to decide whether or not I should save up
for a K7 (hopefully SMP...) or just go for it and buy a K6-3. Thank
William Rhodes for the heads up
Good thing MS wants cross-platform enough to kill the NT PowerPC port due to lack of funds from Motorola. And I'm sure a similar relationship was established with Digital for the NT Alpha port.
how about:
... run Slashdot.org.
:)
Is the K7 a whole new architecture, or just another x86? I'm wondering if we'll have to wait for the kernel mongers to have to develop for it, or if Linux (or any x86 OS) is usable right away.
Heh, another case of "I didn't read the article, everyone recap it for me" ...
- A.P.
--
"One World, One Web, One Program" - Microsoft Promotional Ad
"Remember when the U.S. had a drug problem, and then we declared a War On Drugs, and now you can't buy drugs anymore?"
I am currently running a PPro/200 machine, primarily for developing music software.
I'm definitely at the point where I want more horsepower - so these articles on Slashdot are very interesting.
Does this new AMD chip do floating point anywhere near decent? Is it going to ever catch up with the Intel line in terms of FP power, or am I just better off sticking with Intel for my FP-hungry apps?
I don't know too much about the non-Intel alternatives... I got this PPro/200 a few years ago because it was the best thing for my music stuff... if there's an alternative, I'd sure like to know.
; -- the corruption of government starts with its secrets. a truly free people keep no secrets. --
Appreciate it. I'll wait for the K7 before I decide on non-Intel chips for my uses.
In the meantime, might be interesting to overclock my PPro, if it's possible...
; -- the corruption of government starts with its secrets. a truly free people keep no secrets. --
Thanks, I'll check into it. It may be fun, though this system already overheats if I don't turn on the drive-bay fans I've installed ...
; -- the corruption of government starts with its secrets. a truly free people keep no secrets. --
FUD is used improperly a lot of the time, but the root of this thread was the very definition of FUD. Uh-oh, what if you don't buy the known quantity Intel? You might have crashes. AMD and Cyrix processors are unstable. No one ever got fired for buying Intel. Etc. Sure, I too have noticed the proliferation of the term "FUD," but in this case it first perfectly.
(of course, "first" doesn't fit in that last sentence as well as the word "fits" does. dumb tyops.)
Posted by Mr. Assembly:
It has always been backward compatiblty that has kept intel in business (and Microsoft for that matter). So you can keep all that big investment in software that you have. If it wasn't for that we all would have macs on our desks and intel would be a company that made toasters.
Posted by Mr. Assembly:
You should use silicon grease. The reason for this is that it removes all the tiny air bubbles that prevent heat transfer from the chip to the heatsink.
What happens when I plug it into my board with 1MB L2? Does that become L3, additional L2, or is it just dropped?
tugrul
It's always easy to come up with examples that prove one thing or another. When we're talking about matrix algebra and FFT's, it becomes easier yet.
The computational kernels for all of these algorithms are very easy to optimize for any given cache size and number of CPU's (and, for that matter, number of floating point units). If the code you used was optimized for 256K or 512K of cache and whatever cache algorithm the PII uses it doesn't surprise me that it runs badly on a Celeron A. 128K should be enough cache to get decent performance from a matrix multiply.
The FFT's an interesting case, since the radix 2 kernel only uses about 5 flops, so a lot of people use radix 4 and 8 kernels to reduce memory (or cache) traffic. Of course, the x86 FPU architecture gets in the way here; there simply aren't enough registers and the stack architecture's all wrong. Radix 8 and vectorized radix 2 and 4 kernels are an interesting exercise in register and pipeline management. Then there are the twiddles (nth roots of 1). So OK, I'll grant that the FFT is more cache intensive, but there are clever ways to tune it for different cache sizes.
In any event, I would not pick any x86 architecture for large FFT's. For more typical 1024 point FFT's, though, the Celeron has enough cache (128K is enough for about 8K complex points; divide that by 4 to allow for the twiddle factors and other cache busting stuff, and you're still at 2K). But again, the FPU architecture is all wrong.
Intel represents quality??
If quality always won we wouldn't have
Microsoft and we wouldn't have this bloody
x86 architecture that really stinks!
I always thought that the Motorola 680x0
was better, and I'm sure there were a lot
of other good alternatives out there.
Why would you want to upgrade from a 333 Mhz to a 500 Mhz computer today? The only improvement we're seeing today is entirely in the video acceleration and that's entirely rendered in hardware no matter what CPU speed you have. More and more audio is rendered strictly in hardware, too. The memory requirements of today's software far outstrip caches so you're still going to be at the 100Mhz limit for software. I've compiled video rendering software with maximum optimization on several UNIX boxes and the difference in software-only rendering between 150Mhz and 500Mhz is noticable but barely worth it. The flight gear page has some good benchmarks on the situation.
Actually, wouldn't the Xeon chips be the fastest
:-)
shipping x86 chips, considering that the Xeons
not only have integrated L2 cache, but they have
more of it than the K6-3?
Now, the fastest *reasonably priced* x86 CPU
is another matter entirely.
Games games games.
Games.
Run a pencil-and-paper RPG campaign with your far-off friends: Gametable!
One thing we learned in my architecture class was that cache size is not directly proportional to speed increase, in either direction. For a given amount of memory, and assuming some things about cache-implementation, there's a certain amount of cache that provides optimal performance. Less or more than that will hurt performance. Too big CAN slow down your computer (albeit, probably not MUCH). It's too bad that nobody publishes tests for these things, so that we can know how much cache to get. Since money is an issue for me, and celerons have 128k of cache, period, I don't have much of a choice, but it'd be nice to know.
Run a pencil-and-paper RPG campaign with your far-off friends: Gametable!
Reminds me of a friend who had a 486DX2-66, and his crystal failed, somehow, and so he got a new oscillator, but they didn't have the 33MHz crystals, he had to get a 32.5MHz crystal, so he was underclocking his CPU by 1MHz. 486DX2-65
Yay.
Run a pencil-and-paper RPG campaign with your far-off friends: Gametable!
Funny, 166 w/o hardware excel runs at ~25-30 fps for me..
Daniel
Hurry up and jump on the individualist bandwagon!
I was pleased to find a nice glop of heat transfer compound on my K6-2 350 when I got my motherboard and CPU. Is there a type of compound or name brand that is better or has properties that are more desirable? If so I would definitely be interested on hearing opinions about this, seeing that this subject isn't talked about to much.
Thanks
No one knows about the K8, but AMD has said they are already working on it (as of a year or two ago).
--
Aaron Gaudio
"The fool finds ignorance all around him.
"Every man is a mob, a chain gang of idiots." - Jonathan Nolan, Memento Mori
Yeah...as long as you don't need to do any sort of division you'll be fine.
--
Aaron Gaudio
"The fool finds ignorance all around him.
"Every man is a mob, a chain gang of idiots." - Jonathan Nolan, Memento Mori
The moths might be getting let out of the old sporran one of these days!
I'm out of my tree just now but please feel free to leave a banana.
K6-2 were no OpenPIC. Heck, the K6 were not OpenPIC. AMD abandonned OpenPIC with the K5, since no one ever made a MB (that I know of) to support it.
K6-3s are not openpic either. They will not SMP at all.
(Well, not without a whole lot of glue logic added)
Claude Angers
Back in the 166+ days, they got pretty hot and people who were new then (me) and didn't think alot about cooling had big problems.
I knew a lot less about motherboards back then, too, and could probably have picked a more stable one. BCM sucks!
my 2 cents...
-k
You know, I've been running AMD chips for the last
two years and have never had a single problem.
And I still get people asking me if I trust the
AMD chips instead of Intel. I am sick beyond belief of the "Intel is the real x86" garbage,
especially when Intel releases a chip like the
PIII that is all hype and no performance and then
pushes it with with idiotic commercials saying it
will "Make the Internet more fun." WTF!?!?? Get a
grip, Intel is not the only CPU manufacturer that has a clue.
1) K6-3 are not SMP they are OpenPIC I belive(K6-2 were)....but no openPIC boards exist.
2) K7's are not Super7, they are SlotA and have an entierly different chipset.
---------------------------------
My current chip is an AMD K6-2-350, my *THIRD* K6 Chip. Every single one of them has been as solid as a rock. I think that you had better check your facts before spreading your FUD around.
NT
Hmmm, Dual Pentium 166 sits running in the corner quitely chugging away, never a problem. Exactly when is it going to start locking up on me?
/* TODO: Spawn child process, interest child in technology, have child write a new sig */
I just got done mod'ing them, and it works like a charm! Very cool... BeOS 4.0 even knows they're Celerons ;)
I'm gonna get a K7 with my next system. Hello, 200Mhz Bus!
AMD is really giving Intel a run for its money. Best of luck to them.
Plus, those K6 chips LOOK cool..
----------
"They misunderestimated me." --George W Bush, Nov. 6, 2000
Well, I can counter that with my own experience.
Before upgrading my machine with a K6-2/300 in an EPOX/MVP3C mainboard, I tried some Intel chips in another series of mainboards. I had nothing but trouble with the Intel junk.
This K6-2/300 has never frozen, never failed to boot, never failed to detect devices, and so on. The AMD K6-2 is the best CPU I have ever had.
Although I'm sure the K6-3 is cool, I'm still holding out for a K7.
The K7 is not just another x86. They have completely redesigned the whole structure from hardware to software.
I read on CMP the P IIs have just a single pipeline structure and the K7s will have a dual pipeline stucutre combined with L1 and L2 cache and a 200 mhz bus to make it just fly past P IIIs.
Based on what I've read about the K6-3 with its onboard 256 KB L2 CPU-speed cache, the result could be nothing short of _superb_ performance.
Unlike the Celeron A's with their 128 KB L2 cache, the K6-3's 256 KB L2 cache has already shown superior performance for the same clock speed as opposed to the Pentium II/III series. For example, if you're running Quake II with the 3DNow! drivers, a K6-3 will result in performance that is actually _better_ than a Pentium II at the same 450 MHz, if www.anandtech.com's website claims are true. It'll be interesting to see if we will see patches for Quake II that will take advantage of the Pentium III; if that's the case, it'll be a VERY close contest indeed.
Raymond in Mountain View, CA
..333 is not enough? Hydrodynamics simulations?
<^>_<(ô ô)>_<^>
oh come on. It wasn't just because it had the word intel on the processor that caused your box to be up for 300 days. AMD has been around for a long time and they know what they are doing. I have owned AMD's since their excellent 486. I have never had a problem that was caused by the processor itself. They have always been very reliable and have always had the better price/performance ratio. Yes, i did buy an AMD because it was cheaper. It's also a very good processor. Unlike some people, i guess i don't feel like spending 140 dollars more for basically the same thing. (and don't bitch to me about FPU, it's FPU is fine for everything but games, and all games i play had 3dnow implemented anyway)
Damn, some people. You'd think people who are so obsessed about "choices" would embrace AMD instead of the "monopoly" of intel. Hypocrisy, pure hypocrisy.
Don't get me wrong, I run an AMD and an Intel system, and I'll be the first to go get a K7 when it comes out. But quit thinking that the current top processor company is going to sit idle while another company passes them by. Not gonna happen.
Once the king of the mainframe world, and replete with a bloated infrastructure and inbred complacency, no one ever thought this giant would have to wake up one day and actually face reality. Reality is that if there's someone smaller, better, and offers value, you'd best take them seriously. If anything, this simply reaffirmed that company size and customer base are no guarantee of continued success.
I've got a 300Mhz PII right now and it's pretty good except for when I play Unreal. My rule of box-buying is to wait at least two Moor cycles.
I'd say wait.
For me, it's disappointing to buy a new machine and then not be blown away.
Interesting how Intel doesn't seem to have anything in the works to answer the K7 challenge (other than advertising dollars). Check out the chip roadmap to see what I mean.
I guess if you're a rich man, go for the K6-3 and maybe get a few percentage real world difference over your K6-2. But I'm waiting for K7 - will be a big jump over my P166 :-) Which will be relagated to firewall/web server/file server duties to replace my 486DX2/66 which will probably become an X-terminal (like I need another one?)...
-t.
I'd think that we can push socket7 to at least 600
;) Woo
Jim
If I'm not back again this time tomorrow...
I'd love to be able to run dual K6-3s. They'll be so cheap... it would be the best deal. I know some people are ready to get dual K7s if K6-3s won't do SMP (which I'm sure they won't). However, I'm fairly sure that at least until next year around this time, K7 processors and motherboards are just going to be way to expensive for most people. Sigh.
He said, "You'll be able to tell your grandchildren that you helped assemble the first NT supercomputer," and I cringed.
One of the disadvantages of chipsets that supported socket 7 at 100 MHz FSB was that they had a caching area size that was small, such as 128meg for the Ali chipset. That puts a ceiling on useful RAM expansion. While most won't need more than that, some do. With the K6-3 doing the cache on it's own now, my question will be what the caching area of this new cache control logic will be. The Intel Celeron in Socket 370, and the Pentium-II L2 cache, have an area confirmed to be 4gig.
Hopefully the K6-3 cache will run at full core speed even with a 100 MHz FSB.
now we need to go OSS in diesel cars
What happens when I plug it into my board with 1MB L2? Does that become L3, additional L2, or is it just dropped?
--
Your existing L2 cache will become L3 cache.
-Wee
Ash and Hickory, straight-grained and true, make excellent bludgeons, dandy for the cudgeling of vegetarians.
I would wait for the first price cut on the K6-3 and buy then. That should tide you over for a while until the K7 has come down a bit in price.
BTW. You can now get Dual Alpha 21264 with 4 MB cache and 1 500 MHz processor for under 10K. Another processor is $3500 (about the same as a 450 Xeon with 2MB cache, but the Integer peformance is 75% higher and FP is 100% higher)
Yep, if your motherboard supports the K6-2 400(CTX core), then it will automatically support the K6-3, and the L2 cache on the motherboard will become L3 cache. This is according to AMD. I know that the FIC PA-2013 and VA-503+ will support this.
hmm, you know you have had too much caffine when you come up with something like this:
The names K...2K.
sorry.
aaaaaaaaahhh!!!! man... you have a way of looking into the future... That just hurts me to think of all that power.... heheheh... gotta get it when it comes out...
well well.. Im glad to see that all that Intel Marketing has you convinced that they are the best... "This way in??" How about "no way out."
How dare you compare a AMD to a Cyrix... Thats blasphemy... I hope those " pathetic little chips " keep bothering you.. because Intel is losing its monopoly while your polishing your Intel Inside...
Noone seems to care about the fact that a lot of Super-7-boards do not support core-voltages of less than 2.1 volts, which could result in the neeed for a new motherboard to run the faster types of AMD K6-3s which shall be produced in a .18-micron-process and therefore possibly need a core-voltage of 1.9 or 1.8 (or even less) volts.
Do you have any information on that matter? I did not find any.
regards
tabit
I think my super-7 ASUS goes up to 660 MHz at max. overclocking settings... wonder if it's ever gonna be used :-)
:-)
Otherwise, the K6-3 sounds just great for super-7 owners (and no, anandtech's comments that you don't need 100MHz fsb MB are kind of crap, because what older MBs will *properly* support 400+ MHz CPU speed?)
Btw, the current L2 cache you now have becomes L3, just like on alphas
Stop the madness, people! The classic Intel P166 is helluva fast enough. :-)
Portland Group sells one. However, C++ (and C) usually result in much slower code than FORTRAN 77 or Fortran 90. Besides, Fortran 90 is much nicer to program - for numerical programs.
For this kind of task, I would strongly advice against buying AMD chips - their FPU is very slow compared to Intel's. You would be better off with a PII or Celeron. K7 might improve thinngs, but it isn't released yet.
I run all my linux servers on K5's and K6 and K6 2's. Never have any sort of random reboots or lock up problems. Currently, the most important server has been up for over a month, last rebooted to add firewall support to the kernel, was up 20 odd days before that, and was rebooted for that to allow IP aliasing support in the kernel. Get your dork face off your PC and out in teh real world for a little while.
Does the name Pavlov ring a bell?
It's about time AMD rolled out the K6-3. I'm confident it'll kick the Pentium III's ass. I'll probably end up building a few K6-3s for people.