Slashdot Mirror
New PIII: SMP In, Serial Number Out
florin writes: "This article from GamePC talks about the new cB0 stepping of the Intel Pentium III processor. The FC-PGA format has finally been validated for use in multiprocessor systems. After much confusion about this issue, it is good to see 'Now Dual Processor Capable' clearly marked on the retail box. Another item of interest is that Intel has gone ahead and stripped the controversial processor serial number feature from this new PIII, like they announced they would do on their upcoming Willamette CPU. "
AMD already have around 17% market share, and onece Fab30 in Dresden is fully ramped it will be running 5000+ WSPW, which at current die sizes will give them enough capacity to take 30%+ market share. AMD is actually gaining market share at the HIGH end, since their process technology is currently well ahead of Intels (all .18 micron, with copper interconnect in Dresden, moving to .13 micron by next year). Intel are still struggling to upgrade their fabs from .25 micron to .18, and are still using aluminum interconnect - it'll be a year before they move to copper.
Go to any computer store and all the fastest computers (900MHz+) are going to be AMD.
On top of the process advantages, the Athlon/Thunderbird is a modern core design, and has plenty of head room for clock speed upgrades. Intel's new Willamette core may conceivably start sample in a few months time, but won't be start to be available in any sort of volume to replace the PIII until Q1 2001. Intel expect it will take 3-8 quarters (huge range!) for Willamette to fully replace PIII. In the meantime, AMD Mustang will be out in Q4, and Sledgehammer in late 2001.
BTW, AMD's 760-MP (aka 770) dual way SMP chipset is due out in Q3 to coincide with DDR availability. Micron (whose chipset technology is licenced to Via) are coming out with high end SMP chipset support for Thunderbird.
They wanted to add more cache than they could affordably fit onto one die, and thus into one chip. They had before done socket chips with external cache, the pentium pros, but they were large and unwieldly.
Now they refined the manufacturing processes again to allow them to fit the cache on die and they're going back to the much cheaper, better, socket design.
The slot cost more because it was bigger, required a seperate PCB, and a bunch of components required to make seperate chips interoperate. The socket chip avoids all of this, not only making it cheaper, but allowing it to dissipate all of its heat through one small plate, instead of three seperate ones. Much more easily cooled, thus higher yields. And it's also easier to strap a big fan onto a socket chip than a slot one, and the important thing is that the fan, the big part, points away from the board.
And yeah, the K6-3 was nice, a lot of K6-2s are supposedly 3s which were remarked and had some features removed at the silicon level, because people were more willing to buy the 2s than the 3s. Weird.
The hardware RNG is a much better source of entropy than any other source on your machine, including your Sound Blaster. You can just get more randomness from it in less time. I wish they'd allow access to the raw ouput so we could do a full assesment.
And the serial number *was* a Bad Thing. People change NIC etc too often to reliably track them through it, and *lots* of people won't have NICs at all. Processors are far more reliable for this job. And there's sort of an excuse for those serial numbers, whereas there was never a plausible excuse for the CPUID.
--
Xenu loves you!
...had nothing to do with privacy, or consumer preference tracking, or software copy-protection.
It was to give the system support people an easy way of doing inventory, even remotely. Accounting wants to know what happened to that new expensive machine? Ask the network where processor #foo is, and find out.
(For those of you who think that's a completely stupid idea, I not-so-respectfully suggest you try working for an organization that requires equipment inventories. The ID scheme would have done away with those stupid little tags that get stuck on computers.)
It was only when the marketing groups of the world got hold of the idea and said, "Hey, what else can we track besides inventory... people! People and their purchases! Neat!" that the idea went downhill.
Elbereth speaks the truth; hardware addresses are already unique.
You cannot apply a technological solution to a sociological problem. (Edwards' Law)
How on earth am I going to buy CPUs mailorder and get the right microcode? none of the online stores are set up to handle buying a particular uCode rev - it's pretty hit or miss - Intel needs to have these fixed CPUs out with a different SKU number if people want to be able to order them reliably - in the mean time I'm waiting for SMP Athlons - if this Intel-drought continues for much longer I'll jump to AMD asap
Screw the PIII - some on-line stores are giving prices for the AMD Spitfire (also known by the moniker of Duron):
600MHz - $US89
650MHz - $US115
700MHz - $US159
Kinda makes the days when a PII/400 was over a thousand dollars seem pretty crazy, huh?
So expect to see some FUD soon about how only Intel can provide the true high end, SMP solution. Regardless of the fact that SMP is worthless for 99% of what people do with computers. (Don't take this the wrong way -- I have dual celeron -- just to say that most people run '95 or '98, and of those few who run NT or something better, most of the time they are surfing the web or editting text.)
What is ironic about this, is that Intel has no competitive SMP solution itself. Sure, you can use i840 -- and pay out the nose for RDRAM. And the performance is *still* not up to that provided by the BX chipset.
But BX, even though it is still the best performing chipset on the market (SMP or otherwise), is not really a viable modern solution for SMP, either. BX runs hot at dual 100; I have air piped directly over it from the intake fan. If/when I try for more, I will probably need to stick a fan on it. And of course, going above 100Mhz is not supported AGPwise on any BX board, nor even PCIwise on most.
But none of that will stop Intel. "Only genuine Intel is fit for the real high end!"
imho, don't bother.
PPros were cool chips, but for their time. Even if you could setup a quad PPro 233 (how high did they go) system and get it setup, it wouldn't be as fast as a dual celeron box if all you wanted was raw cycles, and the celerons have FPU tweaks which would make them better at rendering.
And if you need the huge cache and want to do server type things with them, then buy a Xeon now, they're fairly cheap, only $200 above a regular P3 when you count the special board and all, and with its much higher clockspeed it'll kick whatever you could do with the older CPUs.
Intel has come to thier senses.
For the past 6 months, I have rejoiced, silently, every time good news about AMD came out.
"AMD grosses $1 Billon"
and also celebrated when Intel had a screw up
"Intel recalls 840 chipset"
This is not because I hate Intel. I like Intel. The Pentium family of chips has revolutionized home computing. I have a dual celeron machine (BP6) that I am typing on right now. I have a P II 450 machine at work that rulez.
But, Competition is good
If AMD wasn't putting the screws on Intel, then they wouldn't give a damn what the consumer wants. They'd be as bad as Microsoft. But, with AMD giving Intel real competition, things couldn't be better in PC hobbyist land. Chips are cheap as all hell, and there is real choice in the market.
And Intel is forced to pay attention to what we want. We want Dual PIII boxes for cheap. Intel has listened and responded.
The free market rules, when there is competition.
-geekd
...there was good news, too. People who valued their privacy, and wanted to be viewed as more than "consumers" tied to a number, found solace in Intel's other announcement about the P!!!. Intel said when announcing the serial #'s that they'd also be implementing in a later stepping a hardware Random Number Generator. Privacy advocates praised that as much as they [expletive deleted] the dreaded Internet-readable serial #. Those new P!!!'s, originally due out late last year, were supposed to include a sensor to read "thermal noise," the completely random motion of atomic particles, and make those random values available to encryption applications. The importance of this is that truly random numbers cannot be calculated in software, they can only be approximated in software, through the use of "pseudo-random number generators." Those PRNGs are currently one of the weakest points (theoretically, at least) in implementing crypto, because a flaw in the "randomness" of their numbers could render encrypted data vulnerable. But natural phenomena like the motion of atoms is truly random, and the proposed Intel sensor would have therefore contributed substantially to the security of data. And yet, nothing has been heard of it since early last year, when it was announced along with the CPU IDs. The most likely scenario is that the utter backlash against the CPU serial #s caused Intel to completely drop its strategy of integrating "e-commerce" and data security features, which is a shame because hardware random number generation would have been almost as big a step forward for privacy as the CPU ID was a step back. But I still like to think that the NSA had something to do with it, and mutter under my breath "Damn gubbmint, takin' ma guns and ma crypto..." ;-)
"The more corrupt the state, the more numerous the laws."--Tacitus, *The Annals*
All right, that's enough outta me... off to bed before somebody pokes an eye out.
Free music from Jack Merlot.
Actually, what really killed off the idea of the CPU serial number was the fact that _Microsoft_ was not going to support that feature in Windows 98, NT 4.0 and 2000. You can hate Microsoft all you want, but when 85% of the OS market is not interested in the CPU serial number, that other 15% isn't going to be interested, either.
Raymond in Mountain View, CA
To quote the article at gamepc : "VIA's dual processor solution is still quite a while away"
Don't tell that to Tyan. They've got the dual coppermine mobo with a VIA chipset ready to go.
This is supposed to ship within days, and there are a dozen vendors which are set to have it. I was quoted a date of May 20, but that may have slipped. Egghead says they have 1000 on order.
140 bucks for an SMP mobo?! The only question is how do you get a Tyan to overclock. The answer usually is, "you don't". But they are usually extremely stable, so it's amazing they are using VIA!
kabloie
Wasn't one of the advertised Slot-1 advantages the ability to put more in for multiprocessing because now the chip was vertical? Now they're saying back to the traditional chip socket form is better for more processors. What gives?