Slashdot Mirror
Intel Roadmap
Karl "Kielbasa" Wise sent us an excellent article that showed up on Sharkyextreme detailing Intel's CPU stuffs planned out for future. RDRAM, Socket 370, and other tidbits.
← Back to Stories (view on slashdot.org)
← Back to Stories (view on slashdot.org)
That's what the vector computers do, such as the older Crays or Fujitsu ones. They don't have this wuzzy cache stuff, they just have a few GB of equally fast memory, and then a several hundred instruction deep pipeline.
It's great for matrix multiplications, but it sucks for problems that won't fit into an extremely deep pipeline (most discrete algorithms, eg. 90% of the stuff 90% of people use)
If you have high latency on your memory without a deep pipeline to make up for it (in order to prefetch loooong ahead of time) you'll suffer. And if you run the software most people do who don't do weather forecasts or fusion weaponry, you'll even suffer if you had that pipeline (which you don't). If RDRAM has high latency, it won't make the desktop. Not even marketing will make that happen.
I'm sure that in a few years, when everyone has 300MHz SDRAM, Intel will regret the decision to stay with Rambus.
--
Unfortunately, much like your first sentence, they also have a history of trying to get things out too quickly, and end up with buggy results (eg. logic errors, arithmatic errors, etc.)... (c:
Your Brain + EEG + LEGO Robots = Brainstorms
just how much of this
will ever actually
make it to market?
intel tends to be
a bit too optimistic
about release dates.
So is this roadmap
more or less realistic, or
V-A-P-O-R . . . ?
(this sounds like flamebait
but it's really more or less
an honest question)
Irritable, left-wing and possibly humorous bumper stickers and t-shirts
I hope he doesn't have one. I'd have more respect for him if he wasn't one of those pretentious assholes.
Hey Rob, Thanks for that tarball!
"Going to war without France is like going deer hunting without your accordion." - Jed Babbin
Uh, because of RDRAM's "granularity", it was CHEAPER for Sony to use it over SDRAM or DDR. Read something written by someone other than Tom Pabst--btw, where did he get his EE???
W P_memory.shtml
http://www.toshiba.com/taec/components/Generic/
---
So a $200 upgrade becomes a $400 upgrade.
Now I have to be honest in that I didn't search very many places to see of an AT board was available, or if I could find an ATX board that took SIMMs, or with three ISA slots.... Partly, though, I didn't look as I recall a news item a year or so ago saying that Microsoft wouldn't "bless" a motherboard with more than two ISA ports. Still, over the last fifteen years of upgrades, I've rarely seen one that was as simple as it was supposed to be when you bought the original. You decide to upgrade one thing, and next thing you know, you've got to replace half the machine.
I've got a desk drawer that is a memory graveyard. I've got some 256k SIMMs, some 1 Meg SIMMs, some 4 Meg SIMMs. Just tossed two 32 Meg SIMMs on the pile the other day. I think I even have an old 2 Meg board up in the attic. (ISA, though I'd bet anything that it wouldn't work in a modern ISA machine.)
The cake is a pie
Hell, even if I wanted to upgrade the memory, it would have been nice for me to have the option of throwing in a new 64 Meg chip to get 128 Meg total for the prce of 64 Meg. Instead, I have to flush what I have.
There's lots of stuff that just doesn't need upgrading. A ISA 100/10 NIC used only for a two machine network? An ISA 56k modem used only as a backup when the DSL goes down? An ISA SCSI card used only for a cheap scanner? Why replace any of those? Oops, I have to replace one. The new motherboards only have two ISA slots.
One thing that is nice about the IDE interface is that it is completely backwards compatible. I've got an old 250 Meg drive that just does fine as a Linux swap partition. Why not? Why flush it? In fact, pretty much every drive I ever bought is still in use. I've got a 250MB, a 500MB, a 2 Gig and now two monster drives. All work great with the new hardware. Wish memory was that easy. I've got around 100 MB of memory that is basically useless to me now.
The cake is a pie
Notice the price for the 1Ghz Celeron? $750 bucks.
(And that is per 1000 units, probably going to be alot more, especially when the middle man gets through with you)
Granted the Athlon is alot more expensive right now. But notice that the 850's are about $700 bucks and they were released about a month back.
Now assumung Intel is on schedule and releases thier 1Ghz Coppermine 5 months from now, how expensive do you think the Athlon 1Ghz is going to be after 5 months? Or an even better question, how obselete is it going to be?
You gotta wonder what Intel is thinking...
"Imagination is the only weapon in the war against reality." -Jules de Gautier
Good point, but mulitpliers are getting higher, and (intel) caches are getting smaller! (At least at the moment.)
-Peter
I am relly starting to get worried about the ability of future machines to actually be upgraded at all. Seems like Intel is trying their damndest to eliminate that possibility. Is there any hint in actually making my investment a little more viable?
Slashdot social engineering at it's finest
Perhaps it is referring to fiscal quarter 3, of fiscal year 2000? If Intel's fiscal years start in July of the previous year, then tomorrow is the last day of 3Q2000.
(Tight fiscal quarters =anagram>Strategic half-squirt.)
[
> If so, there are going to be a lot of 1GHz Athlon processors sitting on the store shelves months ahead of the Intel shipments.
I don't know what the OEMs have access to, but I've been watching pricewatch.com ever since Intel and AMD fell over themselves trying to be the first to announce a 1GHz chip.
Intel is getting stomped. Sometime in the last 24 hours a link for Intel 866MHz machines showed up at the site; right now there are now five entries for PIII 866s ($975-$1035) and none for Xeon 866s.
Meanwhile, there has been a single listing for an Athlon 1GHz chip ($1399) for 2-3 weeks, there is now a single listing for an Athlon 950 ($985), and there are two listings for A900s ($885, $986) and about 30 for A850s ($718-859).
Without boring you with the details, I can add that there is a similar mismatch between price and availability for 800MHz chips from the two makers.
It looks to me that Intel is suffering a strategic rout in the high-end x86 chip market.
--
Sheesh, evil *and* a jerk. -- Jade
Intel makes good products, sure. But lets face it, they've been gouging customers for years. It's harder IMO to run a monopoly on the hardware side than the software side for reasons I won't mention.
:)
So when someone comes along with a decent, similar product, they can undercut your margins significantly. Thats called competition. So instead of taking the hit from AMD and riding it out, they try to pull a fast one on the entire computer world, and use this RDRAM fiasco, FUD marketing, and strong-arming their customers (well, thats conspiracy ala tom's hardware, but I would expect nothing less from a large corporation) to keep their near monopoly-status. It most likely won't work.
Face it, the shareholders get really pissed when profit levels drop. Maintaining their previous profit margins for the long term is near impossible. Today the only way to do it is with proprietary products like RDRAM, and similarly with their chip packaging and even MMX.
So instead of intel dealing with it properly, they shoot themselves in the foot. Capitalism works once in a while, I guess
Since I mentioned RDRAM, what is the deal with this article? It looks like marketing by intel. Every review I've seen of RDRAM either trashes it or gives it marginally better rankings than SDRAM, but even then the price difference doesn't make it worth it. Yet this site thinks its great, and states that SDRAM won't be able to keep up. Huh? DDR SDRAM is loooking pretty damn good to me. I don't want the extra latency from RDRAM anyway. What a load of FUD.
Thank you, my rant is finished.
Get our asses beat by AMD time and time again
Allow Transmeta to take over the mobile market
Pump billions of dollars into technology that we will never get to work
Keep talking about Merced or any 64-bit processor but never do anything with it
Sue/Get Sued by anyone with a heartbeat (also Microsoft! not sure if they have a heartbeat..)
Fake profit earnings after getting killed by AMD/Transmeta/Cyrix
Make a "KILLER" motherboard with modem/sound/video built-in!
Sabotage future Celeron-type chips from being overclocked or SMP'd
Make more bogus roadmaps
Mike Roberto (roberto@soul.apk.net) - AOL IM: MicroBerto
Berto
This article seems to be out of date on the day it's published. Or maybe this is subtler than it appears. For example:
Overall CPU speeds will increase as the year goes forward. No surprise there. 1GHz Pentium IIIs may hit the after market in late Q3 or Q4 2000. At this time, it does not look like Pentium III CPU speeds will go over 1GHz this year, or possibly ever, though Intel assures us that there is still performance headroom in the Coppermine design.
Now is this information pertaining to on-the-shelf availability of the 1GHz PIII's? If so, there are going to be a lot of 1GHz Athlon processors sitting on the store shelves months ahead of the Intel shipments. And if the news that the 1GHz PIII is all that we will see this year, then I suspect that AMD may have more of a march on Intel than we thought. Of course, Intel may well be banking more heavily on Willamette getting out earlier than planned to make up for the shortfall, in which case it will be interesting to see how AMD develop the Athlon line with its new faster caches in order to keep parity.
Cheers,
Toby Haynes
Anything I post is strictly my own thoughts and doesn't necessarily have anything to do with the opinions of IBM.
Why yields on Rambus have indeed gotten better, it is going to be a loooooooong time before they are reasonable. You can test the parts for basic function ok, but when you have to test the full 128 Meg module for speed, you have to have the whole puppy assembled with heat sink cement and all. One slow chip and the whole thing is slow, and you CAN'T tell beforehand, and you can't take it apart to replace that one part either. Gotta sell it as slow. Kingston reports yields of 50%, others as low as 10% for the full speed units. With crappy yields like that, the price is not comming down soon, if at all, on the high speed (and wasn't that why you wanted it in the first place?) units. Somehow, I think that with the price of double clock, folks will use that in their 820 chipsets. Indeed, there are some tests that have shown that Rambus actually UNDEPERFORMS a GX chipset pushed to 133. I don't own or short it, but be careful if you want to buy stock in that dog, sharkyextreme or no... Please remember that every once in a while Intel backs a real loser. Remember Bubble memory... My understanding is that Intel is going through the motions, but is going to keep their options open wrt Rambus.
Q3 00: Release faster processor. Trumpet it as 'great for internet'. Maybe call it the iCPU? Leave prices for current CPUs untouched, continue to make 50MHz improvements cost $200 over previous models.
Q4 00: AMD is at it again. Squish them, and pay off the FTC. Athlon really heating up. Engage in mudslinging campaign. Switch to smaller micron process to decrease die size. Ignore Transmeta's protests that their CPU has lower power consumption. Power is good. Suck lots of it.
Q1 01: Caught modifying CPU to make SPECint run faster. Damn. Use PIII ID feature to track down creators of SPECint and issue corrective measures. e-commerce is starting to wind down, start investing in hardware companies again.
Q2 01: Bunny suits aren't having the impact we want in the market. Find a pop rock band and rip off a new song - show competitor's CPU on fire.
Q3 01: Get posted to slashdot. Site crashes. Rob Malda mysteriously disappears
Q4 01: Can't do any marketing, Microsoft trial finally ended and DOJ needs a reason to keep their budget up. License out some worthless tech to competitors, later sue them for contract violations.
Q4 02: New SPECint benchmarks show our chips as fourteen times faster than the competition.
Q1 03: Another revolutionary new processor is created. Trumpet it as being Windows 2001 compatible (which was just released last week).
Q2 03: Bah, this far out, who the hell cares what we're planning on doing?
... and why it's not that great after all. At Tom's Hardware .
Intel make great chips. There, I said it. They also have a history of making good decisions. So why are they acting like a rabbit caught in the headlights?
For a long time Intel had the highest clock speed chips on the market. Their FPU kicked any part of the anatomy you care to sit on. Their chipsets were awesome. They drove Cyrix into a *very* small corner.
Then, AMD finally gets it's act together with the Athlon. Athlon is faster/better/cheaper (pick two:) than the Pentium 3. And Intel seem to go to pieces. This is not cause and effect, but I honestly believe that losing bragging rights to AMD has caused Intel to mismanage a series of problems that would have been merely embarrasing.
The Camino chipset had/has problems. Merced is so late it may be entirely overtaken by Willamette. Etcetera (it's not a long list of problems, but they seem to be screwing them up so badly I thought I'd say 'etcetera')
Then AMD beat them to the punch with a 1GHz processor. That must have hurt. Even though the Athlon was running it's cache at 1/3 clock rate, AMD got there first. By all of two days.
The (pass me another bucket of 'allegedly's) rumour is that Intel are having difficulty supplying demand for their faster chips (850MHz+), while AMD are happy to ship by the truckful. (More 'allegedly's please...) Other rumours say that some of Intel's second-tier customers are abandoning their Intel loyalty points and buying Athlons just so thay can ship some boxes. Fast boxes. They would love to buy from Intel, but their customers want the boxes today.
And Intel are *still* making stupid decisions. Backing RDRAM is daft. Nobody makes it in serious volume, it's five times more expensive than 'conventional' RAM, and it's proprietry. You can't make it without a licence from Rambus (Who don't have any fabs themselves). And it is at best marginally faster, and at worst astonishingly slower, than SDRAM.
To end the rant: Intel is a great company with great products. But Andy Grove should go and read his own book.