Slashdot Mirror
Intel Moves Up 32nm Production, Cuts 45nm
Vigile writes "Intel recently announced that it was moving up the production of 32nm processors in place of many 45nm CPUs that have been on the company's roadmap for some time. Though spun as good news (and sure to be tough on AMD), the fact is that the current economy is forcing Intel's hand as they are unwilling to invest much more in 45nm technologies that will surely be outdated by the time the market cycles back up and consumers and businesses start buying PCs again. By focusing on 32nm products, like Westmere, the first CPU with integrated graphics, Intel is basically putting a $7 billion bet on a turnaround in the economy for 2010."
I used to work for a processor company. I learned one thing: it's impossible to beat Intel, they just invest so much in technology that even if you come up with a smarter cache algorithm, a better pipeline, or (god forbid) a better instruction set, they'll still crush you.
That used to be true for the last 20 years. The only problem today is that no one really cares anymore about CPU speed. 32nm technology will allow Intel to put more cores on a die. They'll get marginal, if any, frequency improvements. We just need to wait for the applications to follow and learn to use 16 cores and more. I know my workload could use 16 cores, but the average consumer PC? Not so sure. That's why I'd like to see prices starting to fall, instead of having same prices, more power PCs.
--
FairSoftware.net -- where geeks are their own boss
At some point this roller coaster ride has to end. I mean, why not put off development until the NEXT iteration then?
"...Well, there's egg and bacon; egg sausage and bacon; egg and spam; egg bacon and spam; egg bacon sausage and spam..."
We;ve seen leaprog attempts lead to delays before. If this means AMD gets 45nm before Intel gets 32nm, doesn't that give AMD a performance window?
Actually they were able to step up some of there fabs faster then expected.
Despite the doomsayers, counting on the economy turning around by 2010 is a pretty safe bet. It's already being demonstrated that the housing bubble burst around September was not nearly as bad as the media/politicians made it out to be.
Intel is basically putting a $7 billion bet on a turnaround in the economy for 2010."
And if they lose the bet then they can just ask for a bailout like the financial firms and auto industry did. Because Intel is too big to fail.
The biggest issue for Intel is that most people already have computers that are fast enough for them.... Or, they don't have the money or desire to buy a computer.
The 32nm processors, I understand, will reduce the power needed even further, making it sensible for data centers to upgrade.
Or at least, if the economy *doesn't* turn around by 2010, that the shitstorm will be so bad at that point they don't care.
Pug
An Invisible Entity of Vast Power whose existence must be taken on faith alone: Liberal Media
a 7 billion dollar bet? thats peanuts! wake me up when someone makes a 1.5 trillion dollar bet on the economy.
After dealing with intel's op code hell, all I can say is that whoever comes out with a better method of recording op codes and their uses than the labyrinthine intel and amd manuals will become my patron saint.
But seriously is there any good x86 op code references out there?
I'll wait for -32nm.
There's no -1 for "I don't get it."
NEWSFLASH: Intel has been dumping 10 BILLION dollars a year into R&D since at least 1995. Did not RTFA, but if the blurb is to be taken at face value, the reporter obviously did no real research on the topic.
moox. for a new generation.
The shitstorm may be bad for them, but it'll likely be far worse for AMD to begin with. This is perhaps the best time for them to outspend AMD in research.
The Raven
For one thing, Intel has always been ahead of, well, everyone pretty much on fab processes. This isn't saying Intel will skip 45nm, they can't do that as they a;ready are producing 45nm chips in large quantities. They have a 45nm fab online in Arizona cranking out tons of chips. Their Core 2s were the first to go 45nm, though you can still get 65nm variants. All their new Core i7s are 45nm. So they've been doing it for awhile, longer than AMD has (AMD is also 45nm now).
The headline isn't great because basically what's happening is Intel isn't doing any kind of leapfrog. They are doing two things:
1) Canceling some planned 45nm products. They'd planned on rolling out more products on their 45nm process. They are now canceling some of those. So they'll be doing less 45nm products than originally planned, not none (since they already have some).
2) Redirecting resources to stepping up the timescale on 32nm. They already have all the technology in place for this. Now it is the implementation phase. That isn't easy or fast. They have to retool fabs, or build new ones, work out all the production problems, as well as design chips for this new process. This is already under way, a product like this is in the design phases for years before it actually hits the market. However they are going to direct more resources to it to try and make it happen faster.
More or less, they are just trying to shorten the life of 45nm. They want to get 32nm out the door quicker. To do that, they are going to scale back new 45nm offerings.
Makes sense. Their reasoning is basically that the economy sucks right now, so people are buying less tech. Thus rolling out new products isn't likely to make them a whole lot of money. Also it isn't like the products they have are crap or anything, they compete quite well. So, rather than just try to offer incremental upgrades that people probably aren't that interested in, unless they are buying new, they'll just wait. They'll try and have 32nm out the door sooner so that when the economy does recover, their offerings are that much stronger.
Over all, probably a good idea. Not so many people are buying systems just to upgrade right now, so having something just a bit better isn't a big deal. If someone needs a new system, they'll still buy your stuff, it's still good. Get ready so that when people do want to buy upgrades, you've got killer stuff to offer.
So... what's that xf86-video-geode driver for then?
Intel announced today that it was investing $7bln to build new manufacturing facilities in the US to manufacture these chips.
The new facilities will be built at existing manufacturing plants in New Mexico, Oregon, and Arizona. Intel is estimating 7,000 new jobs will be created. BizJournals.com
That would be the Cyrix MediaGX circa 1997.
I am becoming gerund, destroyer of verbs.
Despite the doomsayers, counting on the economy turning around by 2010 is a pretty safe bet.
Nope. Very bad bet.
If it were just a housing bubble it would have been a couple years of recession and we'd be coming out of it about then. The people and institutions who wrote bad mortgages and the people who bought houses too high would be hurt or bankrupted, the housing prices would drop to something sane, construction would slow (or stop for a while) until the unsold inventory and foreclosures had been sold off (or destroyed by neglect or arson for insurance) then pick up, and the capital now tied up in housing construction would be moved (again at a reduced price) to other productive uses. We're seeing a bit of that now.
This time they "securitized" the bum mortgages and "bought insurance" - "credit default swaps" - to the tune of MORE than the Gross World Product, in order to get multi-A ratings on the paper backed by baskets of subprime mortgages. When the housing prices started down a bunch of people defaulted all at once. So those who "wrote the insurance" had to dump a whole bunch of commodities on the market (depressing the prices further) to raise funds to "pay off the insurance". Thus when "Fannie Mae and Freddie Mac exploded" there was a lot of collateral damage in other markets. But that also would have sorted itself out after a couple years.
Unfortunately, the governments of the world, especially that of the United States, decided to try to "fix the problem". And now they're replicating EXACTLY the class of mistakes that turned a similar recession into the Great Depression - but more extremely, more rapidly, and without the safety net of the gold standard. The result, IMHO, is that we're probably in for a depression that will make the '30s look mild and short. And hyperinflation seems far more likely than not.
Thus my sigline.
As I see it, too much has been done ALREADY for a proper recovery to get started around 2010. (For starters, we're only about halfway through the underlying housing market collapse: The subprimes are largely crunched. But the teaser rates on a lot of other mortgages are expiring and even the government's billions of unbacked paper can't push the interest rate down far enough to save them - just to stretch out the agony.)
If Intel is betting the farm on a hi-tech recovery in 2010, somebody else will probably own the farm in 2011.
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
The 90's recession should have been much worse, enough to pull the debt to income ratio back into line. It would have sucked, it may have been nearly as bad as the Great Depression. Instead since then almost every western country has been running their economies on credit cards and home loans leading to stupefying ludicrous bat-shit insane levels of debt. And when they ran out of rational borrowers, they started lending out money to anyone with a pulse with no credit checks and invented all those stupid ways of hiding the risk that you mentioned.
So now that the whole mess has been exposed and house of cards is finally *starting* to fall, there is simply no way to stop it. It's going to hurt and it's going to effect absolutely everyone. No investment or currency will be safe.
09F91102 no, 455FE104 nope, F190A1E8 uh-uh, 7A5F8A09 that's not it, C87294CE no. Ah! 452F6E403CDF10714E41DFAA257D313F.
Do you have any idea how much computational power is needed for decent automated image analysis of the crap on alt.binaries.* to return RELEVANT results to SINCERE, specific, PORN requests?
I can't believe how much mis-classified crap there is out there. It is almost as bad as spam.
Intel is basically putting a $7 billion bet on a turnaround in the economy for 2010."
What a terrible bet.
Respect the Constitution
If you didn't learn that throughout grade school and college, then I don't know what more to tell you.
Nothing the R's did will help the situation. It was all just a final golden hand job from the government to the bankers.
Nothing the D's will do will help the situation. It is all just a final golden hand job from the government to the usual dependents.
Between them the currency is fucked.
Europe is no safe bet, neither is Asia.
Arguing over blame is pointless. They are mostly long dead anyhow. Those being blamed (and doing most of the blaming) are just the latest in a long line of check kiting fools who learned from their fathers how the 'game' was played.
Good luck to us all.
I see the next killer app as part social networking/part barter network. It will likely be craigslist.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Steel, brass and copper jacketed led is always a safe investment.
Good fun in the best of times...
This time they "securitized" the bum mortgages and "bought insurance" - "credit default swaps" - to the tune of MORE than the Gross World Product
Gross World Product? Try Net world wealth. (How did they calculate that?) Granting that's notional value on those derivatives so they can't all pay off, many are conflicting 'bets'. We can let that unwind by simply letting AIG collapse and flipping off all their counter parties. Fools put down million dollar bets with 'fly by night bookies' and got burned. They won't do that again. Which is exactly what we want in the long run.
The banks have always printed money and paid themselves 100 basis points.
Any asset you can't put your hands on and protect will likely be worth very little to you in cold hard yuan.
Anything the government wants to do is likely to protect a particular powerful constituency. It will nearly always be bad for everyone else.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
The smaller feature sizes bring power savings as well. So they're taking the server of yesteryear and putting it in your pocket. They're delivering the technology that'll bring the next billion users online because those folks don't have the watts to burn that we do.
They're also working to solve the whole I/O problem with servers that happens when you get too much processing power in one box.
In fact, they're pretty well focused on not just learning new things and creating new products, but in delivering new technologies that improve the way we work and live. And then letting go of it so we can figure new ways to use it that haven't occurred to them.
That's so different from the next story down where another company is getting raked over the coals for dumping money into R&D, because that other company is so famous for clinging to every ounce of leverage they can get out of every vague interpretation or use of their innovations and so toxic to deal with that they could deliver an all electric hovercraft that cured cancer and nobody would want to partner with them.
Sweet.
Help stamp out iliturcy.
I am wondering why the extra space isn't used for (much more) cache, rather than for more cores. Especially because more software can utilize extra cache compared to extra cores.
RSA is a problem that is much more simply stated than landing a man on the moon. You only say landing a man on the moon is easy because it was done. It was the culmination of many, many years of research to do it and it requires a lot of risk management and luck to do it. You say mathematically that landing a rocket on the moon is easier than protein folding, but try a realistic computer model of the effects of fuel spray and burn inside the combustion chamber.
This is my sig.
Because space travel is mathematically dead simple
It's only dead simple if you have a rocket that works. Design one of those? If it were so easy, SpaceX would have people up there by now, and I don't even know if they have their first orbit yet.
This is my sig.
Mon chien, il n'a pas du nez. Comment scent-il? TrÃs mauvais!
Actually, space travel is very complex. The only "simple" part about it is that, for two body motion and the limits of our ability to control thrutser force and duration, there are explicit solutions to the differential equations. The brain power behind the programming is immensely difficult, but once coded the computational power needed is not excessive.
More to the point, all the pencil and paper math HAD to be done to make the available processors capable of performing the operations. The fact that they had slide rules indicates that the complexity of the brain work was immense to reduce the solution set to something that can be solved near-real-time on a slide rule. If the same mission were done today, we'd have none of this higher math involved. With the available processor power, it would be a brute force numerical solution. That's what most video codecs are, in essence, is a numerical solution to an equation with known boundary conditions. The more compression you want, the less exact the solution is (And hence the compression artifacts).
Short of computationally intensive activities like video decoding, it shouldn't take much processor power to browse the web. It only does because it's faster (from a programmers time) to do things with brute force than to slim them down. It shouldn't require 250-500+ separate requests to open a page, and there shouldn't be 200kB of formatting for a page which contains - maybe - 5kB of text. That's why Skyfire works so fast on cell phones - there's so much crap in HTML pages now, and so many requests, that its faster to make a VGA snapshot of a page and load that as a damned image than it is to download the actual page.
Is it just my observation, or are there way too many stupid people in the world?
If you're going to invest in ammo, do it NOW. Then store it in a "safe state" where it will remain legal while this gets sorted out:
Several states have legislation in progress that, if passed (and upheld), would require ammo to have the equivalent of serial numbers - and ammo without it to be destroyed. All purchasers would go into a database along with the serial numbers.
Of course that's constitutionally problematic - especially the destruction of existing ammo part. But ammo will get hard to buy (above ground) in any state that passes this until it gets sorted out by the courts (by which time you might have needed to have it) and you don't want to be the test case.
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
Someone at Intel told me that 22nm chips will come after the 32nm design rules, code named Haswell.
See this Register article: Intel adds 22nm octo-core 'Haswell' to CPU design roadmap.
Intel's chipset design and fabrication progress is AMAZING. Everything else about Intel is backward, in my experience: Web site, marketing, behavior toward employees, and so on.
Just in case you haven't seen this yet; Job Losses In Recent Recessions
09F91102 no, 455FE104 nope, F190A1E8 uh-uh, 7A5F8A09 that's not it, C87294CE no. Ah! 452F6E403CDF10714E41DFAA257D313F.