Slashdot Mirror
Intel Delays Release of 4Ghz Chips
bizpile writes "The AP is reporting that Intel's faster version of the Pentium 4 will not be available by the end of the year as previously promised. They told PC makers this week that the 4-gigahertz chip will not ship until the first quarter of 2005. Intel spokeswoman Laura Anderson said, 'We felt by adjusting the schedule for the products, we could better meet our customers' volume requirements and their high expectations.'"
"We felt by adjusting the schedule for the products, we could better meet our customers' volume requirements and their high expectations."
Translation = "full of bugs that cant be fixed in time"
Intel is still paying for their decision to go with the netburst architecture IMHO.
They wanted to be able to crank the megahurtz and use that as a PR device (well, not only that but it helped them).
Of course they are also having problem with the 90nm tech (as is IBM -- I think that only AMD has been mostly clear sailing with that), but most of their problems have come from netburst and lack of competitiveness in the budget sector (Celerons get killed by much faster and cheaper AMD chips).
Treehugger? Treehugger... Treehugger!
Clock speed really doesn't matter, anyway. Well, it's not as important a benchmark as Intel would have you think.
If a processor running at 4ghz can only do half the operations per clock cycle that a 2ghz processor can do, than it's no better than the 2ghz processor, and probably worse due to larger instruction pipelines, etc.
The fact that Intel has relied on this "Mhz Myth" has really killed sales of their Centrino (Pentium M) line of processors. Consumers see the (comparatively low) ghz ratings on the Centrinos (typically about 1.5ghz) and compare them to laptops with less expensive P4's (typically running between 2.5 to 3.5ghz) and wonder why anyone would pick the Centrinos.
Laura Anderson said, 'We felt by adjusting the schedule for the products, we could better meet our customers' volume requirements and their high expectations.
Suit to Geek Translation.
"We can maximize the profit we make off of our existing inventory by delaying the release of the new chips until we sell off the current stock."
LK
"Hi. This is my friend, Jack Shit, and you don't know him." - Lord Kano
Mod me up if you hate the color scheme. Here's a fixed link using the "old" slashdot colors:
0 9&tid=118&tid=137&tid=126
http://slashdot.org/article.pl?sid=04/07/30/22152
Back in 2000 when Intel was first surpassed by AMD when they were beaten to 1 Ghz, they rushed an overclocked 1.13 Ghz chip to market that eventually ended up being recalled. I suspect that the reason that the 4 Ghz chip is not yet being released may, in fact, have something to do with reliability. Also it is important to remember that this chip is running on the Prescott core, which will probably use over a hundred watts of power alone. Perhaps they need more time to explore better cooling solutions (that can be delivered cheaply) before rushing a potentially unreliable product to market.
I am in favor of reliable chips (although personally being an AMD zealot myself); I think that the competition between AMD and Intel is important for innovation and fair prices.
The new iMacs are not delayed to clear out inventory. You cannot buy an iMac now (At least in Australia). The production schedule for the current iMac was dependant on IBM producing G5 processors on target. IBM has not been. G5 iMac is delayed.
I would think Intel is having the same problems as IBM has been having. They just can't get the yeild required for a mass market.
"Does this indicate some kind of problem with the further fulfillment of Moore's law (you know, for once, Moore's law failing to apply NOW as opposed to "Moore's law will stop working in 8 months) or has this just been a bad year?"
Option #3 could be that there really isn't a killer app that requires that speed. I have difficulty imagining a lot of ppl flocking to those machines right now. It is a pity for Intel that 3d cards do more for games than cpu's.
"Derp de derp."
Intel is now feeling some pain. They've built a brand around having more M/Ghz - which only matter superfically.
Being a multi-function device means that a CPU does multiple functions. As with ANY multi-function device, a model of CPU will do some things better than others.
X86 chips have traditionally been processing heavy, I/O weak, since hard, on-demand processing hsa been the driver of the X86 industry. (Video games, etc)
Contrast that with the Sun Sparc line of chips, or IBM's mainframe hardware, heavily optimized for I/O throughput. The needs of a rendering farm node are not well in alignment with the needs of a high-capacity file server.
Even within being "processing" demands, there is a wide, wide range. Floating point. Integer ops. Parallel proccessing. Different, even cross-compatible chips and chip lines will behave differently, performing better at some tasks than others.
But, for years now, Intel has been busy spending millions convincing the population that you can boil performance down to a single number, M/Ghz.
The cracks are beginning to show. AMD has made a solid business with "slower" (Mhz) ships that perform better. Their own Centrino line is "slower" but performs almost as well!
Intel needs to get a clue, and develop a set of benchmarks that truly show real-world performance. AMD has done quite a good job with their "+" rating. (EG, my desktop is an Athlon 2000+)
I give it 6 months, maybe a year. It'll be hard, but even Intel isn't so stupid as to put this off too long.
I have no problem with your religion until you decide it's reason to deprive others of the truth.
I note the article is about Intel being unable to supply processors on the promised schedule. Yes, I realise that there's AMD around as well, but I don't see how switching to x86 will solve Apple's problems.
The basic issue with Motorola was that Moto weren't interested in developing new high-end CPUs. Apart from Apple, they were only targeting the embedded market.
IBM, on the other hand, has to develop new high-end chips, because they are required for their P-series (RS/6000) and I-series (AS/400) servers. In fact, IBM has already produced the chip that the next-generation PowerPC will be based on - the Power5. (G5 Macs use the PowerPC 970, which is a cut-down version of the Power4.)
The issue with higher clock speeds - whether from IBM or Intel - seems to be an industry-wide problem with the 90nm process. It's so bad that IBM has announced that "scaling is dead". See also the scary power dissipation of the new Pentium 4 chips.
When was the last time you saw Intel or AMD delaying a product by months, and causing problems for suppliers?
I read somewhere that Intel was going to be delaying release of their 4Ghz chips. I forget where.
When was the last time you saw Intel or AMD delaying a product by months, and causing problems for suppliers?
Prescott?
Or if I wanted to be really mean I could mention Itanium. That was, what, 5 years late?
AMD tends to be very conservative with its timetables, but even they have experienced problems.
As for IBM, they've run into a wall at 90nm. But so has everyone else. Expect to see lots of announcements of dual-core (and multi-core) chips, and larger caches, but no great increases in clockspeed in the next few years.
If this went on long enough and if we truly are at the end of straight line scaling, the industry might become driven by the one-more-GHz per year rule (the new More Law), versus doubling every 18 months. This new law could then hold for decades as it slowly curves down towards a flat line. I don't actually predict this will be the model soon, as the old Moore's Law is more likely to adhered to, but in 24 and then 36 month time frames for as long as possible. Still, if scaling is dead (and some are saying it is) then we could see the new "More Law" adopted as IT shops and Manufactures try to plan for future purchases. Software providers wouldn't be able to count on Moore's Law bailing them out. Bad news for Longhorn if scaling is dead, it might always be perceived of as slow (if /. reports are to be believed).
We are already putting 200+ million transistors on CPUs, but most speed increases come from scaling (speed increases) and memory caching. Now is the time for the industry to go Multi-Core. How about 100 two-million-transistor cores on a chip instead, with 500 separate integer and floating cores that can be shared across cores as needed.
BTW, I do know the real Moore's Law is about the number of transistors on a chip and not speed, but the two have been synonymous in the public's mind since the 80s.
Letter To Iran