Slashdot Mirror
Intel Scraps Plan For 4 Ghz P4 Chip
bizpile writes "It was reported earlier that Intel would be delaying the release of their 4Ghz Pentium 4 chips, but it now appears that they will be cancelling them altogether. The announcement came Thursday and Intel says they are going to rely on approaches besides faster clock speed to improve the performance of chips. Engineers are working to add additional cores to a single chip and improving the efficiency in how the chips interact with the rest of the system. Intel spokesman Chuck Mulloy said, "Those are the sort of things where you get more capability out of a processor by designing specific silicon solutions as opposed to just keep turning the clock faster." In the meantime, Intel is planning on releasing a 3.8 Ghz chip with 2mb of cache."
Mhz do not always = performance!
Wasn't that the entire reason behind AMD's use of the P-ratings? That performance was measured in more than just MHz.
Hell, Intel has spend DECADES convincing the public that MHZ is king and now they are (once again) following AMD's lead.
HA!
-Charles
Learning HOW to think is more important than learning WHAT to think.
Actually, Moore says that chip complexity will double along with relative performance, not clock speed. If Intel goes ahead with dual cores, and maybe quad cores later, then Moore's law is safe...for now
Does anyone really care about clock speed anymore? Yes, I know some applications need all the muscle they can get, such as video manipulation and scientific computing. However, it seems the interest in clock speed has waned considerably since the 1 GHz mark was hit. Basically, unless you are doing high end gaming or one of the aforementioned activities, increasing clock speed does very little for you. Consequently, it seems to me that the inevitable increases don't garner the same excitement they once did--going from 133 to 166 MHz was a big deal. Going from 3.0 to 3.8 GHz isn't nearly as useful, though the percentages are the same.
But what about Moore's law? Is nothing sacred?
Sometimes less is Moore.
Intel continued to use the MHz race because the public was on board, and simply because they were able to maintain a demonstrable lead in the race due to their process technology lead. They preserved their enormous market share and high margins by spending decades convincing the public that MHz was the key.
It will be difficult for them to apply as much inertia into another simple metric that the public will understand and by whose measure they will be able to remain the clear leader. They need to come up with another marketing story that pushes yet another metric that is again closely tied to their process superiority. I don't know what this is, but I'm sure they have a new story that we will see when they do their multi-core HT rollout.
AMD did not exactly "win" simply because they gave up the MHz war so soon. Yes, they were the first, but they didn't have much of a choice since they knew they could not scale to 65nm process geometry like Intel could. They had to alter their architecture earlier. Intel did not, and it worked in their favor for more years.
It is obvious from the past that Intel's marketing story will never resemble AMD's. They are not "following AMDs lead" unless by that you mean they were able to scale clock speed for a longer time than AMD was.
Intel spokesman Chuck Mulloy said, "Those are the sort of things where you get more capability out of a processor by designing specific silicon solutions as opposed to just keep turning the clock faster." In the meantime, Intel is planning on releasing a 3.8 Ghz chip with 2mb of cache."
So to sum up:
1) We've realized it's dumb to just keep increasing the clock speed.
2) Buy our new Pentium 4! It's going to have a higher clock speed!
~Philly
Common misconception. Electrons don't move at the speed of light. In fact, electrons aren't the primary charge carrier in half the transistors in the chip. Holes are (P vs. N).
Charge carriers propagate at about the speed of molasses. Go read this website, it is great:
http://amasci.com/miscon/eleca.html#light
Here's an excerpt --
THE "ELECTRICITY" INSIDE OF WIRES MOVES AT THE SPEED OF LIGHT? Wrong.
In metals, electric current is a flow of electrons. Many books claim that these electrons flow at the speed of light. This is incorrect. Electrons actually flow quite slowly, at speeds on the order of centimeters per minute. And in AC circuits the electrons don't really flow at all, instead they sit in place and vibrate. It's the energy in the circuit which flows fast, not the electrons. Metals are always full of movable electrons, and when the electrons at one point in the circuit are pumped, electrons in the entire loop of the circuit are forced to flow, and energy spreads almost instantly throughout the entire circuit. This happens even though the electrons move very slowly.
https://www.accountkiller.com/removal-requested
You're also off the mark. It is almost certain that there is no electrical pathway that spans the chip without hitting some logic. The number in 90nm (for best performance) is about 12000\lambda (\lambda = 90nm). Often signals propogate much smaller distances in a cycle. I assure you in one cycle no one is making a signal traverse the entire core. Modern CPUs are highly pipelined which is essentially to say that in one clock cycle data is transfered and processed within a very small section of the chip before being passed on to the next stage. This then frees the stage for the next bit of data. see http://en.wikipedia.org/wiki/Pipelining As a side consequence, what you mention is not the limiting the factor. Signals simply do not need to propogate across the chip in one cycle. What has really happened is the drive current available from each transistor has gotten smaller as the transistor itself has shrunk. The wiring capacitance has remained the same and begun to predominate over the gate capacitence. Thus, making the transistors smaller does not make the circuit faster as it once did. Also, as someone else pointed out, the mobility of electrons in semiconductors is no where near the numbers you quote. Electronics simply don't work the way you claim.
Remember intel has done other things to increase speed other than just MHz increase. Such as: 1) Increase Front Side Bus (in the p4's case 400 -> 533 and now 800MHz) 2) Increase Cache (256 -> 512 -> 1024 -> 2048kb) 3) SSE 1, 2 and 3 4) HyperThreading
You're kind of missing the point.
:) ]
What you're not getting here is that it is INTEL that has been behind the clock speed myth. They have spent untold millions (billions??) teaching people that the speed of a computer is best measured by the clock speed of its CPU. For the last decade, that and "Intel Inside" have been their ENTIRE marketing message. The consumers believe that clockspeed matters because Intel is the one that told them so.
Now, for a long time, this has worked really well for them. They pretty much destroyed Cyrix this way, and AMD has been struggling for many years. Cyrix came up with their PR-ratings to try to be competitive, but their chips weren't very good and didn't deliver on their promise, and they sank into obscurity. AMD did the exact same thing with their + ratings, but they were so conservative about them at first that people accepted them. (this gave them some weasel room later, as they have gotten very nearly deceptive with the ratings on some of their CPU lines, particularly the Sempron.) They had to do this because Intel had taught everyone that it was megahertz that counted: AMD couldn't deliver that, just performance. Basically, they got lucky. Had consumers not accepted those ratings as accurate, AMD would probably be gone now. Apple was in the same boat, as well. With a less rabid fan base, they'd be gone too.
Around the time of Rambus, the marketers took over Intel. They realized that the megahertz message was working fabulously well. It appears that they decreed that all future engineering efforts in the Pentium line would be oriented around cranking up the clockspeed. The engineers delivered what they were told to, a chip that could be scaled a very long way, by going to a hyperpipelined approach. I believe their first P4 was clocked somewhere around 1.2ghz, and it was HORRIBLY slow because of the pipelining; a 1ghz P3 absolutely destroyed the P4. In other words, the P4 was a big step BACKWARDS from the P3 in nearly every way.
But then they started to crank the megahertz, expecting to leap way out in front of AMD and, once again, dominate everything. (Nevermind that it wasn't until the P4 hit about 2.4ghz and got an 800mhz bus that it started to actually get good.) RAM speeds in particular had to do a lot of catching up. A hyperpipelined approach suffers terribly from a mispredicted branch. The CPU stalls completely until the pipeline can be refilled, which kills performance. You need the fastest possible RAM to refill the pipeline as quickly as possible. (and this, btw, is why AMD isn't as desperately dependent on fast memory; its pipeline is about half as long as the P4's, and thus it doesn't choke as badly if it guesses wrong about a branch.) [and thanks to Ars Technica for the knowledge to write this last paragraph
So all of a sudden, over the last year or so, Intel suddenly ran into a brick wall. Their entire chip design culture is clockspeed, not performance, and abruptly they can't crank clockspeed anymore. This is a BIG DEAL, because they're going to have to tear apart and rework EVERYTHING internally. This blunder is going to cost them billions, and if AMD keeps executing as well as they have recently, they could lose a great deal of marketshare. They are already losing mindshare, since AMD got to specify the instruction set for 64-bit X86.
Intel is in TROUBLE. The focus of their entire company, their raison d'etre, no longer exists. They forgot they were actually about performance. Many of their existing projects will have to be scrapped, and they'll have to reorient most of the company in very short order, while still maintaining morale.
If anything can save them, it's the Pentium-M, which is an extraordinary piece of technology out of their Israeli branch. In many respects, the M is the direction Intel should have gone five years ago.
Can they make up for this vast blunder? It's a good question, but I wouldn't count them out just yet. If the engineers
I don't know if I'd agree exactly with this comment. While a 3.8 GHz P4 does not perform as highly as a 3.8GHz Athlon chip would, an AMD chip can not physically run at these speeds. The pipeline would not support it.
The slashdot crowd is quick to attack Intel because they're the big guys, but the NetBurst architecture is an extremely powerful and (gasp!) good architecture. While the engineers designing it designed a processor for maximum pipelinability (over 30 stages now) this is not really a bad thing. Pipelining a processor is a good thing in general. Its main claim to usage is that it allows a processor to run at a higher clock speed. That is what pipelining was created for; to break down the time into smaller slices so more can occur in parallell. This process works great when each stage is of approximately equal length, and I have enough faith in the Intel engineers that no single stage was much longer then the next longest stage.
Back to the point though the pipeline does have downsides. A processor with 20 stages will lose ~ twice as many cycles on a branch missprediction (and more on a cache miss, but that number varies further) when compared to a 10 stage processor. However assuming that by using 20 stages we cut the cycle length by even 50% the additional stages were worthwhile. Cache misses are not a "common" event and branch prediction is in the 95+% range now, so the stalls added there are not as large as you'd think.
What the pentium 4 has done was manifest these to a larger scale. Unfortunately the engineers desiging the processor did not realize the massive leakage currents that are seen with processors at the speeds Intel is using. From a computer architect's standpoint they build upon past assumptions, and more stages in a pipe generally help out, so thats what they did. While the end result is not as impressive as they were hoping the end result is not a poor product.
Now what has the NetBurst architecture offered to the consumers? Well one of the main offerings its had is building an SMT processor (hyperthreading in marketing speak). SMT is more then mere marketing hype. It was not an afterthought thrown onto the P4 due to less then stellar performance as people have hinted at. SMT was originally designed for the Alpha ev8 chip that was scrapped. Intel however bought the alpha design team and used the SMT technology (albeit to a lesser extent then some would hope for) in the NetBurst architecture.
What else has NetBurst added? The trace cache is a wonderful feature as well. This removes the x86 decode logic from the runtime pipeline for most instructions.
So where can Intel go from here? My hope isn't so much in the multicore logic that some talk about. While multicore is interesting, I personally would rather see a wider P4 core (more execution units) and have them extend their implementation of SMT to allow for more concurrent threads of execution. a 4 or 8 way SMT processor could show some real results.
And for those of you who are going to question what I'm saying... No I don't work for Intel. And no my desktop processor is not an Intel processor either (I run an athlon 1600 for my workstation). However in my lab I am working on algorithms designed specifically around SMT processors (as well as cache aware/prefetching enabled applications). Intel's processors happen to enable quite a bit of optimization if done properly.
While I never agreed with Intel playing the MHz game, or their ridiculous prices, I would not say that the engineers were completely against the super-pipelining of the NetBurst architecture. While they may have questioned the reasons behind it, the real world performance gain does exist do to it.
Philip Garcia