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."
I knew this would catch up with them. I'm glad Intel is off the MHz thing. This doesn't mean the general populace will be more informed when buying a processor, but at least they might be looking at other features that may matter more (i.e. shared video memory, backside cache, etc.). Maybe.
Per Square Mile, a blog about density
What this means is that Intel will probably be releasing a multi-core HT product in the same market window that the 4MHz part occupied.
Isn't this a full quarter in advance of what we expected? Won't this put their release in the same window as AMDs multi-core release?
You can't increase clock speed indefinitely. There's a fundamental limit we're brushing up against here, and it's called 0.8c.
Electrons on copper travel 3cm per nanosecond. At four Gigahertz, each clock cycle, the electrons can only travel a theoretical maximum of 0.75cm. I don't even think that covers the diameter of a single core these days.
You can't turn up the clock much faster than it's already going without getting into nanotechnology. The only viable solution is to optimize chip efficiency through other means, and add more cores to the chip working in parallel.
occultae nullus est respectus musicae - originally a Greek proverb
looks like we are approaching a moore's law limit. i looked at http://www.spec.org benchmark submissions and found that 2002 4th quarter had the fastest CPU clocked at 3.06 GHz and in 2004 3rd quarter it is 3.6 GHz. This is just 18% increase over 21 months.
ofcourse, the CPU performance is still increasing at a rapid pace; but as far as MHz (or rather GHz) is concerned, we are reaching the limit faster.
WOW, my pc is 2 yrs old and I am not planning an upgrade!
Considering the fact that my 3 year old PC died, I replaced the 1.4GHz Athlon (T-bird) with the Socket 754 Sempron 3100+.
Same RAM, same disk, same video, but a new motherboard.
I *feel* like I'm getting more than a 28% speed boost from it, so it's clearly not just the clock speed that's doing it. Making a chip run faster never was the right idea, and I'm glad to see that they're walking away from that.
Now, if we can just get a core like the Pentium M, but for desktops, then maybe we'll see some real competition.
Raptor
"Procrastination is great. It gives me a lot more time to do things that I'm never going to do."
On the other hand, the Athlon XP range is End-Of-Lined.
Let's talk Athlon64 and Opteron instead.
Intel will have to put the memory controller onto the CPU sooner or later. If they want to go "Not Invented Here", it's going to cost them $$$ BIG $$$ BUCKS $$$ in cache.
Belief is the currency of delusion.
This sounds like a score for the good guys, with Intel finally realizing what others (like AMD) have realized alot earlier.
But...
Lets see what is actually going to happen. There are plenty of previous examples of Intel changing direction, and it is not always for obvious reasons. Remember slot1 and slot2, that Intel praised as a superior way to interface cpu's to motherboards as opposed to sockets, and when all came down to it, it was nothing but a stunt to try and make life harder for competitors.
Could this be a forced move by Intel, because they aren't capable of increasing the clockspeed and keeping cpus stable?
Quote: "Just as a die-hard linux user, I will say that the mHz wars are *so* 1990, the big questions that I ask is "How much cache" and "How many bits wide?"
Um... true, clockspeed isn't the only indicator of system performance. But what in the world does the number of address bits have to do with performance? And what does Linux have to do with anything?
Intel says they are going to rely on approaches besides faster clock speed to improve the performance of chips
Strange, I thought the point of the big numbers was to sell more chips, not to make them faster. Wasn't part of the reason that Intel made the P4 pipeline as long as it is so that they could keep cranking the MHz up for a long, long time so they'd have lots of generations of P4 processors to sell? Because I don't think you really need that long a pipeline for purely performance reasons.
I wonder if AMDs inroads into the 64 bit market have Intel getting a bit scared about the future?
Remember, Intel sells products to millions and millions of people, most of which do not realize that MHz does not equal performance. One of my friends was just complaining the other day how this laptop he wanted with a Pentium M cost more than the Pentium 4 but ran half as fast. Marketing rules all, and when you're trying to crank out a profit, you do what you need to do to sell your product.
Hopefully this means that we are going to be seeing a revolution in bringing multi-core processors to the desktop. Imagine a CPU that incorporates 4 cores, 4gb cache, 4gb ram, and 40gb storage all on a single die. At that point, the only upgrades you would need to worry about would be for mechanical drives like DVDRW and HighDensity Hardrives, and the latest graphics card. Of course some kind of liquid/vapor cooling would need to be used to pop out the full potential of these new processors, but then thats allways been the case. This is definately a right step in the right direction!
My fantasy involves a direct connection from my computer to my skull.
I recall some earlier discussions about how Intel was finally starting to wise up and design processors that are efficient, rather than just raise the clock speed.
The first incarnation of this is the Banias, also known as the Pentium M. It's basically a P3 pipeline, but with P4 branch prediction (and some other technologies). The P4 has to have very advanced branch prediction in order to even HOPE to get reasonably efficient use of its pipeline. Applying this to the P3's shorter pipeline results in a much higher IPC.
In other words, something philosophically like the Athlon.
Since then, I haven't heard anything about it. And then there's this article. Is there any relationship?
I just want a desktop Pentium M system, without having to browse some Japanese-only Hitachi site.
I don't want more power, I want a fast enough machine that runs silently.
I guess it's my fault for waiting for Intel to provide this instead of just buying a Mac.
Nah, they switched when AMD's speed ratings were higher/faster than any current or planned P4s. Really, if Intel could go from 3Ghz to 4 or 5 or 6 Ghz as quickly as it went from 2Ghz to 3Ghz do you think they'd ditch Mhz altogether? For all the campaigning Apple and AMD have done, the average person STILL thinks Mhz is the only think to look for in a PC. There's no way Intel would ditch that marketing advantage if they could help it. With Athlons quickly scaling to 3600 and 3800 speed numbers and little to no possibility of a P4 with a higher number in the near future, Intel's higher Mhz weren't much of an advantage anymore, so they switched to a naming scheme that couldn't be easily compared with Athlons (or previous Pentiums, for that matter).
For 3D Rendering all you need to do *is* just turn up the clock speed. It doesn't matter how fast the memory bus is... or even how much cache is on a chip beyond a certain mimimal level.
You can build super cheap (except for processors) computers to use in a renderfarm.. (I use Lightwave 3D, Modo, and SoftImage XSI)... and hard drive speed / graphics card speed / Memory speed / Cache on die, Do nothing to speed up a render once you hit that "Render" button. Sure... SSE extensions and the like do speed it up if the code is optimized... but there isn't really a way to optimize the code with this new direction Intel is going.
I am Jack's HTTP Server
But the droids blinkered by intel FUD put their fingers in their ears sang "lalalalala" and barked "NO - faster clock speed is a FASTER CHIP!!!"
Now, suddenly: oOooooo - cycles per second isn't as important!
Oh well. It will certainly be very interesting to see what Intel does over the next few years.
Here's an interesting question, related to this topic:
Assuming they go multicore (like IBM and Power[x] chips) what are the limits involved there? What would logically stop the development of multicore chips from increasing their number of cores?
And: What next?
RS
Shoes for Industry. Shoes for the Dead.
I know it's not a multicore device, but this is an example of what's possible. 36 Gflops @ 220 Watts. (24 Gigs RAM, 1TB storage, $10,000) I want one.
"A language that doesn't affect the way you think about programming, is not worth knowing" - Alan Perlis
If Intel's primary motivation behind going from the Nortwood core to the hotter & less efficient Prescott core (longer pipelines result in a Presocott chip with double the cache of an equally clocked Nortwood actually being slower) was that the Pressy would allow them to scale to higher clockspeeds than the Northwood would allow does this make the Prescott a failure?
my sig's at the bottom of the page.
Intel released their Q3 results late Tuesday. In their conference call they were evasive about a suprising drop in their tax rate and also about the amount of their inventory writeoff. Intel claimed their inventory was down $43 million to $3.2 billion with an unspecified writeoff amount. Investors were happy to see inventory did not go up again and the stock went up Wednesday. In several different articles people are working out the mystery of the writeoff amount. Normally Intel's "cost of sales" is a steady number. Any writeoff will add to this number. So you can estimate the writeoff just by seeing how much this increased. With this calculation, it seems Intel had a writeoff of $472 million.
Okay, this means it's time for the CPU performance increases to take a back seat. Maybe now the rest of the computer can have some time to catch up better with the CPU. I am talking about bus and memory bandwidth. This is one hurdle that needs to be overcome.
Low latency and high bandwidth up the wazoo is one aspect that supercomputers for example have over standard pc components, besides massive parallelism of course.
It would be cool to see intel start making inroads from R&D on the memory front. I'm not talking about on-die cache, that is a given. The questions to be answered are how to get the main memory up to snuff with the rest of the system.
If the current state of the art in CPU power stagnated from here until 5 or more years from now, it really wouldn't be an issue if the same efforts during that time were put into lower latencies across the whole sytem architecture itself.
So what am I saying? The CPU has had enough innovation in it's current form. It's time to focus on other lagging components. Pci-x is a step in the right direction, but it is nothing without main memory advances and other mainboard bus architectural improvements.
>>>>>> Chewie, take the professor in the back and plug him into the hyperdrive.
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
It seems to me that there has got to be maximum rate at which we can push the clock.
I have a 3.2 GHz Pentium 4. How far can light travel in one clock cycle at that speed?
186000 miles / 3.2 billion is about 3.7 inches isn't it?
The race isn't always to the swift... but that's the way to bet!
It had the 'interesting branch prediction' because it NEEDED it. That deep pipeline sucked. A missed branch was a catastrophe, so you can BET they spent a lot of transistors there.
My primary focus at that time was on servers; for pretty much any application you could name, a P3 just spanked a P4 for a long time. Intel even shipped a few 1.4ghz P3s with double-sized cache, but then stopped when folks realized that this chip significantly outperformed much "faster" P4s. Yes, there were some desktop apps that really benefited from the P4, like video encoding, but as general-purpose chips, the P4 was inferior for a long time. The double-cache, high clock speed P3, which was an EXCELLENT solution for many problems, interfered with the marketing message, and was killed.
Every prior generation of chip was a substantial step forward, particularly up to the Pentium. Every chip through the Pentium II roughly doubled the performance of the fastest chip of the previous generation. The P3 was a significant improvement, but was more like a 50% bump. The P4, on the other hand, was a step BACKWARDS; the fastest P4s were slower than the fastest P3s when it shipped, and remained so for quite some time. It wasn't until the front speed bus got to 533mhz and the main clockspeed got to about 2.2 gigahertz that the P4 finally, truly started to win on raw speed... and on value (price/performance), it took longer still. And I'm totally ignoring heat and power, which can be big issues in some circumstances.
It's no mistake that the Pentium M is so darn fast for its clockspeed; it is, essentially, the old P3 architecture with a number of enhancements for low power usage. And it is electrically compatible with the P4. All a motherboard would have to do, in order to support it as a desktop CPU, is provide a different socket. I have no idea why you can't buy desktop boards for the Pentium M, it would be trivial to do. I assume it is, once again, interference with the marketing message.
Had Intel not focused so much on clock speed to the exclusion of all else, they could just start selling Pentium-Ms instead: they're ideally suited for multi-core. But they didn't, and now they have two very large problems at once, both technical and marketing. They have to revamp their engineering approach and re-educate their customers simultaneously, undoing 10+ years of momentum in both areas, without destroying their existing business. Not easy.