Slashdot Mirror
Intel "East Fork" Technology Migration
Hack Jandy writes "When Intel's Centrino platform first unveiled, industry experts were surprised to see such great performance of the Pentium M, based off Intel's P6 (Pentium III) architecture. According to sources in the industry, Intel has officially adopted the approach to migrating Pentium M to the desktop (hence, "East Fork") to offset some of its Pentium 4 processor sales. Cheaper, slower, cooler, but higher performing processors are on the way to an Intel desktop near you!"
The cooler they can keep a well-performing CPU, the less noise they need coming out of the box. Let's count this one as a victory for using PCs for PVR/Jukebox-style uses.
500GB of disk, 5TB of transfer, $5.95/mo
Gonna be great to use this platform for servers..
a sp?id=dot handesktop&page=1
Low power usage...
Great performance..
Low heat emission (easy to make passive cooled..)
GamePC made a test not long ago, and it performed on par with p4EE and amds FX5x...
http://www.gamepc.com/labs/view_content.
http://www.reuters.com/newsArticle.jhtml?type=topN ews&storyID=6786951
Since it's from Reuters anyhow... old news too (11th Nov).
No Norm, those are your safety glasses; I'll wear my own thanks...
So yep, they respond very quickly to customer needs and wants.
Even those who arrange and design shrubberies are under considerable economic stress at this period in history.
I don't think that you are seeing the whole story. Basically, Intel has been holding out for IBM's silicon-on-insulator technology because it reduces power requirements a good deal. Unfortunately for Intel, IBM is pretty sneaky when it comes to licensing and often prefer to swap technology rather than accept cash. I'd imagine that IBM is holding out for an x86 cross-license agreement while Intel does not want to give that up.
What you've seen in the past couple years is a game of chess. With each move, the other hopes that they have positioned themselves to better reach a licensing deal. Intel's move to non-clock processor ratings was a big move in this game.
From what I've seen at Intel's developer forums, they're working on some radically different architecture. Something that isn't von Neumann at all. They're calling it "massively parallel" but the industry seems to think that this means multiple cores on one chip. I think that it means thousands or millions of "processing elements" on one chip (think really small processing elements). Their claim is that they'll be able to apply this architecture to everything from mobile to high-end servers simply by adding or subtracting elements as power constraints allow.
Life is the leading cause of death in America.
I noticed that every x86 CPU architecture in the past decade climbed 4-5 times in MHz from inception to the "end of the line" model: 486 - 25..100(???, 133 is AMD's version and those started higher than 25), Pentium - 50..200, Pentium4 - 1200..3600 now and still has a tad in reserve as shown by extreme overclockers; similarly for AMD, K6 - 166..550; Athlon - 500..2.x(?). And now Pentium2/3 - started at 233 and climbed until around 1300, which is higher than 4/5x. But maybe there's been some really notable arch changes since P2? What're your thoughts?
This is really about Intel finally coming to terms with the fact that nobody wants to buy Itanium chips. That's where Intel was headed, and Intel assumed that everyone would follow along. Unfortunately, Itanium's future depended on technology advancements that never happened, and a rate of adoption that nobody was willing to pursue.
This is why Xeon became an architectural dead end: Intel wasn't willing to move the technology forward, because Xeon was supposed to be superseded by Itanium.
Did you know that "Pentium M" is actually based on the same technology they originally called Pentium Pro? It's true. It was a good design. It didn't do all that well initially because its 16-bit performance was abysmal, and people were still running a lot of 16-bit software at the time. Now that everything is 32-bit, Pentium Pro (now Pentium M) is just fine. The fact that it gets used in laptops is a testament to its ratio of performance to power consumption.
Intel would be wise to move forward with this. They ought to ditch Xeon entirely, and perhaps even graft the AMD64 instruction set onto this chip.
Tired of FB/Google censorship? Visit UNCENSORED!
The probelm for Intel is this: By the time they get this chip to market, or certainly not long after, Microsoft will actually ship Windows XP 64.
While the Pentium M may be able to close the gap to the Athlon 64 when running in 32 bit mode, possibly even beat the AMD chip if Intel are successful in increasing the M's clock speed, the Athlon is just waiting to really stretch it's legs. In some situations moving to 64 bits will not improve performance, and could possibly even hamper it, but for the majority of desktop applications and games with optimised code the 64 bit version with the extra registers will trounce the 32 bit chips.
Precisely. I would be suprised if they could make such a chip and keep it both X86 compatible, and fast for todays applications. If it's only slightly parallel then it's no more than a dual core chip with hyperthreading. If it's massively parallel like the grand parent post suggested, then each individual thread is unlikely to run as fast as a P4 or Athlon64 chip today, and that will hurt applications that don't benefit from being mulithreaded (ie. most of todays unoptimised apps).
The Banias cores had 1MB L2 cache and the new Dothans have 2MB L2 cache, yes, but that's not the sole reason for them performing as they do. The current Prescotts have 1MB L2 and Intel's slated to introduce 6xx P4s in January with 2MB L2, but I can promise you that the additional cache won't suddenly cause Prescott to perform simliarly to Dothan clock for clock.
It's a design philosophy based around high IPC, not the large cache, that makes the Pentium M such a strong performer.
Dude, computers have not been even *close* to Von Neumann for several decades.
Von Neumann assumes uniform memory access times - this is largely untrue for any ordinary scalar processor with a cache - ten years ago processors had internal memory write buffers, l1 cache, possibly l2 cache and main memory - today it's even more complicated. But common for all of this is that even a simple hierarchical memory system with a single level cache and the main memory makes your computer very far from Von Neumann.
So, if Intel wants to present something that "isn't von Neumann at all", all they need to do is pull an i386 out of a hat and wave it at the drooling masses.
SOI only helps reduce one particular source of static power consumption. While static power is a big issue at 90nm, SOI doesn't magically solve it. Further, the big problem with Prescott is power dissipation and heat under load--dynamic power consumption. I'm not sure where you heard this rumor, but even if true it's ancillary to the current discussion.
I'll believe that when I see it though, I don't think Intel would do it if only for the reason it's not as marketable.
Of course its marketable. The new model number scheme puts the p4 at 4xx, while the M is 6xx. Thats 200 more. It must be a lot better. I knew they'd do this as soon as they came out with the model numbers.
Note: I'm not a moron. I'm just writing what "joe sixpack" thinks.
Cringely had an article a while back that mentioned Google liking to use Pentium IIIs in their data center. Yes the Pentium 4s were faster, but if you looked at your datacenter as a whole system, including power, cooling, and space requirements, they were better off with 'old' Pentium IIIs. At the time, I think Google was worried they wouldn't be able to source new machines with P-IIIs, looks like Intel is following them this time. Intel seems to be following a lot lately, the megahertz at any cost mantra sure faded fast.
I don't think so.
Intel has basically been hanging itself with the awful lot of rope their own marketting gave them. The "MHz is everything" marketting was an easy thing to push, since most people actually _want_ one number that tells them everything about a CPU.
(True story: I actually spent some time arguing with a marketroid about it, and gave up. He was arguing that it must be Anantech's and everyone else's benchmarks that are at fault, because CPU A is in some apps 50% faster than CPU B, in some apps equal, and in some apps actually a little slower. "It can't be! If CPU A is X% faster than CPU B, it must be X% faster in everything!" Any explanations about differences in CPU architecture and such, went right above his head.)
So it was easy for Intel to push the MHz as the one true speed indicator. And for a while all they had to do was keep putting out CPUs with more and more MHz.
Except after a while it became a trap. Any new design _had_ to be higher MHz, or have Intel's own marketting working against it. All those many millions that went into telling people "buy a higher clocked CPU", now would basically tell them "don't buy the newest Intel CPU chip", if Intel made one with less MHz.
And now Intel finally _has_ to find a way out of the hole it dug itself into.
As for Cyrix (now VIA), it was never really a problem for Intel. Cyrix just fell behind performance-wise on its own. The last proper Cyrix versions were already falling beind in integer performance too, but it was their floating point performance that was abysmal. So what killed Cyrix was not as much Intel, as games going 3D: now everyone had benchmarks everywhere, clearly showing the Cyrix as barely crawling.
And Via's versions fell behind even more. They aren't just slower in MHz, they're also slower _per_ MHz. Other than being low power, they just suck.
And it's not that VIA really _wants_ to be the poor-man's niche, for Chinese families who can't afford an Intel or AMD. People find such niches to survive, but noone really wants to _stay_ in such a niche. Noone actually wants to sell their top CPU at $30 or less, instead of, say, the $600+ that an Athlon 64 FX sells for.
So if VIA could break out of that unprofitable niche, believe me, they would. The problem is simply that they can't.
A polar bear is a cartesian bear after a coordinate transform.
Basically, the Pentium M is a move back to a P3 type design philosophy, away from the 30-stage pipeline madness Intel's gotten themselves into with Prescott. I fail to see how going with a more intelligent design is going with a dumbed down processor.
I agree with you whole-heartedly. Although the only thing I'd add to what you've said is that they're going back to a chip design that they didn't actually design! If anyone recalls, the Pentium was basically ripped off from DEC. Sure, adding SSE and other "add-ons" was a way of extending the life of the base design until Intel could design its own chip from scratch: the Pentium IV.
Figures they'd go back to a design that was more efficient clock-for-clock than what they could come up with on their own.
And before anyone reads too much AMD kudos in this, AMD bought DEC engineers for chip design and traded flash tech for copper fabrication tech from Motorola to help them leapfrog from K6 (Intel-clone) to the K7.
Diplomacy is the art of saying, "Nice doggie!" until you can find a rock.