Intel: We Rushed The Dual-Core P4 To Market

Spinner

Intel engineer Jonathan Douglas has admitted that the dual-core P4 was rushed out the door because of competitive pressures from AMD.

Smithfield made it through testing and out the door in about nine months, which is remarkably quick by Intel standards. The need to get a dual-core CPU into the market as a response to AMD meant that Smithfield lacked features of the dual-core Opteron and Athlon 64 like independent memory buses for each core. In addition, the need to put two Pentium 4 cores on a single die led to additional signaling problems as the transistors were even closer together on the new dual-core CPUs.

Submitted by: profdlp

Source: Arstechnica

mmonnin

Well no ****. It sucks compared to the AMD dual core counterpart. The operton was built ground up to go dual-core from day one.

Lets see here...for one Intel core to send data to the other core it need to go across the FSB and to the Northbridge and back to the other core. AMD cores can talk directly to each other w/o using bandwidith across the FSB, or HTT link. The Intel solution is really 2 cores glued together and put on one chip. Thats basically all it is. It sucks balls. AMD Opertons also have another way to access and share shared variables to ensure mutual exclusion unlike the Intel processors.

If they do not fix the cross core communication, where the data uses the FSB, by the time the dual core Xeons come out later this year, it is going to scale horribally. All the CPUs use the FSB to talk to each other. Now that same bandwidth will be shared with twice the cores. The available bandwidth for each core will be cut in half.

drasnor

mmonnin wrote:

Lets see here...for one Intel core to send data to the other core it need to go across the FSB and to the Northbridge and back to the other core.
...
All the CPUs use the FSB to talk to each other. Now that same bandwidth will be shared with twice the cores. The available bandwidth for each core will be cut in half.

This isn't quite as bad as it sounds. A traditional SMP system has to go to the FSB for everything (including CPU-CPU transfers) so a dual processor dual core Xeon system couldn't do worse than a Xeon MP system with 4 discrete processors. In fact, Opteron is the only processor I'm aware of that lets you connect cores directly and the AMD64 architecture is the only one that can go to memory without going to the FSB first. It's really a testament to how well AMD designed AMD64 and how much we've embraced their vision as our standard in multiprocessor systems.

There's no excuse for signaling problems though. That's reason enough by itself to prevent these processors from selling at all.

-drasnor

Thrax

GOOD ****ING GAME NUBS.

Intel Corp
Failing to learn from past mistakes since 1968

mmonnin

It makes no sense to not allow CPUs or cores to talk to each other directly, esp after the Opteron has been out this long.

airbornflght

i think intel is falling behind. all there is to it.
unless they pull a rabit out of their hat as did amd, then they are gonna be hurting.

drasnor

mmonnin wrote:

It makes no sense to not allow CPUs or cores to talk to each other directly, esp after the Opteron has been out this long.

My point was that it's easy to forget that there isn't anything more direct than the front side bus in any non-AMD system. Also, Opteron has only been out for two years now and Intel didn't even perceive it as threatening until recently. Factor in that AMD has been cooking up Opteron for longer than two years before they released it and that it's design philosophy is radically different than anything anyone has ever done before and you get a tough cookie to emulate. This isn't the same as the K6 days where the Pentium III and Pentium II were natural evolutions of the Pentium and the K6, K6-II, K6-III, K7, Athlon could be brought to market in short order since they used the same fundamental technologies as Intel.

Think about it this way: AMD has been developing the base technologies that make Opteron possible long before it hit the market (e.g. HyperTransport) while Intel hasn't been developing anything new in the core logic department beyond support for the latest I/O. Two years ago AMD hits the market with all of these technologies in one cost-effective high-performance package and markets it well enough to have significant penetration into traditionally Intel sectors (enterprise servers and workstations) and Intel hasn't got anything up their sleeve. The best they can do is try to force a standards change that edges AMD out of the picture (BTX, DDR2) but this hasn't met with much enthusiasm since motherboard manufacturers hate LGA775, DDR2 was a performance flop compared to DDR when it first hit the market (first impressions count: see Rambus) and enthusiasts don't like BTX. They didn't expect Opteron and have been caught with their pants down. Their moves over the last year (canning Netburst, adapting their mobile technologies to their entire product line, processor numbering schemes to bunk the MHz myth, AMD64 emulation modes, and now two discrete processors in one package) are essentially damage control while they try to develop a real competitor.

In short, we're in a transitional period where Intel hasn't got anything good outside of the mobile segment; dual-core P4 is simply a way of getting money and looking good in the interim. My personal opinion is that AMD has a great product: I really like my dual Opteron workstation and have built a dual Opteron server and single Opteron workstation for my place of business. At the same time though, I want to be able to have the choice between multiple great products like I do now with graphics cards, motherboards, sound cards, network cards, memory, etc.

-drasnor

Shorty

Thrax wrote:

GOOD ****ING GAME NUBS.

Intel Corp
Failing to learn from past mistakes since 1968

How long have been we all saying that..

airbornflght wrote:

i think intel is falling behind. all there is to it.
unless they pull a rabit out of their hat as did amd, then they are gonna be hurting.

Ya damn right.

muddocktor

I agree with the points brought up so far, but I think starting next year that you will start seeing the results from Intel's wakeup from La-La Land when they finally realized that AMD had the superior product. Intel does have future stuff coming down the pipeline that sounds pretty darn interesting. Dual core products developed from the P-M architecture with revised SSE2/SSE3, 64 bit capability and HT (dubious benefit IMO because of the much shorter pipeline).

Gobbles

airbornflght wrote:

i think intel is falling behind. all there is to it.
unless they pull a rabit out of their hat as did amd, then they are gonna be hurting.

they are about to.. they are dropping the p4 arch. for a "new" one, which I believe will be the dothan based chip, which we all know intel actually did right. Its efficient, fast, and runs head to head with AMD... Next year should be pretty interesting cpu wise...

muddocktor

OMFG, Gobbles, I can't believe you put Smilin' Bob up as your avatar.

Leonardo

The Register published an article by Ashlee Vance in the Mountain View, summarizing Intel's blunders of the last couple years. Extract:

Over the past two to three years, we've watched as Intel scoffed at the idea of x86-64-bit chips, preferring instead to insist that Itanium wasn't a huge failure. Intel also brushed off the multicore hype from the likes of IBM and Sun Microsystems. All the while, Intel plodded along, upping the GHz of its products and wrapping them with more and more cache. Then, one day, Intel woke up with a fleet of tubby, super-heated antiques on its hands and had to read about AMD signing up IBM, HP and Sun as server customers...The big dog slipped up and had to change course - quick. Problem is the processor game requires long turnaround times.

Intel has awakened and is marshalling their considerable forces. AMD is enjoying their respite and technology lead.

Gobbles

From neowin about intels idf in san fransico

Looks like a hell of a fight coming in the future of processors... I cant wait..

Much has been said about Intel's plans to shift its design strategy on the desktop platform over the last few weeks, but as the Intel Developer Forum in San Francisco that kicked off today the attendees finally got a glimpse at what it's all about. Three new cores have been introduced: The Conroe for desktops, the Merom for mobile platforms, and the Woodcrest for servers. The new processors will combine the best of both worlds from Intel's Pentium 4 desktop line and Pentium M processor for notebooks, sporting both a high front side bus speed while minimizing power consumption.

While the Prescott core P4's utilized a 31+ stage pipeline compared to the Athlon 64's 12, the new cores appear to have a 14 stage arrangement. This will allow for lower power requirements and more processing power per clock. Expect clock speeds upwards of 3GHz on the 65nm cores. While current models operate upwards of 4GHz, Intel claims five times the processing power per watt, which would be startling change in efficiency. Under the new architecture L2 cache is shared between cores. Intel states that the Conroe will have multiple versions with different L2 cache sizes while the Merom cores will feature only one model. This is presumably to allow room for "Extreme Edition" models of the Conroe with more L2 cache.

Intel's design shifts are no doubt exciting. With these steps we may finally see the abolishment of the MHz Myth, which has plagued the likes of AMD into using silly naming schemes in order to prove their processor's merit. There will be more to come as Fall IDF 2005 continues.

mmonnin

Yeah at IDF this week they will unvail some more info on their next architechure beyond Netburst.