Intel Stepping Up to Combat AMD's 4x4

Grooves writes "Intel has said that the company is stepping up the pace of its Core 2 architecture rollout to compete with AMD's 4x4. Two "quad-core" parts originally slated for release in the first half of 2007, Kentsfield for the desktop and Clovertown for servers, will make their debut as early as the end of this year. The Ars article warns that per-core bandwidth problems could end up giving a performance advantage to AMD's 4x4 approach."

Comments

Haste

[MECC]

Make waste...

And so it begins

[Linkiroth]

The great hardware war heats up once again. Right now, the biggest advantage Intel has is that their chips are scheduled for an earlier release. If they wait on the Core 2s, they're screwed. They need to get the Core 2 Duos out before AMD gets out their 4x4s so that people have less of a reason to upgrade when AMD releases their chips.

Re:And so it begins

[vancondo]

They need to get the Core 2 Duos out before AMD gets out their 4x4s so that people have less of a reason to upgrade when AMD releases their chips.

Do most chip sales happen at the release date, or do most people wait for the competitors product to come out spurring price drops to compete? I know I seldom buy anything at the alpha-expensive stage, usually preferring to wait a few months for the inevitable price drop.

Gotta love CPU wars

[ntxb229]

Consumers really come out on top. Better processors at cheaper prices.

Re:Gotta love CPU wars

[AcidLacedPenguiN]

don't you realize that the high performance wars fuel the low performance price drop?
I mean, I don't call myself an expert or anything like that, but I think one of the sole purposes of pushing a high performance part onto the market is to move med/low performance parts into consumer's PCs.
I mean, most of their revenue comes from selling consumer parts, not high performance ones.

Who is paying?

[Ahnteis]

I can't even afford a high-clocked AMD X2. How am I (as a fairly high-spending gamer who builds his own computers) supposed to afford TWO of them? And if *I* can't, who exactly are they targetting with this 4 core nonsense?

I may still buy AMD on principal (yes, some of us do that still) but I really think Intel has AMD beat for the next year or two.

Re:Who is paying?

[Anonymous Coward]

This is why you invest in a water cooling setup. I took a $150 processor (opteron 144) and OC'd it stable to 2.9ghz. This was when I could have purchased a FX-57 for ~$1000. Cheapest Conroe CPU with 4mb cache is around $300, which can be OC'd to 3.8ghz+ on water. You can take an opteron 165 ($300) right now and OC it to 2.6+ on air, or 2.8+ on water. This is done with a water cooling system that costs $200 up front, then $50 each processor upgrade unless you have a universal mount for the CPU block. This works alot better for the rare person who has more brains than money. The only people who are buying the FX or Extreme processors are the ones with more money than brains or hard core overclockers who want the best toys to play with.

4x4 is an inaccurate name...

[CyberBill]

AMD apparantly cannot multiply. 4x4 = 16. The 4x4 architecture is two dual-core CPUs on a single motherboard (2x2=4 cores). This is pretty damn annoying and I wish they would rename it to something a little more accurate to whats going on... If you have a Dual 7950's (which are each just two 7900's), you wouldnt call it 4x4.

You must first ask the right questions

[Ctrl Alt De1337]

Are you a gamer? Are you someone who does intense multimedia work? If not, then a single-core chip is fine, much less a 4-core chip. For the vast majority of home and business desktops, chips that are considered old right now offer plenty of computing power. The Apartment Hunters across the street from the UF campus still use G3 iMacs at the front desk. These 4-core beasts will be niche things for a while, I think, unless a lot of weasely salesmen can (continue to?) convince people to buy more computer than they need.

Re:You must first ask the right questions

[masklinn]

Dude, even a dual core chip is pretty useless for a gamer already, there's almost no game using the second core right now (the only use it has is that it runs all your malware and the Steam client [well one could say that Steam is a malware in its own right though] so that the actual running of the games can be done on their own core).

Games making use of 4 cores? You've got the time to see it coming.

Re:You must first ask the right questions

[skiflyer]

actually for the most part right now dual cores don't speed up any single application. (Yes there are plenty of rendering/scientific applications which it does speed up, but I said for the most part.) Where it really shines is in speeding up the user experience.

Run a compile, a virus scan, and still have your email app or browser pop up immediately when you click it.

The neat thing about the way multi-core programming works is that for alot of things once they make it support two, it'll automatically support 4.... so once we get app support for multi-threads as a common commodity you'll be able to scale performance alot easier (easier, not cheaper)

Personally tho, I'd trade all my single core systems for dual cores systems that ran two cores each at 50% the speed of the single cores, cause like you said, most of us have more computer than we need for most applications, but most of us also multi-task quite a bit, and it's the responsiveness and multi-tasking of my dual core machine that makes me so happy, not the raw speed (tho that's nice too :))

Re:Oooh.... core wars

[vhogemann]

And also consider that many softwares over there are not prepared to take advantage of these extra core... Sure, you'll be able to run more applications at the same time without degradation.

This makes me wonder, will the developers adapt to this new reality. I mean, Intel and AMD can't give us more performance by raising the clock of their processors... so they started to put more cores on them. At one point developers will have to paralelize their code to be able to gain performance.

Bandwidth will be a problem.

[default luser]

Intel knows this very well, they've been having trouble with bandwidth for years while stuck at 800 MHz FSB. The only dual-core Pentium 4 processors to show efficient use of the second core are the EE-series, with 1066 MHz bus.

Even if Intel can successfuly crank the FSB up to 1333 MHz bus, that's still significantly less than they need to feed twice as many processors as Conroe. If this were AMD, they'd just add more memory controllers and more HT links...but for Intel this is not an option.

Intel does offer a Dual-Independent Bus architecture, but this is designed for Woodcrest, and is extremely expensive to implement. DIP does allow Woodcreast to scale effortlessly to 4 cores, and that is why we've seen Intel encourage reviews of their 4-core (2 processor) Woodcrest platforms. Unfortunately, even this DIB architecture will not scale well into 8 cores (4 cores per bus), and Intel's cheaper-to-implement quad-core processors will really feel the squeeze.

but...

[mseidl]

these new chips are still based on the AGTL+ bus, which doesn't compete with HyperTransport. While it'll add perfomance, can you imagine a 4way 4x4 setup with Intel? You will have 16cores sharing one bus w/ half the bandwith of HT, NOT PRETTY!

Re:it's not about gamers

[chez69]

hell yeah it makes sense under the desktop. windows desktop for testing , linux desktop for development, linux dev server, linux DB server all on the same desktop.

it'll be great for developers

4x4 shouldn't worry intel

[buddyglass]

The Ars article warns that per-core bandwidth problems could end up giving a performance advantage to AMD's 4x4 approach.

I see two problems with this. First, most cpu-intensive tasks are single-threaded, and Conroe beats AMD on those. Second, even if it turns out that two Athlon64 X2s scale better than a single quad-core Conroe, the Conroe is a single-chip solution in a single-socket motherboard. AMD will have to price its X2s at less than half the cost of a quad-core Conroe. "Less than half" since they'll also need to absorb the extra cost of the dual-socket motherboard 4x4 requires. I suspect they won't be able to achieve that price point. So, given an AMD 4x4 system and a comparably-performing Intel quad-core Conroe system, the AMD system will cost more and be less attractive to consumers.

Re:4x4 shouldn't worry intel

[swillden]

And what's to stop Intel from releasing its own dual-cpu version of Conroe, along with a dual-cpu chipset?

As a dozen posts and the article and article summary all mention: Memory bandwidth. Intel's going to have a very difficult time getting data to and from four cores fast enough for the cores to be useful. AMD has a big edge there.

Imho, things look pretty bad for AMD in the short term.

IMHO, things look pretty good for the consumer in the short term. And in the long term, too, as long as AMD and Intel continue competing with each other.

4 Intel cores != 4 AMD cores

[Nom du Keyboard]

I'm not sure that a 4-core Intel processor, at least in the first iteration, is going to be all that wonderful. Intel still has the front-side bus problem that Dell used as justification to go to Opterons for 4way servers. Additionally, the Kentsfield approach of simply joining 2 Conroe dies in the same package may not prove much better than when they created the first dual core Pentiums by joining 2 Netburst cores in a single package. It is inherently not a well integrated design.

That being said, undoubtedly Kentsfield will be at least incrementally faster than Conroe, so that helps with bragging rights. And small, cache-based code (think Cell processor SPEs) could run well on it. But unless it is priced exceptionally close to Conroe prices, would not be my first choice.

AMD is likely to do 4-cores the right way the first time around, rather than ship a Marketing Solution.

Re:Latencies and more

[Khyber]

Available bandwidth translates into less latency, FYI, always has for networking, data storage/retrieval, and graphics applications like 3d Rendering and generation. So, AMD, with it's superior bandwidth, may not need to prefetch simply because it can have that much more data crammed down it's pipeline on the fly, whereas Intel has to pre-cache it. HyperTransport doesn't go over the FSB as far as I'm aware. (But I'm not that aware, so please correct me)