Intel quad core Kentsfield CPU's

[jamesgoddard]

With the Intel quad core Kentsfield CPU more of less complete - see running pics (http://www.theinquirer.net/?article=32154) - do people think that Intel will pull the release date forward to spoil any 4x4 plan from AMD? After all the Kentsfield is only 2 x Conroe on a single 1066mhz FSB. I am thinking a release this fall is very possible.

 

[iterations]

I don't think the release date will be pulled in from Q1 2007, but this speaks to Intel being able to have a very strong launch. They already have complete, working, single-socket, consumer Quad Core systems working in house. Now they have months to tweak it and continue to improve it's performance/watt ratio.

The release date likely won't change becuase Core 2 is going to be so dominant, Intel would not want to compete with themselves. The only way I see Kentsfield being pulled into 2006 is if AMD pulls a huge rabbit out of their hat and releases something that realistically competes with Conroe. (And no, based of what I've seen, 4x4 aint it, and K8L isn't coming until the 65nm switch for AMD in Q1 2007 anyways).

Cheers!

 

[ltcommander_data]

Now they have months to tweak it and continue to improve it's performance/watt ratio.

I've said it many times, and I'll say it again. Intel must take the extra time to get Kentsfield and Cloverton working on a 1333MHz FSB. If they release with a 1066MHz FSB, the results will be disappointing (ie not decent scaling), and they can't afford having that type of stigma (ie. Smithfield like) associated with their first quad core product and probably the first quad core product in general.

K8L isn't coming until the 65nm switch for AMD in Q1 2007 anyways

The first 65nm Rev G products from AMD are not going to be K8L. The earliest we'll see K8L is probably H2 2007.

 

[iterations]

I've said it many times, and I'll say it again. Intel must take the extra time to get Kentsfield and Cloverton working on a 1333MHz FSB. If they release with a 1066MHz FSB, the results will be disappointing (ie not decent scaling), and they can't afford having that type of stigma (ie. Smithfield like) associated with their first quad core product and probably the first quad core product in general.

Yeah I agree. I've been hearing rumors about "Bad Axe 2" which I'm guessing has to do with what you are referring to. A Chipset refresh for 1333FSB to coincide with Kentsfield would be ideal.

The first 65nm Rev G products from AMD are not going to be K8L. The earliest we'll see K8L is probably H2 2007.

Ah, I was being optimistic, heh.

 

[zarooch]

Agree... i already tried to explain this same whole thing to 'linux0' which he didn't accept but anywayz... people have there own views... apart from this... i think that AMD should have accepted Intel superiority in performace/watt as well as price... and instead of releasing this 4X4 solution... they should have gone back to the table and come up with a new uArch. and then implement it on 65 nm process tech. so that they can equally compete with Intel's new uArch CPU(s)... AMD is lagging... 65nm for next year... 45 nm don't know?... well Intel is very well focused and they have a greater lead the competition won't be balanced from AMD side... but still it would be intresting to see how AMD comes up again.

 

[KarmaKiller]

Hey, I agree that 4x4 seems to be a bad answer. BUT do we have some benchmarks? (ej: 2x Opteron 285 in a nForce 4 Pro mobo). We don´t know how much AMD will drop the prices either.
So I think is hard to critisize 4x4 -yet- without a decent backup.
Cheers!

 

[ltcommander_data]

In regards to 4x4, what I don't a lot of people in general (not you specifically) don't appreciate is the additional costs associated with that approach. 4x4 is a easy route for AMD since it doesn't cost them anything, they just continue to sell their current processors, just twice as many, which is a good thing. However, how are motherboard makers going to take it? You look at an enthousiast board and those are already very crowded. It's going to be an interesting design challenge to put a 2nd socket and more memory slots onto a standard ATX board. The movement now is also toward smaller boards and smaller computers which doesn't help matters. The production cuts for such a board will also be very high due to the extra traces and probably the necessity to add extra layers for stability.

A lot of people point to 4x4 as a way for people to just buy the board and 1 processor now and add another later. The thing is that AMD seems to indicate 4x4 is only for the FX series, while people generally point to the ability to put a 35W X2 3800+ in it. Even if AMD opens the platform up it's unclear whether 512kB cache processors like the X2 3800+ will work. I've heard reports, which I don't know if true or not, that the 512kB models not only have their cache size reduced to save transistors, but also only have 1HT link. This means they can't operate in a 4x4 setup. Only the 1MB models retain 3 HT links like the Opterons. This would of course mean low cost, low power 35W X2 3800+ 4x4 setups aren't likely.

I'd be curious to see how much extra capacity AMD has to supply a 4x4 setup. They are already selling processors as fast as they can make them, and now they plan to be selling two of them at once. Having a single high-performing processor is much more worthwhile in terms of power, yields, and cost than selling 2 processors.

 

[ltcommander_data]

b) Revise the server chipset (as AMD is doing) to support a dual socket dual FSB consumer level version, and build their own "4x4" ... very unlikely, but not outside the realm of possibility. Why, well because Conroe/Woodcrest are identical -- the chipset logisitics are already worked out.

Intel shouldn't be doing this officially for obvious reasons, but I'm hoping that Apple releases their G5 tower replacement on Woodcrest. So far XServe is confirmed on Woodcrest, and the iMac on Conroe. Apple seems to want to go Conroe on the Tower, but that would put it far too close to the iMac. They can't possibly release all their Tower models on Extreme Editions. It'd also be a downgrade in a sense going from 2x dual core G5s to a single dual core Conroe. With Apple's able to run Windows now, a Woodcrest Tower may be an interesting option for gamers. It'll no doubt look nice as well. Not flashy with UV effects, but very stylish.

 

[jamesgoddard]

To stay focused on the Kentfield topic, Ltcommander_data is right on the money on this one.... for dual core Conroe 1067 is showing enough BW to feed two cores. Part of this is due to the large cache, great prefetch logic, and shared memory subsystem and the need for higher BW subsides. But double that and the BW becomes a limiter again, 1333 is not "non-doable" as we see with the Bently platform, but apparently on the consumer chipset it becomes a limiter.

Bently has been confirmed to only work at 1066fsb when using quad core CPU's... The problem I think is that the fsb hits a hard limit somewhere just over 1066 when you have 2 electrical loads running on it, thus Kentfield will be 1066mhz... Unless Intel can make a true 4 core single chip CPU.

 

[zarooch]

To stay focused on the Kentfield topic, Ltcommander_data is right on the money on this one.... for dual core Conroe 1067 is showing enough BW to feed two cores. Part of this is due to the large cache, great prefetch logic, and shared memory subsystem and the need for higher BW subsides. But double that and the BW becomes a limiter again, 1333 is not "non-doable" as we see with the Bently platform, but apparently on the consumer chipset it becomes a limiter.

Bently has been confirmed to only work at 1066fsb when using quad core CPU's... The problem I think is that the fsb hits a hard limit somewhere just over 1066 when you have 2 electrical loads running on it, thus Kentfield will be 1066mhz... Unless Intel can make a true 4 core single chip CPU.

any links?

 

[ltcommander_data]

Bently has been confirmed to only work at 1066fsb when using quad core CPU's... The problem I think is that the fsb hits a hard limit somewhere just over 1066 when you have 2 electrical loads running on it, thus Kentfield will be 1066mhz... Unless Intel can make a true 4 core single chip CPU.

Yes, the three electrical loads is the limiting factor.

http://www.xtremesystems.org/forums/showthread.php?t=101798

However, my feeling is that with dual cores running at 1840MHz FSBs you'd think there is enough signal strength to get 1333Mhz FSBs for quad cores. They might need to revise the layout a little bit, give a bit more voltage for the northbridge, and recommend higher phase voltage regulators, but that shouldn't be a problem for enthousiast or server boards. I can see Intel being reluctant to make Cloverton 1333MHz, since in the server market 24 hour 100% stability is critical. However, I'm sure Kentsfield can get away with it. That's not to say Kentsfield can be less than 100% reliable, it's just that the defect rate can be slightly higher in the same ratio as consumer products usually are to server products.

 

[ltcommander_data]

Evidently the yield on the Broadwater is not coming in good enough to support running at 1333 MHz in volume, so they will need to run it at 1067, this is my guess.

The limited ability for the 965 chipsets to run at 1333MHz FSBs is simply because they were never officially designed to support such speeds. This problem is compounded by the fact that mainstream products are not required to have the overdesign necessary in the other components to guarantee faster FSB operation. This all goes back to the blatant lack of a enthousiast i975X replacement. Argh. It's going to be an embaressment for Intel if they are forced to launch 1333MHz processors on ATI RD600 boards only.

 

[DavidC1]

c) Pairing up with ATI (they have a 1333 FSB solution ready today), putting together a "Kentsfield" like enthusiast solution, with 4 cores and pushing that as numero UNO.

2 problems there.

1. It will result in ATI having the highest performance part, not Intel
2. Intel has more conservative methods to validation so they won't accept ATI's if it doesn't meet their standards.

 

[iterations]

Intel is in the middle of moving chipset production into their 90nm 300mm fabs (which means the cost for 90nm has been fully depreciated) so stock speeds for stable chipsets should improve, from their current 130nm counterparts. This should enable chipset revisions that are electrically stable at higher FSB speeds.

 

[zarooch]

which means that it can be overclocked by increasing the FSB?

 

[9-inch]

With the Intel quad core Kentsfield CPU more of less complete - see running pics (http://www.theinquirer.net/?article=32154) - do people think that Intel will pull the release date forward to spoil any 4x4 plan from AMD? After all the Kentsfield is only 2 x Conroe on a single 1066mhz FSB. I am thinking a release this fall is very possible.

Once again, intel repeating the same nasty hack for Kentsfield (Pentium-D anyone?). :roll:

 

[iterations]

Not a nasty hack at all. It is a smart way to do a single socket quad core chip. Eventually there will be a quad core solution all on one die. This approach keeps costs in line and yields up.

You should be familiar with this technique, 4x4 is basically the same philosophy except AMD is using two sockets, which unfortuantely makes mobo production much harder and more expensive.

 

[9-inch]

Not a nasty hack at all. It is a smart way to do a single socket quad core chip. Eventually there will be a quad core solution all on one die. This approach keeps costs in line and yields up.

You should be familiar with this technique, 4x4 is basically the same philosophy except AMD is using two sockets, which unfortuantely makes mobo production much harder and more expensive.

Are you forgetting the possibility of NUMA here? This will improve bandwidth even more. Anyhow, It's far away from intel's nasty hack.

 

[iterations]

Not a nasty hack at all. It is a smart way to do a single socket quad core chip. Eventually there will be a quad core solution all on one die. This approach keeps costs in line and yields up.

You should be familiar with this technique, 4x4 is basically the same philosophy except AMD is using two sockets, which unfortuantely makes mobo production much harder and more expensive.

Are you forgetting the possibility of NUMA here? This will improve bandwidth even more. Anyhow, It's far away from intel's nasty hack.
Not forgetting about it in the least. That is why AMD can offer a two socket solution. But it is just that, a two socket solution. Kentsfield is making quad core single socket a reality, regardless of your tainted opinions.

Anyways this thread is about Kentsfield, so don't bother crapping in it anymore. It is clear that Intel has killed your puppy or something, so you believe everything they do is bad.

 

[9-inch]

so you believe everything they do is bad.

Is not that everything they do is bad, it's the fact that they should have done better by being a "BIGGER" company than AMD.

Anyways this thread is about Kentsfield, so don't bother crapping in it anymore

Just the way jumpingjack and his Horde did in the K8L thread for talking BS about it? Even if K8L is superior to conrunt these guys will always believe conrunt is better no matter what.

 

[iterations]

Hmm I doubt JumpingJack et. al. would maintain the dominance of Conroe, if real benchmark data is made available for K8L; he seems much too intelligent, well informed and reasonable for that.

The fact that K8L isn't going to be available until 2H 2007, and only exists in PowerPoint for now, makes it kind of silly to compare, as there is nothing to compare but promises made on PowerPoint slides.

Kentsfield on the other hand does exist in real silicon and has for a few months. There will be one die quad core solutions in the near future, I'd imagine. Intel's focus for now is getting Core 2 out on time, then I'm hoping we will see more data start to be made available on Kentsfield/Clovertown/etc... and then single die quadcore, particularly with Penryn 45nm.

 

[ltcommander_data]

I'd thought I'd put forward an interesting scenario for your opinions. Seeing that the 2MB Conroes/Allendales appear to scale slighty better than the 4MB Conroes, there might be a funny situation where a 2x2MB quad core could reach a 1333MHz FSB while a 2x4MB quad core is limited to a 1066MHz FSB. In such a situation, which do you think is the better option?

My opinion would be to go with the 2x2MB 1333MHz FSB route. I think the extra memory bandwidth would be critical in filling a quad core. At the same time, using 2MB dies will increase yields, and more importantly decrease power consumption and temperature. That could possibly allow one clock speed bump over a 2x4MB solution.

 

[apache_lives]

9-inch you are by far the greatest moron in the world, take a look at BOTH sides of the story - AMD FANBOY "K8L WILL KILL IT" and then INTEL FANBOY "IT WILL SMASH K8L" and behold there both the same fight, when you be an AMD fanboy, atack Intel fanboys you atack yourself cause the people you atack are your own - FANBOYS

God dam someone ban this guys ass this is total BS he doesnt add anything good just total crud.

so you believe everything they do is bad.

Is not that everything they do is bad, it's the fact that they should have done better by being a "BIGGER" company than AMD.

Anyways this thread is about Kentsfield, so don't bother crapping in it anymore

Just the way jumpingjack and his Horde did in the K8L thread for talking BS about it? Even if K8L is superior to conrunt these guys will always believe conrunt is better no matter what.