Cedar Mill

[ltcommander_data]

"The fact 3.8 GHz is not being listed, or at least not as a 85W part should speak volumes about the shape it is in."

Actually, as of the latest Intel Desktop Roadmap that I read, the 3.8GHz 671 Cedar Mill is listed as targeted for a 2nd half 2006 launch. I read that less than a month ago. As well, the 671 was always targeted for a higher TDP just as the 570 and the 670 were targeted in for the higher 115W bracket. I also have reason to believe that the TDP for the 671 will be lower than that of the 670. Why is that? Just look at the 661. Like the 670, the 660 also fit into the higher TDP of 115W. But now, the Cedar Mill implementation has the 661 fit into the lower heat bracket of 86W just like the 651, the 650 and lower. This clearly indicates that Intel is making headway in their heat problems if they are able to fit a 3.6 GHz processor in with 3.4GHz and lower processors. It is interesting to notet that when the 3.4 GHz 550 was launched, it too fit into the higher 115W TDP. But as the 90nm process improved, the latest 550's were able to fit into the lower TDP of 84W. Clearly, the TDP of the 671 will decrease not increase.

As well, it is interesting to note that Intel is taking a more competitive approach to measuring TDP. The older processors' TDP were measured for only the processor itself yieldind the brackets of 84W and 115W. The current TDP measurements are platform TDPs which now also include the system power requirements. Intel is certainly making headway if the processor and system only generate as much heat (86W) as the itself did in the previous generation (84W). These values are also now more competitive with AMD, since AMD constantly claims its TDP readings include the integrated memory controller. Now Intel's TDP readings include the entire chipset.

 

[ltcommander_data]

On closer reflection, it appears my earlier comments on the TDP limits were low balled. In fact not only can the processor and motherboard fit into the previous TDP of just the processor, a 3.6GHz 661 processor and motherboard (86W) can fit into the TDP of just a 3.4GHz 650 processor before (84W). In otherwords, examining the FMB TDPs, Cedar Mill processors gain not only a 10% reduction in heat dissipation but also a 200MHz increase in clock speed. Subtracting the motherboard from the 86W TDP, a 3.6GHz Cedar Mill now has a TDP in the mid 70W region down from 92.7W. This is more than a 20% reduction. This is why the 3.4 GHz 950 and the 3.46GHz 955 can fit into the TDP of the 840.

If a 3.6GHz 661 processor + motherboard can fit into the thermal profile of a 3.4GHz 650 processor alone before, it isn't unreasonable to think that a 3.8GHz processor 671 + motherboard will fit into the profile of just a 3.6GHz 660 processor. Therefore it is perfectly reasonable for the existance and thriving of a 3.8GHz Cedar Mill.

In addition, it isn't unreasonable for Intel to introduce a 4 GHz Cedar Mill + motherboard into the TDP of a 3.8GHz 670. As well, the TDP of the 3.8GHz 670 processor alone was 115W, now with the dual cores the maximum design TDP of motherboards is 130W. Therefore, Intel still has additional room for speed ramps.

 

[endyen]

Now Intel's TDP readings include the entire chipset

Got a link?

 

[P4Man]

WTF are you smoking ?

1) What makes you think Intel has revised its TDP definition ? I see no evidence of this whatsoever. Intel defines TDP as the maximum sustained power consumption system integrators should target when designing the cooling. They state the cooling solution should be able to cool the CPU to below maximum Tcase when consuming TDP. They also state the cpu can indeed consume more, and if this happen Thermal Monitor is to be relied upon to save the CPU from melting.

IOW, intels TDP is below what their cpu's can consume using real world applications, as noted in their own spec sheets.

2) the notion intel includes power consumption of chipset, let alone motherboard in its TDP is not only completely unsubstantiated, it is also totally absurd. Different chispets consume different ammounts of power (nForce4 for instance is power hog), its impossible to take that into account in the spec sheets of the cpu. Its also totally useless. And of course its totally untrue.

3) Thanks for pointing out intel has released 84W prescotts, I forgot about those. So we have up to 3.4 GHz 84W prescott versus up to 3.6 GHz 86W Cedar Mills. Very, very unimpressive.

4) You keep dreaming about 4(+) GHz P4s, where intels CEO has publically said there will not be such a chip, period. Cedar Mill clearly shows us why. Yes, they could up the TDP again, TDP is not the problem though, *actual* peak power consumption, thermal density and electromigtation are. Have you not noticed Vcore has remained basically unchanged in spite of the shrink ?

Wake me up when Cedar Mill reaches 3.8 GHz, and Ill be you a beer if they ship more than 3 samples of a 4 GHz product.

5) the notion that any of this would somehow make Intel competitive in power consumption of power/W is absurd:

<A HREF="http://www.xbitlabs.com/articles/cpu/display/pentiumd-820_3.html" target="_new">http://www.xbitlabs.com/articles/cpu/display/pentiumd-820_3.html</A>
or here
<A HREF="http://www.lostcircuits.com/cpu/p4_820/11.shtml" target="_new">http://www.lostcircuits.com/cpu/p4_820/11.shtml</A>

Please tell, what does a measily 10% do to change those charts ? Not a damn thing.

In conclusion, its quite clear Cedar Mill will at best only provide a very modest reduction in power consumption compared to Prescott, while being much smaller, therefore, much hotter. Prescott suffered badly from thermal density and power delivery issues, and Cedar Mill will suffer even worse. Its obvious to me intel is using Cedar Mill only to reduce cost per die, and as a test vehicle for 65nm, but is not devoting any serious resources to make this a competitive part. Instead its betting big on Merom/Conroe, hopefully for good reasons.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[P4Man]

>I still don't get how TbredA worked compared to TbredB.
>You get like super 1700+ TbredAs(JIHUB) vs. the disasterous
>2200+ TbredA.

AFAIR, TbredA was wiredelay limited, giving it a brickwall clock limit of just over 2 GHz, hardly better than even Palomino. AMD solved that by adding a few layers to the design of Tbred-B.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[ltcommander_data]

I'm not smoking anything. In fact I don't smoke.

It isn't that Intel has revised its TDP definitions, its that I believe there are 2 definitions. One is the basic TDP definition that only involves the processor. The second is the FMB (Flexible Motherboard) TDP definition that I believe is used in the corporate market with the Stable Image Platform Program (SIPP). This definition indicates to businesses the power usuage they should expect from the processor and the average chipset. No doubt the chipset is one of their own, in this case probably a 915P or a 945P. I would think its a P model instead of a G, since the G's would consume more power if the integrated graphics is activated.

http://www.tomshardware.com/cpu/20051007/a_sneak_peak_at_intels_65nm_pentium_4-02.html

"While the current mainstream FMB (Flexible Motherboard - this is an Intel specification for processor and system power requirements) allows for a TDP (Thermal Design Power or a maximum heat dissipation) of 95 Watts, the Cedar Mill core is specified for up to 86 W now. As far as we know, this includes Pentium 4 631 to 661 (3.0 to 3.6 GHz), but it might easily be the case for faster models as well."

If you read the article carefully, it is very clear that the 86W TDP is a FMB definition that includes the system power requirements. That is why they are comparing the 86W Cedar Mill FMB TDP with the Prescott 95W FMB TDP. If they were comparing the 86W Cedar Mill FMB TDP with just the Precott processor TDP they would have said 84W. Once again, notice that the Cedar Mill 3.6GHz 661 + system power requirements generates the same amount of heat as a 3.4GHz 650 processor alone would have.

How should I read the line "Flexible Motherboard - this is an Intel specification for processor and system power requirements"?

"In conclusion, its quite clear Cedar Mill will at best only provide a very modest reduction in power consumption compared to Prescott, while being much smaller, therefore, much hotter."

Either way, with or without the motherboard, that modest power reduction you refer to allows not only Intel to reduce power comsumption by 10% but also increase clock speed by 200MHz.

And yes, I agree that Cedar Mill is only a test platform for the 65nm process. That is why Intel currently has no plans to release the 6x3 VT enabled processors instead relying on the Prescott 6x2 to introduce VT to consumers, which don't really have that big a use for it right now. The Cedar Mill is to the Pentium 4 what the Tualatin was to the Pentium III. The last of a generation. Conroe will redefine the state of the processor market with its 4MB or 8MB shared L2 cache, L1 cache interlinks, micro and macro fusions, additional and redesigned ALUs and FPUs, full speedstep implementation, and option for HT support.

<P ID="edit"><FONT SIZE=-1><EM>Edited by ltcommander_data on 10/09/05 02:44 PM.</EM></FONT></P>

 

[P4Man]

>How should I read the line "Flexible Motherboard - this is >an Intel specification for processor and system power
>requirements"?

Simply as its written. The allowed TDP referred to clearly relates to the CPU TDP as far as I can read. But even if Im wrong, a few watts really dont change the argument.

>Either way, with or without the motherboard, that modest
>power reduction you refer to allows not only Intel to
>reduce power comsumption by 10% but also increase clock
>speed by 200MHz.

Ahem.. no. TDP goes *up* from 84 to 86W (2%) for only a 5% increase in clock. Wow. Basically, same power consumption per GHz, much smaller core, therefore, significantly higher thermal density, as I said all along.

>The Cedar Mill is to the Pentium 4 what the Tualatin was to
> the Pentium III

*Absolutely* not. Tualatin allowed P3 to scale from 1 GHz (pretty much no OC headroom) to 1.4 GHz (1.6 to 1.7 GHz overclocks being the norm), a non trivial *40%* clock speed increase with the same power consumption.

>Conroe will redefine the state of the processor market with
> its 4MB or 8MB shared L2 cache, L1 cache interlinks, micro
> and macro fusions, additional and redesigned ALUs and
>FPUs, full speedstep implementation, and option for HT
>support.

That remains to be seen. My guess ? Intel is overreacting, and going from too high clockspeed and power density, towards too low clock (and performance) with ample power density headroom that is not needed on desktops. So conroe may indeed offer superior performance/w, much like Dothan today, but I doubt it will clock high enough to dethrone K8 performance wise. 100W actual power consumption is manageable for high performance desktops, especially (large die) dual core CPUs. Its nice to be at almost half that much, but not at the expense of performance. Its not like ppl are buying Dothans or Turions for desktops these days.

Come 2007, K8 may or may not face thermal issues like Prescott/Cedar Mill and intels low clock, low power approach might pay off, but I doubt Conroe will be the performance champ in 2006. If that leapfrog is to happen again, IMHO it will take an integrated MC, which wont happen until the end of 2007, early 2008. But feel free to bookmark this and laugh at me in a year or two

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[endyen]

Very humourous. TDP and FMB are two totally different animals, you can niether breed nor interchange them.
FYI TDP is an engineering specification, following the NEMA standard ( continous duty @ 66% of full load).
FMB is a spindoctors sales tool, based on M$ office use.
The two are not interchangeable.

 

[endyen]

Very humourous. TDP and FMB are two totally different animals, you can niether breed nor interchange them.
FYI TDP is an engineering specification, following the NEMA standard ( continous duty @ 66% of full load).
FMB is a spindoctors sales tool, based on M$ office use.
The two are not interchangeable.
BTW, the authors of this review made no mention of FMB.

 

[V8VENOM]

I believe it is called KFWRD internally or "Keep Funding While Research and Development".

Dual core, 65nm, these are just marketing tools in the hope they can overcome physical problems that they know they should be able to solve, but can't as of yet.

Speculation is that those that might have been able to solve these problems have moved on and/or retired -- or worse yet, realized the solution is beyond their capabilities and/or current materials.

The wall has been hit, a tad sooner than many expected (myself included). But that is always the case real solutions to the real world.

Imbedded systems is still viable and may keep the tech world going a little longer. But if real solutions aren't found REAL sooner, tech industry will be in even more of a rapid decline.

 

[endyen]

Oh be serious. Intel just took a bad direction with scotty. They could have saved themselves a whole lot of grief by doing the die shrink on northwood. Then they could give it FD SOI, and a 266 fsb. The 90 nano woody would probably get them to 5ghz. On 65 nanos, they could be looking @ more. Instead they just paniced.

 

[P4Man]

>I believe it is called KFWRD internally or "Keep Funding
>While Research and Development".

LOL.. as if youd know any "internal" acronyms. And no, intel engineers would not use such an idiotic one: obviously *any* R&D has to be funded.

>Speculation is that those that might have been able to
>solve these problems have moved on and/or retired

You mean *your* clueless speculation.

>Imbedded systems is still viable and may keep the tech
>world going a little longer

Now if only you knew how to spell that, some might perhaps believe youd have the faintest idea what that market is about.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[V8VENOM]

Oh the experts have arrived, the experts have arrived, the experts. This is pathetic -- an article is published by Tom and you all of a sudden start spewing the crap in the article -- are any of you capable of thinking for yourself or do you just take whatever drivel you read and the spin on it? What a bunch of spin doctors.

Oh please, a spelling mistake -- life goes on.

 

[P4Man]

>Oh the experts have arrived, the experts have arrived,
>the experts

We have a saying here: in the land of the blind, one eyed man is king.

>this pathetic -- an article is published by Tom and you
>all of a sudden start spewing the crap in the article

Who.. me ?

= The views stated herein are my personal views, and not necessarily the views of my wife. =