socket 478 + 1 = 479

I'm no EE based, but AFAIK, it depends on the contact pin current rating and bandwidth requirement -Xdb @ YGHZ

oh man, I wanted AMD to take back the crown this time.

if intel and AMD take turns going on the podium, we can expect better competition, better product, cheaper price from both.

But if AMD loose for the second time, intel is going to jack the price up for sure.

LGA479?? I thought they said last time when LGA775 release that they wanna make 1 universal socket... now wanna do LGA479 pulak...

Based on info I have kucingfight is correct on the size but pin/pad/ball count is 478 not 479.

Penryn on desktops/workstations (Wolfdale etc) will be on existing LGA775/771.

For Nehalem the pin count I know so far is more than kucing's number.

thus the 478/479 is for Mobile only?

Yes, I think it's called Socket P.
I'm not sure if "P" stands for "Penryn", but for LGA775 a.k.a. Socket T the "T" stood for the cancelled "Tejas" proc.

All my info could be wrong; there's too many damn codenames to remember.
FYI all the codenames are geographical places where Intel has a strong influence.

I'm not very EE either, but the Conroe used LGA775, and Penryn will have the same power envelope and higher FSB so a reduction in pin count would be quite strange to me.


That sounds more like it. And it does sound likely that Nehalem would easily have >1000 pins, looking at current AMD solutions. Unless CSI is a really, really high clocked serial bus, lol.

I can pretty much confirm the info on Penryn.

There's actually a few variants of the Nehalem core and the only info I have is on the higher-end server parts. Kucing might be right if he was talking about the mainstream versions of Nehalem.

Information on Nehalem is pretty scarce even internally (and some are pretty interesting!). But do expect quite a few changes from the current PC platform.

Kucing: I think you mentioned before you work with 1 of the substrate suppliers - is that correct?

i agree with u...

amd can no fight any longer...
intel had oredy release it core2quad but amd still fumbling behind...
but it is no good either...there will be no competition...

Intel roadmap from 65nm to 32nm tech. ( AMD killer )

not just that, amd just introduce 65nm architecture but intel prepare 45nm and incoming year 32nm, new architecture, more core, higher FSB, more cache, build-in memory controller...

amd will "bungkus" lio...

Never ever write off AMD; they're the Rocky Balboa of the processor world (always get up even after getting hit in the chin).
They've got great engineers + a little help from IBM.

Respect to the men in green.

AMD used to sux too, back in the old old days. When they are poor and tiny. Hot AMD that burned itself up.

but their athlon XP and AMD64 took the world by surprise.

despite their lack of resources, they were able to take down intel for many years. Until Core came along.

So, now... with the profit they've obtained from athlon and AMD64, they are way richer than the thunderbird era.

so expect them to cook up something super top secret and powerful. Real soon.

Something to SURPRISE us again!!.

Nothing super secret left about K8L which is coming soon-ish.
Whether it's super powerful is yet to be determined.

well, there is still a super secret - Where is the damn benchmark?

I don't know, but from what i assume were the best case comparative figures that have been released by AMD so far, it seems that it's not going to wipe out intel, especially not Penryn. I hope there will be surprises, but at this point i don't think it will be significantly better than any intel offerings at the time, at least in the single chip desktop space. In the server space, would bet that the K8L will win, especially in 4 socket or more systems.

Also as an aside, i just realized that the high-k + metal gate tech solves a lot of problems with using germanium channel transistors, which will probably replace silicon in the not so near future since it's a lot faster. So this new gate tech seems to be a much better long term improvement than most people realize so far.

This post has been edited by ikanayam: Apr 3 2007, 06:59 AM

According to knowledgable ppl germanium is very difficult to work with - the fabs needs to be a couple of factors cleaner than normal fabs. Plus germanium itself is expensive compared to good old Si.

Yes. A large part of that problem was that the previous dielectric used for Si transistors (SiO2) don't work very well with Ge channels. The high-k hafnium based dielectrics fix that part of the problem. Ge (~900C) melts at a much lower temperature than Si (~1400C), and the use of metal gates fixes that problem because they don't have to be doped at 900C like the polySi gates.

It is extremely unlikely that they will use Ge wafers because Ge is very unstable compared to silicon (which is the other big problem). Just the transistor channels will be Ge. The improvements are substantial, potentially ~2x electron mobility and ~4x hole mobility improvement. Cost of the material itself is not that big a deal, Hafnium used in the high-k dielectrics is not exactly cheap or abundant either. But these things won't be used a lot, just on the transistor channels, which are only a few atoms thick and a few tenths or hundreds of atoms wide. Plus current chips are already using SiGe (15-20% Ge IIRC) to strain the silicon channel.

Fabs today are a LOT cleaner than yesterday, so i don't see why tomorrow's fabs can't be cleaner than today. What i think might be a bigger potential problem is the fact that Ge has a much lower band gap than Si, so it's a lot more prone to leakage issues caused by temperature increases.

This post has been edited by ikanayam: Apr 3 2007, 09:03 AM

more Penryn speculation. it just a mere speculation, dont take it as a fact, spam that author if u wish.

*who need OC anyway, with 400fsb, it give me hell lot of performance boost from current C2D.

This post has been edited by edwin3210: Apr 3 2007, 09:09 AM