K10 too comlex for Intel to have fabbed at 65nm?

[wingless]

Heres the link: http://www.fudzilla.com/index.php?option=com_content&task=view&id=6353&Itemid=1

I really would like to find the OFFICIAL quote by Intel that says they could never have made Barcelona at 65nm because it was too complex. This would suggest AMD's process technology is pretty damn good. 65nm X2s would suggest otherwise though. AMD does use a 45nm IMMERSION lithography process which is fairly highend. Maybe theres hope for them yet come Bulldozer (if they stay in business that long). Either way Nehalem is sort of like a much improved K10 with Hyper Threading according to this guy. Whatever happened to AMD's reverse Hyper Threading tech they talked about a few years back? (Reverse HT: making multiple cores act as one to improve single-threaded performance several fold)

Quote:

Shared L3 cache and integrated memory controller


Nehalem is a scary word that promises a huge step forward in the CPU business, but in reality it is something that we've already seen, years ago.

The four cores inside are modified Core 2 45nm Penryn cores, the integrated memory controller is something that we saw with the K10 / Barcelona and Agena CPUs a while ago, and the shared L3 cache for all four cores is definitely something that we’ve seen from the Barcelona / K10.

Yesterday's conference call with Intel reminded us of a conversation that we had with some Intel engineers in the Summer of 2007 when they said that the K10 is a great concept, but it is tough (impossible) to make at 65nm. They used these words: we (Intel) cannot make the K10 in 65nm and Intel is known to have the best transistors in the word.

This now makes perfect sense, as Nehalem at its 45nm really looks like an advanced version of the K10 with integrated memory controller and L3 cache, but to be fair to Intel, Nehalem can cope with two threads per core and AMD's K10 and K10.5 cannot.

So, Nehelem is actually a new glimmer of hope for the K10.5, the new 45nm generation from AMD, as if Nehalem can work well at 45nm, there is a big chance that the K10.5 might be a success.

Nehalem looks like an advanced K10 / Barcelona
Written by Fuad Abazovic
Tuesday, 18 March 2008 10:00

 

[amdking]

um, seems to me he did not even know what he was talking about....

the integrated memory controller is something that we saw with the K10 / Barcelona and Agena CPUs a while ago

um....IMC was in the K8 (and of course K10) but his "a while ago" should have referred to K8....ROFL!!!

 

[JCLW]

It still remains to be seen if AMD can make a monolithic 65nm quad core K10, in volume, at a competitive speed.

This is the official quote that made the rounds:

At 65nm the die would be too big to hold four cores and it would be so expensive it would not make sense, our 45nm process technology will allow us to do a monolithic quad-core design.

http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=201804316

 

[FlahsMemory]

I don't think none of the stories got it right. What about Tukwila? 2 Billion transistors, 4 cores,and 65nm. http://news.bbc.co.uk/2/hi/technology/7223145.stm

Intel is basically saying it was stupid for them to do it at 65nm because of how expensive it would have been and how the public would have disliked it( new socket).

Also even though AMD can do K10 they do not have nearly as much customers to satisfy as intel. The cost of intel producing a monolithic core at 65nm for the general consumer base would have been exponentially great than AMD due to massive defect density and lower clocks. That is why I think intel is making a lot more 45nm fabs in preparation for monolithic quad even at 45nm. They still know it is going to be challenging and to combat that they are also going with modular design where they can just pull pieces off and throw them in different chips.

 

[Kuroimaho]

It's not a matter of could they but would their margins be high enough on those ? I bet it wouldn't be enough for Intel execs to nod on it. Their multidie approach worked out better in the end.

 

[funnyperson1]

Yea I don't think its a question of whether they could have done it. Its whether or not it would have been smart to do it. We can all see how well this worked out for AMD.

 

[AlabamaCajun]

Intel is just now heading into the territory with 6 core Larabee which puts 3 dual Penryn cpus on one die with a wad of L3 cache. Apparently Intel saw the gain in L3. The next platform or "Tock" will put 4 independent cores on one die each having only 256K private L2 and 2x32K L1 per core. Each core will supposedly be hyper-threaded. In place of larger L1 and L2 caches, Intel plans to use 8M of L3. AMD already has this and now can focus on making it work. What we hope to gain from AMD now is solid overclockable and fast procs. Just get within striking range of Penryn before Neha gets into consumer sockets. Back on Intel's side, you will need the new Mobos to use versions of Neha with IMC and quick path interface. I've seen leaks that a skt 775 version of Neha may be available.
With K10.5, we should be able to use our AM2+ boards as is with a possible BIOS upsate. If we want to use the DDR3 version then we will have to get a new MOBO.

 

[Hardin]

Intel is just now heading into the territory with 6 core Larabee which puts 3 dual Penryn cpus on one die with a wad of L3 cache. Apparently Intel saw the gain in L3. The next platform or "Tock" will put 4 independent cores on one die each having only 256K private L2 and 2x32K L1 per core. Each core will supposedly be hyper-threaded. In place of larger L1 and L2 caches, Intel plans to use 8M of L3. AMD already has this and now can focus on making it work. What we hope to gain from AMD now is solid overclockable and fast procs. Just get within striking range of Penryn before Neha gets into consumer sockets. Back on Intel's side, you will need the new Mobos to use versions of Neha with IMC and quick path interface. I've seen leaks that a skt 775 version of Neha may be available.
With K10.5, we should be able to use our AM2+ boards as is with a possible BIOS upsate. If we want to use the DDR3 version then we will have to get a new MOBO.

I'm not so sure that we will see a socket 775 nehalem. The big thing about nehalem is the imc and 775 motherboards already have a northbridge. So it might work on the socket but we would still have to buy new motherboards.

 

[dealmaster]

I personally think it's a pretty vast oversimplification to say Nehalem = K10 + Hyperthreading.

 

[Shiggity]

I'm not so sure that we will see a socket 775 nehalem. The big thing about nehalem is the imc and 775 motherboards already have a northbridge. So it might work on the socket but we would still have to buy new motherboards.

You won't see a 775 socket nehalem. A 775 board will not support quickpath and the CPU will not work without that.

I don't think nehalem will be considered that fast when it first comes out, the clockspeeds won't be that much higher and most apps don't even take advantage of 4 threads (current quad cores). Where as nehalem cranks up the memory bandwidth with tri-channel DDR3 and quickpath and you get double the threads. (Every core on a nehalem CPU will be double threaded, so 8 threads per quad core cpu) 16 threads on their dual quad core platform =D

As we have seen before, the most noticeable increase in performance always comes from increased clockspeeds, but we won't really see much higher speeds from penryn -> nehalem. However this is only due to programs not taking full advantage of quad core cpus. It will be a long long long time before we see apps that can use 16 threads simultaneously.

*I'm referring to threads as the number boxes in the window's task manager for CPU usage. Dual Core has 2, quad core has 4, nehalem quad has 8, and the nehalem dual quad core setup has 16.*

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3264&p=6 (bottom picture)

You saw right away from AMD that sacrificing clock speed for more memory bandwidth was a collossal mistake.

 

[AlabamaCajun]

At one point several sites posted this info http://wzus.ask.com/r?t=p&d=us&s=a&...u=http://www.dvhardware.net/article21965.html
that a 775 version of Neha was in the works for lower end Nehas and Legacy support. Of course this would mean no quick path or IMC. It's possible the info sprung up from the Dunnington based on merged Penryn cores ans shared L3 but those are socket 771.

Edit; from Ed's comments there will be two socket but not 775. http://overclockers.com/tips01256/

 

[Kuroimaho]

I personally think it's a pretty vast oversimplification to say Nehalem = K10 + Hyperthreading.

Don't they call C2D a reworked PIII ? They can't beat it just call it names.

 

[Shiggity]

At one point several sites posted this info http://wzus.ask.com/r?t=p&d=us&s=a&...u=http://www.dvhardware.net/article21965.html
that a 775 version of Neha was in the works for lower end Nehas and Legacy support. Of course this would mean no quick path or IMC. It's possible the info sprung up from the Dunnington based on merged Penryn cores ans shared L3 but those are socket 771.

Edit; from Ed's comments there will be two socket but not 775. http://overclockers.com/tips01256/

They said a nehalem 775 socket will require a new motherboard, waste of time by intel imo.

 

[AlabamaCajun]

Don't they call C2D a reworked PIII ? They can't beat it just call it names.

There was some info about Intel's Yonah using the Tualatin PIII as the base with new upgrades to give it more speed, better mathco and bigger caches. The Tualatin was a great chip on Intels part but they veered into a new P4 and P4HT processors that limited any revolutionary changes. The Tua and a Banis Pentium M CPU became the Yonahs leading to the Core greatness. Some reports as the Pentium Pro was the root?

This wiki, http://en.wikipedia.org/wiki/Intel_Core, is at least close if not gospel on the emergence of Yonah in 32Bit.

They said a nehalem 775 socket will require a new motherboard, waste of time by intel imo.

This just shows we can't trust anything on the net until someone buys it (frustrating).
I think by now Intel has dropped any plans (if they had plans to in the first place) to drop one onto the skt 775. I think the problem is that the CxQ we see now won't even physically fit into the 775 carrier. What I think may happed it Intel creates a line a CxDs that do fit into 775 by dropping the IMC and QPI since neither have any function on a FSB based board. (Just my Opinion). What I have heard of is a 715 socket that has a Neha based core and QPI but I did not see any details. I don't remember where I saw the 715 but in searching for "Intel NeHalem", I found it by accident.

 

[Kuroimaho]

Calling it a reworked P3 is a slight exaggeration, sure they scrapped the Netburst design and worked from banias, but based on this you could call K10 a reworked thunderbird.

Old news were socket 1160 and 715 around 2 years ago for Neha. Recently 1366 appeared and 715 vanished from the PDFs.
Might be that they choose triple channel for high end and required more pins so they bumped both high end and the rest. Then there is Beckton and its 1567pin socket and the mobile 989.

 

[AlabamaCajun]

AMD was a direct progression K7-K8-(dropped K9)-K10 so yes the bones of a Thunderbird are in there somewhere.
Intel's development was called a parallel to P4 with P-M being the root. I think it also has parts from PIII, P4 the P-M and P-Pro.

Edit; After a little searching http://lambcutlet.org/page/2/

Pentium M’s (namely “Banias” on 130nm and “Dothan” on 90nm) introduced in early 2003 had already revamped the “dead” Pentium III line in producing a low-power, high IPC CPU in pretty much the form that the RWT guys speculated way back in 2001 in how to make x86 run cool. “Yonah”, fabricated in the 65nm semiconductor process node allowed Intel to bring two lower power high performance cores into one piece of silicon!

Remember this is just the source of the original Banias core with some major updates. From what I got as the reason P4 was not the main root is the power hungry core was to far into the HT cores. Intel made a mistake going to P4 and just killing the PIII off. I remember how good the Tuas were but they were expensive. PIII 1000 was a hot CPU which like the Barton Athlons were hard to cool. Tuas came with a heat spreader but really gave off a lot less heat. P4 from day one was extremely hot even at 1.4G just 200mhz above the 1.2G Tua but a lot hotter. I still have 2 Tua rigs running today.

 

[Evilsizer]

im not sure why i find this so funny, since amd doesnt own there own fab's like intel does. they saying intel would be better using IBM to make there cpus like amd does,umm no. that would just have put intel behind, when you control the manufacting of your own product. you can get changed implemented quicker and setup testing lines for changes you plan to make. having someone else do the work for you in amd's case. mean that they prolly have to wait longer to get test chips with the changes they want to make.


P-Pro = root for P3, P3=root for Banis/Dothan/Yohan, then things get a bit fuzzy. as other changes were made to yohan to come up with core 2. now as most of it was reported then, the changes were added instruction sets and tweaks to the die. take it as you want, im still calling Core 2 a P3 on steriods.

 

[Cheator]

Calling it a reworked P3 is a slight exaggeration, sure they scrapped the Netburst design and worked from banias, but based on this you could call K10 a reworked thunderbird.

Well its a decendant of a P3 taulatin core (its the same base, but a small aprt of the base). Calling it reworked, I think is blatantly inaccurate, however it has a small truth to it.


Calling a K10 a reworked TB is less accurate thank that, since even the K8 and K7 difference is rather large, not to mention the X2 athlon versus the Athlon 64 single core.

 

[AlabamaCajun]

im not sure why i find this so funny, since amd doesnt own there own fab's like intel does. ~

AMD owns 3 Fabs, one in The States and 2 in Germany.

Lets face it, these two companies used what worked and learned from failures to get the current brood of chips we have. Really, they both have done outstanding jobs of building the latest tech and both have had recent blunders. Phenom and Penryn are in cycle now to correct problems but in the long run both will be great platforms. Comparing a Mustang GT to a Model T is about like comparing a Q6600 to a Tua PIII. You have to consider the middle ground to the evolutions.

 

[ddoggma]

I know what you getting at, but the Model T would have to be like a Pentium Pro or something. The Tua P3s were/are good and capable CPU, even today. I have a 1.26S in my HS and it works great,has for years now. Does cam encoding and Ts2 and files with no problems. Long live my 1.26S! Oh, it does 1.40 mhz no problem also. But I dont need the extra headroom at the moment, as she avg about 30% load. I allready have her replacment sitting here waiting [P4/2.8 northwood]. But I wont take her offline till she fails somehow.