Sandy Bridge Socket

[Idontcare]

Originally posted by: alyarb
you won't get 4 physical 32nm cores until sandy.

Sadly you are probably right, quad-core sandy is likely to beat AMD 32nm to market.

Could you imagine if the process technology gap was this large in the GPU market?

 

[imported_Lothar]

Originally posted by: alyarb
you won't get 4 physical 32nm cores until sandy.

Well...I will be upgrading my computer soon within a month then.
Sandy bridge is too long on the horizon for me(2011). I might consider Ivy bridge which is the die shrink or just jump straight to Haswell/Bulldozer II(or whatever AMD's equivalent is).

 

[IntelUser2000]

Originally posted by: alyarb
who is complaining? most of us want more FP for multimedia transcoding. most single-threaded mainstream apps are fast enough on a modern architecture, and intel feels the same way. that's why they're focusing on turbo mode and dynamically allocated power efficiency than another risky, revolutionary pipeline that throws inordinate amounts of die area at branchy bullshit. that would only encourage people to write shittier code and postpone the eventual transition to LRBni or what-have-you. if you read the last paragraph in my post, i kind of addressed that.

Problem with that is GPUs are starting to address that(read the GT300 Fermi article). There will still be things that CPUs are superior at even in FP, but most people will still be better off with per core improvement in integer. Kind of the reason people aren't excited about Nehalem(even though performance advance is great, but not really in areas people want).

As for code, history has been shown not relying too much on compiler advances and code efficiency, but instead on hardware was a success.

Otherwise VLIW architectures, or at least some form like Itanium would be more successful now. Lazy programmers stay lazy.

 

[RussianSensation]

Originally posted by: Lothar

Are you saying "No Nehalem 32nm quads at ALL" or that there will be none on socket 1156???

Well I can't tell you with certainty that socket 1366 or 1156 will never get 32nm quads since Intel can always change their strategy. But Core i7 920/i 7 860 are going to remain at the same performance level in the hierarchy until at least Q3 2010. So there will hardly be any developments on the quad core front for socket 1156 for 12 months from today! If you need a new computer, there is no point in waiting. Both 920 and i5 750/860 are excellent choices.

 

[Idontcare]

Originally posted by: RussianSensation

Originally posted by: Lothar

Are you saying "No Nehalem 32nm quads at ALL" or that there will be none on socket 1156???

Well I can't tell you with certainty that socket 1366 or 1156 will never get 32nm quads since Intel can always change their strategy. But Core i7 920/i 7 860 are going to remain at the same performance level in the hierarchy until at least Q3 2010. So there will hardly be any developments on the quad core front for socket 1156 for 12 months from today! If you need a new computer, there is no point in waiting. Both 920 and i5 750/860 are excellent choices.

Or it could just be that we are the ones who are unawares of Intel's existing strategy. We are all guilty of assuming 32nm westmere will be 2C and 6C but no 4C...however I don't exactly recall Intel ever stating this, nor do I remember seeing anything that would imply they are not planning a 4C westmere product.

 

[RussianSensation]

Since 32nm quad cores for 1156 are not even on Intel's Latest roadmap, it's hard to just assume that suddenly they will reappear. Furthermore, since AMD is uncompetitive to Core i7, Intel has little to no incentive to release 32nm quad cores to cannibalize the sales of their 45nm processors on the mainstream. Instead, it would make more sense to ride out their 45nm process in order to reduce the capital costs of the 45nm manufacturing equipment and use 32nm Gulftown as a way to test the new 32nm process (i.e, wait until it matures).

Originally posted by: Idontcare
We are all guilty of assuming 32nm westmere will be 2C and 6C but no 4C...

I am not assuming. It has been stated that no 32nm quad cores will appear on 1156 in 2010:

Keep following; if you want a quad-core Westmere, your only option will be in the LGA-1366 socket with Gulftown. Core i7 will get replaced with a six-core, twelve-thread processor in early 2010. There won?t be a 32nm quad-core part on the desktop until the end of 2010 with Sandy Bridge.

 

[ilkhan]

I think IDC meant that Intel may disable two cores and sell gulftowns as 4c products. I don't think it's likely, and I doubt IDC thinks it's likely, but we think its *possible*. Intel keeps saying gulftown/clarkdale/arrandale are the only westmere products. AFAIK Intel has never said gulftown has 0% chance of ever being sold as a 4c cut-down product.

Speaking of, has anyone seen a gulftown die shot yet? I want to see how the die is arranged compared to bloomfield/etc.

 

[Idontcare]

Originally posted by: RussianSensation
Since 32nm quad cores for 1156 are not even on Intel's Latest roadmap, it's hard to just assume that suddenly they will reappear. Furthermore, since AMD is uncompetitive to Core i7, Intel has little to no incentive to release 32nm quad cores to cannibalize the sales of their 45nm processors on the mainstream. Instead, it would make more sense to ride out their 45nm process in order to reduce the capital costs of the 45nm manufacturing equipment and use 32nm Gulftown as a way to test the new 32nm process (i.e, wait until it matures).

Originally posted by: Idontcare
We are all guilty of assuming 32nm westmere will be 2C and 6C but no 4C...

I am not assuming. It has been stated that no 32nm quad cores will appear on 1156 in 2010:

Keep following; if you want a quad-core Westmere, your only option will be in the LGA-1366 socket with Gulftown. Core i7 will get replaced with a six-core, twelve-thread processor in early 2010. There won?t be a 32nm quad-core part on the desktop until the end of 2010 with Sandy Bridge.

I don't see there where Intel says anything there about 4C westmere products not being planned as being in the wings perchance AMD delivers something with Thuban that requires a response.

Marketing tells us what marketing wants us to think/know; I'm not about to make the assumption that marketing is telling us everything.

What I do see is you equating Anand's statement as being that of Intel...which is an assumption on your part. Correct me if I'm wrong there.

Naturally you are allowed to be of the opinion that you aren't making any assumptions. No need to bold it and scream it at me though. I read just fine.

 

[imported_Shaq]

Originally posted by: IntelUser2000

Originally posted by: RussianSensation

- There are no 1156 32nm quads on the roadmap in 2010.
- Sandy Bridge is the "tick" in Intel's strategy, which means a new architecture (like Nehalem is to Conroe).
- 32nm Gulftown will just be a shrunken 6-core Nehalem with no architectural improvements and $1k+ price tag to boot as far as I read.
- Sandy Bridge is expected Q1 2011 so it's more like 1.5 years away.
- Even if Sandy Bridge was compatible with 1156, the lack of proper power regulation/circuitry as a result of a new architecture would make it incompatible with the current P55 chipset. Let's not forget how often Intel changes chipsets!

BTW, Sandy Bridge and Nehalem are Tocks. It must have confused people because "Tick" is usually the first in line(you know, from a clock, tick comes before tock), but since they count Pentium D 65nm/Core Duo as Tick, new architectures are Tock.

I hear Gulftown will have one lower than the EE. So you might be able to buy at $600.

EE: 2.4GHz base, 2.66GHz Turbo Mode
Non-EE: 2.4GHz base, 2.53GHz Turbo Mode

Turbo won't be impressive on Gulftown.

Originally posted by: drizek
Yes, it is a new architecture, but it won't be like Nehalem was to Conroe. Performance/clock is not going to go up significantly from what I understand, certainly not the 0-60% improvement in performance that came with nehalem.

The problem on Nehalem is that per core performance didn't go up significantly. Even accounting multi-threading, the improvements lie in average of 20%.

We'll see what happens.

I didn't know they would have a slower, cheaper Gulftown than the EE. That increases my chances of getting one(from 1% to 20%). I never spent $500 or more on a CPU before though so it is still unlikely. It's good news to hear it will be available however.

 

[RussianSensation]

Originally posted by: Idontcare

Naturally you are allowed to be of the opinion that you aren't making any assumptions. No need to bold it and scream it at me though. I read just fine.

Sorry didn't mean to portray an increased tone of voice You are correct in that Intel's strategy is dynamic and will change depending on the competitive marketplace offerings. So the possibility of 4C Lynnfield on 1156 is highly unlikely in 2010 but may be somewhat probable.

 

[Idontcare]

Originally posted by: RussianSensation

Originally posted by: Idontcare

Naturally you are allowed to be of the opinion that you aren't making any assumptions. No need to bold it and scream it at me though. I read just fine.

Sorry didn't mean to portray an increased tone of voice You are correct in that Intel's strategy is dynamic and will change depending on the competitive marketplace offerings.

Water under the bridge, its cool, my bad for needlessly over-reacting to your post, you didn't deserve my tone either.

Originally posted by: RussianSensation
So the possibility of 4C Lynnfield on 1156 is highly unlikely in 2010 but may be somewhat probable.

Speaking of which, you notice the 'dales are 32nm and the 'fields are 45nm? So I wonder if a 4C westmere would end up being called a lynndale?

Did we ever come to any conclusion whether or not the whole "Core i7/i5/i3" branding scheme actually finally made any sense now that we have so many more pieces of info beyond just i7?

 

[ilkhan]

it makes sense to a degree. Pentium will continue as duals without turbo or HT. i3 are duals with HT (4 threads) without turbo. i5 are 4 threads with turbo. i7 is 8 threads with turbo. To a degree it doesn't matter, just referring to them by series 3xx/4xx/5xx/6xx/7xx/8xx/9xx provides more info than the i3/i5/i7 does.

IDK if you hadn't noticed this before or were just being amusing, but: 'dale refers to duals, 'field refers to quads (not process tech, IE: wolfdale(45nm)/kentsfield (65nm)/yorkfield(45nm)). "clark" being used on both simultaneously is an annoying and confusing coincidence. I suspect 'town and 'ton will live on for a while beyond gulftown/beckton.

 

[Idontcare]

Originally posted by: ilkhan
it makes sense to a degree. Pentium will continue as duals without turbo or HT. i3 are duals with HT (4 threads) without turbo. i5 are 4 threads with turbo. i7 is 8 threads with turbo. To a degree it doesn't matter, just referring to them by series 3xx/4xx/5xx/6xx/7xx/8xx/9xx provides more info than the i3/i5/i7 does.

Yeah, that's where it continues to be a pointless brand/labeling scheme to me...I jsut remember the Intel PR people told the world that it would all become clear and make sense once the rest of the labeling scheme was released.

So now the conclusion appears to be "yeah it is now more clear, but at the same time it appears to be needless and irrelevant given the #xx model qualifier in which the # communicates the info anyways".

Originally posted by: ilkhan
IDK if you hadn't noticed this before or were just being amusing, but: 'dale refers to duals, 'field refers to quads (not process tech, IE: wolfdale(45nm)/kentsfield (65nm)/yorkfield(45nm)). "clark" being used on both simultaneously is an annoying and confusing coincidence. I suspect 'town and 'ton will live on for a while beyond gulftown/beckton.

I wasn't trying to be amusing, so you are right that I had actually not realized that before. It does make a lot more sense now, thank you for explaining that to me.

Dunnington was six-core, I guess they really should have called the other two as Gulfton and Becktown to keep the 6-core = ton and 8-core = town decoder ring going. Or maybe Dunnington really should have been Dunningtown?

At any rate, 'dale vs. 'field = dual vs. quad...got it :thumbsup: thanks for politely vanquishing my ignorance there.

 

[ilkhan]

Intel (until now) hasn't had much consistent across core/xeon line-ups, dunnington/beckton may just have been pre-consistent-naming-system. :shrug:

 

[IntelUser2000]

I've been thinking the difference between 'dale vs. field' is "mainstream desktop"
vs. "high end desktop".

See, the first "field" CPU was Smithfield, a dual core MCM based on two Prescott cores. Then again we might all be dreaming up something non-existent.

Chipsets have some patterns too.

"3 series"(P35/G35/Q35): Bearlake
"4 series": Eaglelake
IGP inside Clarkdale: Ironlake

 

[Idontcare]

Originally posted by: IntelUser2000
I've been thinking the difference between 'dale vs. field' is "mainstream desktop"
vs. "high end desktop".

See, the first "field" CPU was Smithfield, a dual core MCM based on two Prescott cores. Then again we might all be dreaming up something non-existent.

Chipsets have some patterns too.

"3 series"(P35/G35/Q35): Bearlake
"4 series": Eaglelake
IGP inside Clarkdale: Ironlake

So we can expect the next chipset to be named either Putterlake, or Birdielake, or Salmonlake. :laugh:

 

[IntelUser2000]

Originally posted by: Idontcare
So we can expect the next chipset to be named either Putterlake, or Birdielake, or Salmonlake. :laugh:

BTW, if you guys haven't noticed I've been trying to move non related topic to the more appropriate thread(the "other" Sandy Bridge one) with no success. :roll:

 

[Idontcare]

There's another sandy bridge thread?

Yeah, if the OP agrees then it probably makes sense to merge the two threads since there is relevant info in both. A mod can do it, but I think the OP has to request it or at least approve it once someone else requests it.

 

[IntelUser2000]

Yea, the one you put up.

Anyways, I was just comparing the die sizes between Pentium III 0.13u "Tualatin" and the vastly improved version, the Pentium M "Banias"(both at Intel's 0.13u).

Amazingly, the die sizes are similar

Pentium III Tualatin: 80.4mm2
Pentium M Banias: 84mm2

That's even more amazing since Pentium M has extra 512KB more cache. Even accounting for the possibility that the process tech has matured and offered probably 10% more die size reduction, its quite an engineering feat. 512KB cache took 14-15mm2 die size so the comparison is really:

Tualatin: 80.4mm2
Banias "512K": 69mm2

Banias improved performance over Coppermine core based(0.18u) Pentium IIIs by 25-40% per clock.

 

[cbn]

Could you imagine if the process technology gap was this large in the GPU market?

It would be a absolute blowout for one video card company if that happened.

AMD really needs to figure out how to get to smaller nodes faster.

CPUs are obviously different than GPUs in that specialized programs are needed to harness all the die area. But still I would like to see them beat Intel to 22nm, 16nm or 11nm.

 

[cbn]

Do we know anything about Sandy Bridge other than it will require a new socket? What features will it have? Extra L3 cache? More L2 cache? Or will this be another micro-evolutionary step in the same way Nehalem was?

Will a mainstream socket appear along side X68 or will we have to wait another year like we did for P55?

My money is on Dual Module Bulldozer for the "sweet spot" in the 32nm age. Trouble is that crown probably won't last long as Intel might be on 22nm 6 months later.

 

[Idontcare]

Same L2$ and less L3$.

Westmere:

Sandy Bridge:

Source: http://pc.watch.impress.co.jp/docs/column/kaigai/20091127_331818.html

 

[cbn]

GPU and CPU on the same die? This is going to be the top chip?

 

[Zensal]

GPU and CPU on the same die? This is going to be the top chip?

I believe it is going to be all chips.

GPUs seem to be taking on more roles lately than just graphics processing. This looks to be the start of specialized cores being integrated along side regular x86 cores. So, for things that work better on GPUs, programmers can use that, and for things that work better on x86, there will be plenty of cores for that also.

 

[cbn]

I believe it is going to be all chips.

GPUs seem to be taking on more roles lately than just graphics processing. This looks to be the start of specialized cores being integrated along side regular x86 cores. So, for things that work better on GPUs, programmers can use that, and for things that work better on x86, there will be plenty of cores for that also.

That is interesting because AMD Bulldozer for enthusiast socket doesn't have an integrated GPU according to the roadmap.