Sandy Bridge E vs Ivy Bridge

[Nemesis 1]

Let's not forget that IVB's initial SKUs will have an upper bound on stock frequency/performance based on the pricing/performance tiers set up where SNB-E represents high-end/enthusiast and IVB represents a more mainstream product. And we already know where SNB-E lies based on released Intel roadmaps.

Why would intel set Ib clocks according sb-e clocks . Sb-e will have 6 and 8 cores . Intel will set IB clocks to match up with Highend BD . Intel didn't hold back SB in performance for the sake of 1366 socket cpus . No reason to believe they would start with IB .

 

[iCyborg]

won't IB end up undermining the sales for SB-E?

And sorry for my foolishness, but if Ivy Bridge draws less power, it's not going to overheat as much and in turn, overclock better?

So who would want to jump on SB-E if it ends up costing 3-4 times more?

re sb-e sales, IB will not effect them as SB-E will be priced above them. Inital costings of the inital SB-E is $300/$600/$1000. IB is proberly at the same price points as the existing SB CPUs, with proberly a short term extra for being "new".

As to IB overclocking, that is one of the reasons people are hoping it is as good as expected, but then intel will use some of that power savings to have CPUs that run at a lower total power just for better marketing reasons.

As to who wants SB-E, anyone with money and wants the best performance. Regardless if it is not cost effective.

I think it's a valid question. I don't like this latest Intel strategy even if it doesn't really affect me. For the last 6 months, and we're apparently in for 4-6 more at least, you can choose between mainstream SB and high-end 1366. Pretty much no one recommends 9xx over SB, it's overall slower except for some highly multithreaded scenarios, and even the cheapest 970/980 can buy you 2600K, a good 1155 MB and a couple beer rounds. SB-E will rectify that for a short period of time until IB comes, and then we'll have the same situation again, unless IB-E comes within a couple months of IB, but I don't see the point of SB-E in that case either. Or is it targeted primarily for the server market...

Specifically, for the bolded part: i7 970 is priced above 2600K, but very few people would recommend it unless you're doing something very specific. And for most people, best performance comes from 2600K and not the 3-4 times pricier 990X. And it looks like SB-E vs IB will be very similar to the current 1155 v 1366.

 

[Blue Shift]

I think it's a valid question. I don't like this latest Intel strategy even if it doesn't really affect me. For the last 6 months, and we're apparently in for 4-6 more at least, you can choose between mainstream SB and high-end 1366. Pretty much no one recommends 9xx over SB, it's overall slower except for some highly multithreaded scenarios, and even the cheapest 970/980 can buy you 2600K, a good 1155 MB and a couple beer rounds. SB-E will rectify that for a short period of time until IB comes, and then we'll have the same situation again, unless IB-E comes within a couple months of IB, but I don't see the point of SB-E in that case either. Or is it targeted primarily for the server market...

Specifically, for the bolded part: i7 970 is priced above 2600K, but very few people would recommend it unless you're doing something very specific. And for most people, best performance comes from 2600K and not the 3-4 times pricier 990X. And it looks like SB-E vs IB will be very similar to the current 1155 v 1366.

Not really. Since IB is a die-shrunk SB with only minor architectural changes, the performance differences compared to SB should be much smaller than Nehalem -> SB. The advantages of IB are smaller chips (less cost per chip to manufacture, and lower power usage) and potentially higher clocks. SB-E should still be faster.

 

[Lonbjerg]

I think it's a valid question. I don't like this latest Intel strategy even if it doesn't really affect me. For the last 6 months, and we're apparently in for 4-6 more at least, you can choose between mainstream SB and high-end 1366. Pretty much no one recommends 9xx over SB, it's overall slower except for some highly multithreaded scenarios, and even the cheapest 970/980 can buy you 2600K, a good 1155 MB and a couple beer rounds. SB-E will rectify that for a short period of time until IB comes, and then we'll have the same situation again, unless IB-E comes within a couple months of IB, but I don't see the point of SB-E in that case either. Or is it targeted primarily for the server market...

Specifically, for the bolded part: i7 970 is priced above 2600K, but very few people would recommend it unless you're doing something very specific. And for most people, best performance comes from 2600K and not the 3-4 times pricier 990X. And it looks like SB-E vs IB will be very similar to the current 1155 v 1366.

You are forgetting something.
When the socket 1336 launched, it bested the former king:
http://www.anandtech.com/show/2658/1

I got a i7 920 at launch and it has been chugging along at 3.5GHz ever since.
Current socket 1155 CPU's arn't real tempting.
Socket 2011 might be the first real incentive for me to upgrade.
Not bad since we are talking 2008:
http://www.anandtech.com/show/2658/1

 

[iCyborg]

Not really. Since IB is a die-shrunk SB with only minor architectural changes, the performance differences compared to SB should be much smaller than Nehalem -> SB. The advantages of IB are smaller chips (less cost per chip to manufacture, and lower power usage) and potentially higher clocks. SB-E should still be faster.

So 3D tri-gate is a minor arch change? SB was on time, so I would think that delayed IB would bring at least some similar improvement. Otherwise maybe Intel should do 2 tocks for 1 tick, if tocks take less time and bring more performance...
SB-E is even less of a change from SB, so one way or another, it will still be a choice between a more modern CPU on a newer process and older high-end; you still have to sacrifice something.

 

[iCyborg]

You are forgetting something.
When the socket 1336 launched, it bested the former king:
http://www.anandtech.com/show/2658/1

I got a i7 920 at launch and it has been chugging along at 3.5GHz ever since.
Current socket 1155 CPU's arn't real tempting.
Socket 2011 might be the first real incentive for me to upgrade.
Not bad since we are talking 2008:
http://www.anandtech.com/show/2658/1

I'm not forgetting since I bought 920 in Dec 2008 and I'm still using it (at 3.6, in your face, , just kidding). I don't plan to upgrade to SB-E or IB (nor Bulldozer), and I actually like that even after 3 years it still rocks, so it's just an academic matter for me

 

[GammaLaser]

Why would intel set Ib clocks according sb-e clocks . Sb-e will have 6 and 8 cores .

All I've seen from roadmaps is that SB-E will launch with 4- and 6-core versions. Are there any materials which actually mention 8-core versions?

Intel will set IB clocks to match up with Highend BD . Intel didn't hold back SB in performance for the sake of 1366 socket cpus . No reason to believe they would start with IB .

Look at the OC'ing headroom of 2500K/2600K. There's clearly sufficient parts to bin them at higher stock frequencies. But the question you have to ask yourself is, why would Intel release parts that at stock may cannibalize sales of higher-tiered parts because they offer more performance at the same # of cores but for 3x less $$?

I think Intel will leverage IVB process advancements first and foremost for its lower power consumption, which means plenty of OC'ing headroom for any unlocked SKUs. Later, process advancements emerge as performance gains through higher stock frequencies and/or more cores. It may not make sense that the latest process generation does not correspond to the highest pricing tiers but that's the strategy Intel is pursuing. Looks like they are aggressively wanting to push new technology into small form factors first to try to fend off ARM.

 

[Riek]

So 3D tri-gate is a minor arch change? SB was on time, so I would think that delayed IB would bring at least some similar improvement. Otherwise maybe Intel should do 2 tocks for 1 tick, if tocks take less time and bring more performance...
SB-E is even less of a change from SB, so one way or another, it will still be a choice between a more modern CPU on a newer process and older high-end; you still have to sacrifice something.

tri-gate has absolutely no effect on the architecture whatsoever. IB won't bring ipc improvement like sandy did..
SB-E is probably a bigger change than IB.. Sb-E adds a lot more cache then IB will.

 

[AtenRa]

I am thinking higher than that. Quad IB should push 4Ghz stock and well into the 5's (close to 6) max OC.

Yes, if IB will be a 95W TDP and keep the same power characteristics as SB, then we could see a 4GHz CPU to replace Core i7 2600K.

 

[Terzo]

Look at the OC'ing headroom of 2500K/2600K. There's clearly sufficient parts to bin them at higher stock frequencies. But the question you have to ask yourself is, why would Intel release parts that at stock may cannibalize sales of higher-tiered parts because they offer more performance at the same # of cores but for 3x less $$?

I think Intel will leverage IVB process advancements first and foremost for its lower power consumption, which means plenty of OC'ing headroom for any unlocked SKUs. Later, process advancements emerge as performance gains through higher stock frequencies and/or more cores. It may not make sense that the latest process generation does not correspond to the highest pricing tiers but that's the strategy Intel is pursuing. Looks like they are aggressively wanting to push new technology into small form factors first to try to fend off ARM.

I'm pretty sure you two are in agreement, it's just that Nemesis says Ivy will be clocked to match Bulldozer if necessary (at which point they would be stealing sales from AMD rather than their own higher tier parts). He also said the same thing about lower power consumption...post is referenced below for your convenience, with some highlighting thrown in.

I don't think so . I think Intel would use the same clocks and take advantage of the power savings as Being GREEN. There would however be the higher clocked K models that would enjoy a 18% clock increase befor it reaches present 2600k power usage . thats actually a huge amount of clock room . SO I look for Only the the higher end Ivy bridge to be clocked higher to deal with BD if BD is competitive . We could see a based clocked IB K model at a 3.9 ghz with 4.5 ghz turbo . That = todays 2600k in power usage. Thats about the best we can hope for other than minor changes that add up in % of performance gain at same power usage.

 

[BonzaiDuck]

Don't count on IB working with existing S1155 mobos even if it shares the same socket. History of S775 shall repeat itself again.

Sure, but as I said, that was as much due to nVidia producing chipsets for Intel cores. My best guess is that Intel -- if releasing a CPU for a particular socket, would make their own chipsets upwardly compatible with it. We're talking about a dominant firm in the industry that plans everything -- sometimes years in advance.

I'll bet a Mexican dinner that -- if Intel is releasing an "IB" chip for socket-1155, it will work in a P67 or Z68 motherboard, even if it needs a BIOS revision.

As for this . . . stuff . . . . I hear about "which market is target" for this or that processor or socket, it depends on which cores will be Locked, partially unlocked, or fully unlocked. But this is an entirely new ball-game. They'll produce the chips least expensive to produce for sale to mass OEM budget and midrange sales. It doesn't mean they won't have a flagship model for that particular socket or for "flagship" motherboards. It doesn't mean that savvy, budget-conscious "Enthusiasts" won't migrate to this or that CPU or chipset.

They've started unlocking these cores even at $300 price-tags for a new regime of "Turbo-savvy" chipsets which can in turn be used in OEM motherboards. Some here have thought that Intel has kept a leash on the technology, that they could've done some of these things years earlier. But that would be consistent with either a duopolist or a dominant firm that can call the shots even with a competitor like AMD, because they'll always have some idea -- some "intel" if you want to call it that -- of what the competition is capable of.

 

[GammaLaser]

I'm pretty sure you two are in agreement, it's just that Nemesis says Ivy will be clocked to match Bulldozer if necessary (at which point they would be stealing sales from AMD rather than their own higher tier parts). He also said the same thing about lower power consumption...post is referenced below for your convenience, with some highlighting thrown in.

Yeah...I really wanted to address that SNB-E is supposed to launch with a 4-core version. If SNB-E was 6/8 core only it may make more sense for Intel to launch IVB with presumably 4 cores or less and push the frequency as much as they can. Of course, BD plays a role in this as well, too bad we don't know how BD stacks up against SNB right now

 

[podspi]

It may not make sense that the latest process generation does not correspond to the highest pricing tiers but that's the strategy Intel is pursuing. Looks like they are aggressively wanting to push new technology into small form factors first to try to fend off ARM.

Does anybody know if Intel continuously improves their process as time goes on (or do they just work it until it is great, set it and forget it?). If they improve as time goes on, it makes sense they'd want a high-volume, but lower-bin (to improve yields) product to be "pipecleaner" for newer processes.

Way out of my area of expertise, but that would explain why they go for mainstream products on the new process first.

 

[Nemesis 1]

Steppings!??

 

[GammaLaser]

Does anybody know if Intel continuously improves their process as time goes on (or do they just work it until it is great, set it and forget it?). If they improve as time goes on, it makes sense they'd want a high-volume, but lower-bin (to improve yields) product to be "pipecleaner" for newer processes.

Way out of my area of expertise, but that would explain why they go for mainstream products on the new process first.

I'm sure there's definitely recipe tweaks that go on to improve chip yields throughout the lifetime that a process is used. I recall an Intel graph that plotted die defects vs. time for a handful of different processes and all of them had a steep drop as the process was developed and ramped up, and continued to decrease slowly until it was phased out by the next process.

In this case, IVB consists of both a shrink and tri-gate so this is especially going to be true as it's one thing to make prototypes of a couple of transistors in a lab and another to be able to consistently produce millions of very complex chips each containing billions of these things.

 

[Nemesis 1]

Intels uses a one fits all process . Once they decide on what tweaks to use . Its automaticly used in all the fabs . Any real changes intel makes usually comes with a new stepping. As in the past. But all tweaks are universal thru there fabs or thats as understand it .

 

[GammaLaser]

I recall an Intel graph that plotted die defects vs. time for a handful of different processes and all of them had a steep drop as the process was developed and ramped up, and continued to decrease slowly until it was phased out by the next process.

I knew if I looked a bit I could find it!

http://download.intel.com/pressroom/kits/32nm/westmere/Mark_Bohr_32nm.pdf

 

[Idontcare]

Does anybody know if Intel continuously improves their process as time goes on (or do they just work it until it is great, set it and forget it?). If they improve as time goes on, it makes sense they'd want a high-volume, but lower-bin (to improve yields) product to be "pipecleaner" for newer processes.

Way out of my area of expertise, but that would explain why they go for mainstream products on the new process first.

There's two kinds of "yield" from a high-level perspective.

There's parametric yield (clockspeeds and power-consumption binning) and then there's functional yield (defectivity, does it add 2+2 and get 4, etc).

I knew if I looked a bit I could find it!

http://download.intel.com/pressroom/kits/32nm/westmere/Mark_Bohr_32nm.pdf

^ addresses improvements in functional yield, not parametric yield.

Regardless, the short answer is yes, they do. Everyone does. When AMD did it they publicly referred to as CTI. I don't know of an IDM or foundry that doesn't actively seek to improve parametric yields. (or functional yields for that matter.

And Nemesis, the phrase you are seeking is "copy exact". Intel has had a copy-exact philosophy for fab alignment since the early 90's IIRC. Intel is not alone in this policy, it's industry wide but it varies from company to company in terms of just how "exact" the copy exact need be.

What generally happens is that you qualify multiple sources (be it the toolsets or the chemical suppliers) and then you "copy equivalent" from fab to fab in ways that are cost conscience.

There is no sense in requiring the fab in Israel to use the exact same source supplier for DIW or phosphoric acid as the fab in Oregon if a local supplier can provide equivalent product that meets the spec requirements.

But you can be sure they both will use the exact same make and model of immersion litho tools at $30m a pop.

 

[RussianSensation]

http://www.donanimhaber.com/islemci...lemcisi-Core-i73960Xin-ilk-test-sonuclari.htm

"If the slides are legit, Sandy Bridge E will arrive on the shoulders of the Core i7-3960X Extreme Edition. The CPU will have six cores and 15MB of cache, putting it 3MB ahead of the 990X. Interestingly, the Gulftown CPU actually has a higher base clock speed than its incoming replacement: 3.46GHz versus 3.3GHz. The 3960X has a higher Turbo peak, though. It'll scale up to 3.9GHz, while the 990X tops out at 3.73GHz. One of the slides also mentions a Core i7-3930K, which lacks Extreme branding but still features an unlocked multiplier.

According to the slides, the i7-3960X is 12-15% faster than the 990X in Cinebench, POV-Ray, and ProShow Gold. There are much bigger gains in SPEC CPU2006 and in Sandra's multimedia and memory bandwidth tests. In the latter, the 3960X's quad-channel memory controller offers more than double the bandwidth of the triple-channel 990X." - TechReport

 

[AtenRa]

According to the slides, the i7-3960X is 12-15% faster than the 990X in Cinebench, POV-Ray, and ProShow Gold. There are much bigger gains in SPEC CPU2006 and in Sandra's multimedia and memory bandwidth tests. In the latter, the 3960X's quad-channel memory controller offers more than double the bandwidth of the triple-channel 990X." - TechReport

That confirms (if the slides are correct), what I and others saying that Clock for Clock SB-E will have almost the same performance with SB.

I would say that Socket 2011 is not about the CPU and performance (Except for Multithreaded apps and 6-Core CPUs) but the platform features

 

[bryanW1995]

I will have to agree,

Ivy Bridge will have minor improvements in the CPU micro architecture over Sandy Bridge (AVX extensions and uArch optimizations) and clock for clock will have almost the same performance.

It is in the iGPU department that we will see the biggest differences with up to 16 EUs (12 for SB), PCI-e Gen3.0, DX-11 support and more.

Because of the 22nm process along with better power gating in iGPU and in Memory, power usage will be much lower both in idle and load than SB.

Are we talking about ~ 7 % increse like penryn had, or something even smaller than that?

 

[Edrick]

That confirms (if the slides are correct), what I and others saying that Clock for Clock SB-E will have almost the same performance with SB.

I would say that Socket 2011 is not about the CPU and performance (Except for Multithreaded apps and 6-Core CPUs) but the platform features

I agree. The x79 platform will be a major reason to upgrade. As well as the use of 6 core CPU's (and possibly 8 later on).

 

[Edrick]

Are we talking about ~ 7 % increse like penryn had, or something even smaller than that?

I am willing to bet clock for clock will be less than 7%.

But I expect clocks to be upwards of 10% faster.

And the IGP will see the largest performance increase of SB.

 

[Nemesis 1]

IB 1155 will be only as fast as it needs to be. The focus is on power consumption.
IF BD is within 5% performance of the 2600K . The IB equivilant wil be 20%+ faster than the SB2600K Intel has headroom for about 30% performance increase at same power usage. SO until we see real BD performance numbers IB speculation is lame. 3D trigate gives intel some real headroom at 22nm. As for IGP improvements throw those out the window based the EU counts that were Let out way to early by AT to be real or = to the DX11 EUs that will be on IVB. IB will be 2x better than SB igp in performance at the min.

 

[386DX]

IB 1155 will be only as fast as it needs to be. The focus is on power consumption.
IF BD is within 5% performance of the 2600K . The IB equivilant wil be 20%+ faster than the SB2600K Intel has headroom for about 30% performance increase at same power usage. SO until we see real BD performance numbers IB speculation is lame. 3D trigate gives intel some real headroom at 22nm. As for IGP improvements throw those out the window based the EU counts that were Let out way to early by AT to be real or = to the DX11 EUs that will be on IVB. IB will be 2x better than SB igp in performance at the min.

I agree with you IB is going to have alot of headroom so Intel's gonna be able to play with frequency/turbos to make it perform where it wants without killing there higher end. I predict that Intel is going to try and replace the SB 2500K with a dual core IB + HT that will be faster then 2500k at single thread and within 90% at multi-thread at a price point of around $140. The 4 core IB will probably come without HT and be faster then the 2600K at both single and multi-thread and be priced around $216.