About Sandy Bridge E and Ivy Bridge

[kalistan9]

Sandy Bridge E just came out, and I was wondering if Ivy Bridge would out-perform it? Does the Ivy Bridge have as many cores as the Sandy Bridge E?

I was confused because it lists the Sandy Bridge E as being enthusiast, top performing and Ivy Bridge as mainstream.

I don't see how Ivy Bridge is 22nm and trigated yet listed as being lesser on the Enthusiast's Market. Why would Intel have superior die technology yet make the Ivy bridge having inferior cores or cpu speed, etc. If anyone could explain to me whats up in the market, and which would be the best buy, I would appreciate it? Thank you!:\

 

[gevorg]

Ivy Bridge will definitely outperform it at single to four-threaded apps. Unlikely to outperform it at apps that can efficiently use 5+ cores.

 

[kalistan9]

Why the hell don't they make a 6 core or 8 core processor based on 22nm and trigate since they have the technology matured

They should be making an 8, or in the very least 6 core cpu. It shouldn't be too hard of a change for the die process.

 

[gevorg]

Checkout the specs of Sandy Bridge-E, there is more to it than just 2 more cores. Ivy Bridge-E will be a true step up from Sandy Bridge-E.

Its not always about the # of cores, AMD made a mainstream 8-core CPU and it can't even beat 4-core 2500K in most benchmarks.

 

[nitromullet]

Why the hell don't they make a 6 core or 8 core processor based on 22nm and trigate since they have the technology matured

They should be making an 8, or in the very least 6 core cpu. It shouldn't be too hard of a change for the die process.

They will... That will be Ivy Bridge-E.

Ivy Bridge is a tick to the Sandy Bridge tock, so no major changes other than a die shrink.

http://www.anandtech.com/show/4318/intel-roadmap-ivy-bridge-panther-point-ssds

Ivy Bridge will be a die shrink of Sandy Bridge and represents a “tick” in Intel’s product line. That means the micro-architecture is mostly the same as Sandy Bridge, but it will be manufactured using 22nm process instead of 32nm.

First we'll see 6 core Ivy Bridge-E on LGA 2011 for enthusiasts later in 2012 (or early 2013). Then in mid 2013 we'll see the next tick from Intel which I imagine might include 6 core mainstream and performance oriented cpus on 22nm.

 

[frostedflakes]

Why the hell don't they make a 6 core or 8 core processor based on 22nm and trigate since they have the technology matured

They should be making an 8, or in the very least 6 core cpu. It shouldn't be too hard of a change for the die process.

For mainstream processors, they seem to be opting for higher integration instead of adding more cores. They seem to be more interested in using the saved die space from process shrinks to integrate more robust graphics, PLL clock generator, voltage regulators on the upcoming IB, random number generator, stuff like that. And considering their mainstream quad cores still kick the crap out of 6-core and 8-core chips from their competitor, it's hard to complain about this strategy. IB probably won't bring any huge IPC increases (maybe a couple percent due to slightly larger cache, small tweaks like that), but if it offers significantly more overclocking headroom than SB, which is already a really good overclocker, Intel will have a real winner on their hands for enthusiasts IMO, even if it "only" has four cores.

 

[greenhawk]

Why the hell don't they make a 6 core or 8 core processor based on 22nm and trigate since they have the technology matured

What you want is IB-E, the replacement for SB-E in about a year (going by intel's current 1 year per release roadmap).

As to why not make a 6/8 core for the main stream, it is a mix of the following.

-marketing
-what people use
-what the market will pay for
-intel pacing release so they can get back R&D costs

edit: got to love posting without refreshing first.

 

[greenhawk]

Ivy Bridge would out-perform it?

Does the Ivy Bridge have as many cores as the Sandy Bridge E?

I was confused because it lists the Sandy Bridge E as being enthusiast, top performing and Ivy Bridge as mainstream.

Why would Intel have superior die technology yet make the Ivy bridge having inferior cores or cpu speed, etc.

If anyone could explain to me whats up in the market,

which would be the best buy, I would appreciate it? Thank you!:\

1- Hopfully a little bit in single threaded applications. No chance in properly multithreaded applications

2- Officially, no. Ivy maxes at 4. SB-E has 6 on the cpus released (though a cut back SB-E is due next year with 4 cores)

3- It is Intels first cpu for mass production on 22nm, It is therefore smaller and easier to manufacture. Thinking of it more of a provind ground of 22nm/3d transistors. (ie: outside of the lab / into mass production)

4- Currently SB, as it is well priced, avaiable and well tested. SB-E is new and expencive, IvyBridge is still 6ish months away and it is unknown if it will have any release day issues like SB did. Not expected, but not 100% either.

 

[dma0991]

IB is all about the GPU and power savings and not a whole lot of improvement on the performance side. Why should Intel go through the trouble which costs a lot of money to develop a 6 core mainstream IB when AMD is barely even a challenge. Besides, IB is a Tick+ which is merely a die shrink compared to SB with the exception that there is Tri-Gate thrown in to improve it.

Probably you might find mainstream 6 core offerings from Intel when Haswell arrives but from what I know Haswell will still remain at 4 cores with better integration and improvement in IPC instead of improvement in core count.

 

[superccs]

They will... That will be Ivy Bridge-E.

Ivy Bridge is a tick to the Sandy Bridge tock, so no major changes other than a die shrink.

http://www.anandtech.com/show/4318/intel-roadmap-ivy-bridge-panther-point-ssds



First we'll see 6 core Ivy Bridge-E on LGA 2011 for enthusiasts later in 2012 (or early 2013). Then in mid 2013 we'll see the next tick from Intel which I imagine might include 6 core mainstream and performance oriented cpus on 22nm.

With little to no competition on the high end why would Intel want to make any of its chips irrelevant if it doesn't have to. Bank some tech to smash your competition when it shows up.

 

[IntelUser2000]

2- Power Point presentation officially, no. Ivy maxes at 4. SB-E has 6 on the cpus released (though a cut back SB-E is due next year with 4 cores)

Corrected. They can always release it near the end of the lifecycle.

I don't see how Ivy Bridge is 22nm and trigated yet listed as being lesser on the Enthusiast's Market. Why would Intel have superior die technology yet make the Ivy bridge having inferior cores or cpu speed, etc. If anyone could explain to me whats up in the market, and which would be the best buy, I would appreciate it? Thank you!:\

The way I see it, all the talk and criticism towards Intel about low power products have refocused the 22nm process enhancements towards low power. So if you want mobile Ivy Bridge, great for you! Ultrabooks, even better! The way its going, 3820 chip is going to be matched in performance by the top tier mobile Ivy Bridge chips, for the first time ever. Sandy Bridge was close, but it'll definitely close the gap in Ivy Bridge. Even if 3820 outperforms the 2700K by 10% Ivy Bridge mobile might get close.

 

[BrightCandle]

To some extent its bad market segmentation on similar products and a delayed release causing confusion. Ideally Intel would have released SB-E at a similar time to SB allowing a choice between mainstream and professional markets. That would have made it clear what the performance level and capabilities would be. It didn't work out that way, SB-E clearly hit some snags on its route to market.

But with Ivy looming SB-E will likely have less per core performance than SB-E, so it has always been with processors. The move in process brings with it benefits that the best harvested dies from the previous generation can never out do. The only difference between a lot of CPUs you buy is simply how fast/well they tested and were binned. They are all the same product. SB-E is a SB core with the GPU removed and a bucket load of cache added along with more memory channels. With Ivy E being a distance away it looks even odder than 2011 is the so called high end. I think its just bad release timing.

We aren't looking at massive performance improvements from Ivy at the moment anyway and SB-E is in some circumstances a marked performance improvement over SB with the 2 extra cores. Its also quite a bit cheaper than a 990X and similar to a 980, so overall it works out that 6 core performance is cheaper on SB than on 1366. But they led with the mainstream and its definitely going to hurt their sales of the high end chips.

 

[PlasmaBomb]

With little to no competition on the high end why would Intel want to make any of its chips irrelevant if it doesn't have to. Bank some tech to smash your competition when it shows up.

Which is why SB-E is an 8 core die with 2 cores disabled. If AMD were to show up with something competitive I'm sure Intel could "find" two more cores...

 

[exar333]

Which is why SB-E is an 8 core die with 2 cores disabled. If AMD were to show up with something competitive I'm sure Intel could "find" two more cores...

Power consumption.

They could re-enable the extra cores, but clocks would need to be dropped to likely ~2.4-2.6. Intel likes to keep single-threaded performance high, so I doubt this be released until IB-E.

 

[greenhawk]

Corrected. They can always release it near the end of the lifecycle.

Unless the IvyBridge cpus being released have disabled cores, I stronly do not think intel is going to re-create a new mask to allow the creation of a 6 core s1155 cpu. So no correction needed. Making a mask is not cheap and definitly is not worth doing for a limited run of cpus at the end of a life cycle. Espically as the Ivybridge replacement already has a date lined up for its release.

 

[greenhawk]

Which is why SB-E is an 8 core die with 2 cores disabled.

More to do with being a Xeon reject as I see it. The SB-E is more likly chips that did not make the cut to be a Xeon / 8 Core unit. As the quality of production improves we might see a 8 core before the end of life from the SB-E, but intel generally wants to keep to it's price points of $1000 for the top chip. Looking at the Xeon range, it is allowing some to be well over $1000, and those ones have all the trimings (20MB cache, 8 cores ect).

 

[Edrick]

More to do with being a Xeon reject as I see it. The SB-E is more likly chips that did not make the cut to be a Xeon / 8 Core unit. As the quality of production improves we might see a 8 core before the end of life from the SB-E, but intel generally wants to keep to it's price points of $1000 for the top chip. Looking at the Xeon range, it is allowing some to be well over $1000, and those ones have all the trimings (20MB cache, 8 cores ect).

Read:

Power consumption. They could re-enable the extra cores, but clocks would need to be dropped to likely ~2.4-2.6. Intel likes to keep single-threaded performance high, so I doubt this be released until IB-E.

Desktop parts needs to be kept at 130W or under. Xeons go higher.

 

[IntelEnthusiast]

More to do with being a Xeon reject as I see it. The SB-E is more likly chips that did not make the cut to be a Xeon / 8 Core unit. As the quality of production improves we might see a 8 core before the end of life from the SB-E, but intel generally wants to keep to it's price points of $1000 for the top chip. Looking at the Xeon range, it is allowing some to be well over $1000, and those ones have all the trimings (20MB cache, 8 cores ect).

You know it is hard to have a reject for a processor that hasn't been release yet? The Intel® Core™ i7-3960X is the top of the bins from Socket 2011 at this time.

Christian Wood
Intel Enthusiast Team

 

[greenhawk]

Desktop parts needs to be kept at 130W or under. Xeons go higher.

The wattage limit is more being respectable to the market and it's demands. 130W is just the limit selected seeing as even at that level, customers became volcal about the heat / cooling needs.

http://en.wikipedia.org/wiki/List_of_CPU_power_dissipation

Higher end Pentium D - 130W
Core 2 Quad extreams ~ 130W
Core 2 Extreme QX9775 - 150W
all (?) Bloomfield i7s - 130W

AMD appears to be mostly a 125W cap, but an odd ball at 140W is present in the range.

It comes down to what the market segment will accept, and without doing fancy things, getting the heat from the silicon into the enviroment is not a simple task at high wattage levels (small surface area and only having access to one side of the chip).

 

[greenhawk]

You know it is hard to have a reject for a processor that hasn't been release yet?

The Intel® Core™ i7-3960X is the top of the bins from Socket 2011 at this time.

Yes it is hard, but at the same time, the manufacturing process is such that if 8 cores on a future chip passes testing, those would be put aside as inital samples for the Xeon release. At least until the current process is tuned and quality improves. I do not see intel, even with their budget, going out of their way to kill two working cores on a chip to make a cheaper chip for a market segment vs putting that chip aside for a later release.

While I am on the outside looking in to the silicon industry, the testing and grading of silicon is rather straight forward. You tested it, disable sections not up to par, speed test the working sections then disable working parts as needed to fit a pre-determined unit configuration. Then just need to package into a chip and wrap it up for sale.

In short, while it might not be "in production", I would bet the testing/grading is setup to look for them and is bining them accordingly in preperation of having enough to called it "in production".

 

[Idontcare]

Desktop parts needs to be kept at 130W or under. Xeons go higher.

Which just seems so weird to me.

Consider that the guy who is going to drop $1k for an extreme-SKU CPU is probably intending to drop that CPU into a rig that is populated with at least one, and possibly three, 300W GTX580's or HD6990's...whether the CPU is binned to use 130W or 300W is quite superfluous.

 

[Edrick]

Yes it is hard, but at the same time, the manufacturing process is such that if 8 cores on a future chip passes testing, those would be put aside as inital samples for the Xeon release. At least until the current process is tuned and quality improves. I do not see intel, even with their budget, going out of their way to kill two working cores on a chip to make a cheaper chip for a market segment vs putting that chip aside for a later release.

While I am on the outside looking in to the silicon industry, the testing and grading of silicon is rather straight forward. You tested it, disable sections not up to par, speed test the working sections then disable working parts as needed to fit a pre-determined unit configuration. Then just need to package into a chip and wrap it up for sale.

In short, while it might not be "in production", I would bet the testing/grading is setup to look for them and is bining them accordingly in preperation of having enough to called it "in production".

Keep in mind that these are the C1 steppings with broken Vt-d. The C2 steppings will fix this and those will be your Xeons. So yes, these 3960s are top binned as of now and not rejects of the Xeons.

 

[Khato]

The real question is - why offer anything more than 4 cores for a mainstream processor? So that you can have more cores power gated the majority of the time? Are there any 'mainstream' applications other than video encoding that really make use of more thank four cores? Since that usage is likely going to be obsolete anyway with the performance and quality improvements of the Ivy Bridge Quick Sync implementation.

Another interesting query - how long do you expect Intel's mainstream line to stay at a maximum of four cores? Haswell? Broadwell? Skylake? Skymont?

 

[greenhawk]

Keep in mind that these are the C1 steppings with broken Vt-d.

I did not know that. I just assumed it is like all other features of the current cpu market. Features turned on or off after the silicon has been processed to keep manufacturing costs down.

Guess the only thing they can pull out of the current quality binning process is silicon for extream testing (ie: tests that do not care about Vt-D or any other design issue).