Transistor count of Core i7 2nd generation, "Sandy Bridge" processors

[idata]

There are two conflicting information regarding the transistor count of Sandy Bridge CPUs.

216mm2 die size, 4 core, 2MB L3/core, 1.16 billion

http://www.eetimes.com/electronics-news/4213428/Intel-details-Sandy-Bridge-at-ISSCC http://www.eetimes.com/electronics-news/4213428/Intel-details-Sandy-Bridge-at-ISSCC

216mm2 die size, 4 core, 2MB L3/core, 995 million

http://www.tomshardware.com/reviews/sandy-bridge-core-i7-2600k-core-i5-2500k,2833-2.html http://www.tomshardware.com/reviews/sandy-bridge-core-i7-2600k-core-i5-2500k,2833-2.html

Why such a big difference? Why is it the information coming from technical presentations and consumer review sites differ? If they are different chips, what are the differences?

Please, do not counter with obvious answers such as cache sizes or graphics EU sizes are different. I've heard both numbers, and both claim its the exact same chip. I know only Intel knows this.

[idata]

Well, it will be logical to first ask those who presented those conflicting informations for their argument or proof. Present those informations here to enlightened whoever is interested in participating in your question or query.

It's only a suggestion.

[idata]

First, Intel did not include a transistor count in their usual CPU specifications that are available on Intel's web site, in contrast to first generation i7 CPUs, the Core i7-900 series, for example.

Next, since the second generation Core CPU's have an "on die" graphics processor, in contrast to all previous Intel CPU's, is it appropriate to include the transistors in the graphics processor in the total transistor count? You would think it would be simple to separate the two and make note of that. That data must be out there somewhere.

One might think the difference between the two figures is one with and one without the graphics processor. The difference, 21 million transistors, is way to small to account for the graphics processor.

For example, if we were to start with a CPU with 1.0 billion transistors, 10 million transistors is 1% of that number. So 21 million transitors is 2.1% of 1.0 billion, so plus or minus that amount is actually a small difference, IMO.

There is either an error in one of those figures, or the higher figure is including something the other is not.

[idata]

Parsec, where are you getting 21 million transistors from? 1.16 billion - 995 million = 165 million

And again, 1.16 billion and 995 million is supposedly for the same chip, 4 core/12 EU/8MB L3

They use to have the information up at Intel's site but not anymore. BTW it claimed 995 million.

[idata]

"...according to the paper presented at ISSCC Tuesday by Ernest Knoll, a designer at Intel's design center in Haifa, Israel. Sandy Bridge features 1.16 billion transistors and a die size of 216 square millimeters, Knoll said."

It's hard to say who's wrong but I'd be inclined to take the word of one of the designers. The problem here is that the information by eetimes is from a reporter that may have confused the data with the 6-core processor. I couldn't access the white paper nor did I care to register at the site to get it but that would be the place to verify the info. It wouldn't be the first time a reporter got the information wrong.

[idata]

DavidC1, Yikes, you are correct sir!! So much for my decimal point shifting. The 165 million difference amounts to a 14% to 16% difference/.

Here's a table I found from what ought to be a reliable source:

The article that had this table also stated that each SB core had 55 million transistors. Why the one article said 1.16 billion, I don't know, although that's close to a six core Gulftown CPU. IMO, the 995 million figure is correct.

[idata]

Carwiz, the 1.16 billion transistor count is too little for 6 core. Gulftown with 12MB L3 has 1.17 billion transistors. Sandy Bridge E 6 core is said to have 15MB L3. Sandy Bridge E also has extended I/O support.

It's also not the first time I've heard the number.

[idata]

I just used that as an example. We know what the reporter wrote. If you want to know for sure, buy the white paper. I could care less and think the transistor count is tied to marketing. The actual on-die transistor count will vary by design and function even within the same CPU like the 2600 and 2600K. With the Bulldozer coming, Intel needs to leak information without showing their hand. Maybe both numbers are wrong. I'd be more interested in Intel's offering that's slated for the second half of this year. I'm sure it's going to be an answer for the Bulldozer. And I'm certain transistor count won't be part of the comparison except by die designers.

[idata]

It's a incorrect to suggest that 2600 and 2600K would differ in transistor counts. They don't make seperate die for what's a trivial difference. Also, I feel you are suggesting disabled features might be counted differently. Well, that has never been done before. What's physically on the chip whether disabled or not is included in the count.

Unrelated, but part of marketing's job is so consumers understand the significance of the features they are trying to sell, not outright lie.

You don't know the answer to the question, may not care. I'd still like to know.

[idata]

More bits of articles about Sandy Bridge transistor count, which really may not help, but at least reveal the source of one of the figures:

The quad-core desktop Sandy Bridge die clocks in at 995 million transistors. We'll have to wait for Ivy Bridge to break a billion in the mainstream. Encompassed within that transistor count are 114 million transistors dedicated to what Intel now calls Processor Graphics.

SAN FRANCISCO—Intel Corp. disclosed more technical details of its 32-nm Sandy Bridge processor at the International Solid-State Circuits Conference here Tuesday (Feb. 22)... The 32-nm Sandy Bridge processor integrates up to four x86 cores, a power/performance optimized graphic processing unit (GPU) and DDR3 memory and PCI Express controllers on the same die, according to the paper presented at ISSCC Tuesday by Ernest Knoll, a designer at Intel's design center in Haifa, Israel. Sandy Bridge features 1.16 billion transistors and a die size of 216 square millimeters, Knoll said.

Shmuel "Mooly" Eden, vice president and general manager of Intel's PC Client Group, introduced the new chips... Eden said the chip has 1.16 billion transistors on the chip, certainly talking about the quad-core version. Interestingly, his number was slightly different from other numbers I've seen for transistors, which have said the quad-core version has 995 million transistors on a 216 mm2 die, while the dual-core has 624 million transistors on a 149mm2 die.

And finally, a link to an ultimate, all-in-one list of Intel CPU's, including transistor counts:

http://www.techarp.com/showarticle.aspx?artno=337&pgno=10 http://www.techarp.com/showarticle.aspx?artno=337&pgno=10

Will we ever really know? I'm so confused...

[idata]

Parsec, I've e-mailed EETimes hoping for their response. I've asked them to ask Intel about it.

I've personally asked TechArp to correct the specifications on few of the products. Some of them they just have to take a guess as official information is hard to come by. Though specifically regarding this I'm pretty sure they have gotten the same number as the review sites did.

[idata]

Yes David, this is all kinda weird. Weird in that why would Intel not include the transistor count spec in the official CPU information page on their web site? If it is a little below or above one billion, either is a huge accomplishment IMO. If it was actually less than say an i7-900 series four-core CPU, which are 731 million, so what? Better overall performance with less transistors would be a great accomplishment as well. Or due to the Western culture's paradigm of more and bigger always meaning better, would that be perceived as not better? Is there a marketing concern over that? I could understand that, as the semi-informed would always consider more as better.

Another factor could be the inclusion of the on-CPU graphics processor, whose transistors are not part of the regular CPU functioning. Are the graphics core transistors part of the count or not? I'm sure they are, but what does all this nit-picking really do for us? Nothing IMO.

Can you imagine if the reason for not including the transistor count as 995 million is so the Intel executives that stated it as 1.16 billion are not embarrassed? Frankly, I could care less, blame an engineer for providing an incorrect figure, done and over.

Don't get me wrong here David, I like all the specs too, but does it really matter all that much?

[idata]

It's a incorrect to suggest that 2600 and 2600K would differ in transistor counts. They don't make separate die for what's a trivial difference.

For your information, I'd like to direct your attention to the i3-2105 and the i3-2120. These are two processors in the Sandy Bridge line and the same series that have a difference of 120 million transistors. On the surface, the difference is HD-3000 vs. HD-2000. They have like specs. This is repeated throughout Intel's processor history so in my opinion, it's even more incorrect to suggest that Intel is locked into a specific die, TC or size for any series. Even for a reason as "trivial" as graphics presentation as you seem to think.

Intel has admitted publicly (if you can read between the lines) that they add un-necessary circuits to dies and that makes transistor count trivial in itself. To wit, the SATAII problem. The problem was caused by un-necessary circuits.

[idata]

CarWiz, Apparently David was not aware of the difference in the graphics core between the i7-2600 and i7-2600k. Your point brings up an interesting question, IMO.

That is, how is a CPU constructed as a whole, out of individual pieces, or is the die (if that is even the correct term) unique to each processor. Common sense tells me that when working with "parts" (processing cores, cache, IMC, graphics core, etc) as small as these are, I wouldn't think they could be individually manipulated, even by machines. For many other reasons that come to mind, I would think each CPU model must have it's own die.

I am not sure if the the "unlocked" feature of the 'k' models something that is set or enabled during the CPU creation process, or is part of the architecture of the CPU itself.

To nitpick your mention of the Cougar Point chipset issue, we are talking CPUs here, not chipsets. I wonder how many extra, unused transistors exist in a CPU, which should not be included in a count, IMO. I doubt that the unused ones are added for any reason beyond they simply occur when working with something as tiny as the dies themselves are. Given that we are dealing with tens and hundreds of millions of transistors, what's a few thousand extra here and there? These numbers are basically meaningless, since how would you verify that or count them?

[idata]

Apparently David was not aware of the difference in the graphics core between the i7-2600 and i7-2600k. Your point brings up an interesting question, IMO.Umm, I did say I knew it, though the way it was conveyed might have been indirect. Like I said, both numbers are for the 4 core, 12EU, 8MB L3 version.