Intel goes FinFET on 22nm – somewhat
Lots of cache to burn
Apr 7, 2011 in Chips, Microprocessors, Rumors
There is a lot of speculation about Intel’s (INTC) upcoming 22nm process, and we finally have some hard information to share. The first product on the process, Ivy Bridge, will have other tricks too, Intel is rolling out a lot at once, but doing it carefully.
When we first broke the news about Ivy Bridge and its use of silicon interposers for on chip memory stacks, a lot of people did not understand the changes it brings. Not only will it change what a CPU is, it also can be used selectively. There are also huge risks, yield, and therefore cost being the most notable, but lifespan could also be a potential gotcha.
Intel is doing the smart thing here and putting the stacked memory only on some models, most likely the high end SKUs. This allows them to charge a premium for the parts, alleviating the high cost while bugs are worked out. If it screws up totally, they can pretend it never happened and launch with Haswell later on.
On 22nm, Intel is taking somewhat of a different approach, they are using FinFETs, but only in the cache. You can’t modify a chip enough to change transistor types once it ships, but this says to SemiAccurate’s blue ribbon panel of semiconductor experts and chimney sweeps that there was definitely a plan B up until the last minute, and one that could swapped out in a hurry.
All 22nm Intel chips will definitely have FinFETs as an option, which products they are used on is likely to be determined by how well the technology works out in practice. Regular arrays of transistors that can be replaced without touching the core is unquestionably the right way to do things, and also gives us a very unique view into Intel’s thought process.S|A
32 Responses to “Intel goes FinFET on 22nm – somewhat”
Trackbacks/Pingbacks
-
-
Apr 28, 2011
[...] [...]
-
-
Apr 26, 2011
[...] SemiAccurate. Tweet [...]
Hello,
I think that the Intel Sandy bridge IGP isn’t that bad but it isn’t that good either. AMD mobile Llano has got the power consumption down and it has a GPU that in some cases is about 2X what Intel has. I like ARM cpus but they are neither fast enough nor run the x86 video games I like so much. Even at 22nm Intel is going to get PWN3D by AMD Trinity. 50% to 70% Lower power consumption isn’t going to help Intel much because the low power cpus are probably going to be quad cores which will take away that advantage. On a side note Terga 2 is way too slow and Terga 3 is better but power consumption is going to be too high. If ARM doesn’t move faster then Intel might catch up and AMD will beat it.
IN 2013/2014:
I think AMD will be the low power company
Intel will better performance but will always be a high performance/High power cpu company. I don’t think their horrible ULTRABOOK!!! will sell very well. ARM will become the wildcard because I think there is a good chance they will announce a new CPU for 2014 which will offer about 50 to 70% performance clock for clock and comparable performance per SOC.
Hmm, as for totally off topic there seems to also been rumors floating around of I7-980 standard model to show up on June and the next sandy bridge model I7-2700K.
Good things comes first considering Intel trying something new next they should fix the existing first.
Actually, I’m following the style of an anonymous commenter in a Roborat’s blog, Aimed Corporation.
¿Como se WTF, senõr troll?
The Intel tax is the die size, extra power consumption, IP cost and code cost that the massive block of silicon that translates x86/x64 to RISC implies. ARM lacks this ‘tax’ and thus will put Intel CPUs into the same grave occupied by other once dominant designs, like the z80 and 68000.
All Intel can do is fight (and lose) the law of diminishing returns- made worse by the fact that everything Intel touches except the CPU is a complete failure. Intel needs an on-chip GPU that is not garbage, and the only way it can do that is to buy Nvidia, or steal ATI’s designs under the excuse of ‘patent sharing’.
Meantime, improvements to ARM SoC designs are running well ahead of Moore’s Law, including GPU performance.
Multi-core won’t save Intel, because that moves the advantage even more strongly in ARM’s direction. Worse, traditional PC programmers are too thick to program to multiple cores, quoting garbage like “Amdahl’s Law”- a ‘law’ that strangely doesn’t inhibit PS3 and x360 game programmers from gaining near 100% use of all their cores, in complex mixed function code tasks.
Intel still pushes single thread performance, and intends to continue paying shill sites to benchmark primarily in this direction. ARM programmers will increasingly extract every drop of possible performance from their multiple cores. With no Intel tax to suck power, money and performance, ARM has a perpetual ‘downhill’ advantage.
Let’s stay on topic.
The discussion is about Intel and AMD CPU architecture and not the PS3 Cell Processor. Thes ASF proposal is about running multiple concurrent applications, each maximizing core resources.
The PS3 and Xbox run one programme (game) at a time.
If you are interested in the Cell processor, here is a link:
hxxp://www.blachford.info/computer/Cell/Cell0_v2.html
The XBOX uses the PowerPC core.
ARM performance is still crap compared to x86. Good luck trying to scale that up without losing power efficiency.
Also, ARM will always be behind Intel because ARM CPUs are made with what can be considered “previous generation” semiconductor process compared to Intel’s. This won’t change – Intel will always be ahead of others in process technology.
ARM is a small design house. Do you REALLY think that they can beat Intel in R&D? Previously Intel didn’t care about low-performance parts like cell phone chips. Now that they do, ARM is toast – ARM has zero chance of competing with Intel’s massive, well-funded R&D department.
By the way, ARM “tax” is the license fee; something that ARM in their current monopoly is increasing on a whim:
http://www.eetimes.com/electronics-products/processors/4114549/ARM-raises-royalty-rate-with-Cortex
We should all be happy that Intel is entering this market and ending ARM’s monopoly.
Right, and that’s why Intel has so freaking many design wins for cellphones… and that’s why their lowest TDP Atom4Cellphones CPU runs at 300mhz@4w, and still can’t compete with any ARM CPU on the market on speed, TDP and performance-per-watt.
Ladies and gentleman, from the company that gave us Itanium, Pentium Pro, and Pentium 4… and failed to give us terascale, larrabee, or any decent graphics chip(unless your standards are so low that you actually consider Sandy Bridge decent), I give you: another product that can’t possible fail, because Intel is so big, rich, and L337.
AMD needs to counter this tech with a new announcement or two, like the following one, proposed by the company a while back:
ASF is an x86 instruction extension for the AMD64 architecture. Basically, its objective is to simplify the synchronization of concurrent programmes.
Why ASF? Well, multiple cores are here to stay and a way is needed to maximize their use, by running applications designed to use all available cores (limited by Amdahl’s Law).
Read paper at:
hxxp://www.amd64.org/fileadmin/user_upload/pub/eurosys10-full_stack.pdf
@maddoctor
Unfortunately there are some people soo stupid they don’t even know the basics of economics or actually anything, and they make stupid statements on several fora. I actually laugh when I occasionally see such stupid people commenting.
I guess Intel will maybe not release this chip with FinFET, if they can’t get the chipset even right (p6x), I doubt they will find all the bugs on time with these upcoming chips.
Maddoctor is moron. Look what happened before to Nvidia cards when ATi cannot compete, $800+ is for highend cards and some reaches $1000+ for super rare version.
Now you can buy $700+ for dual card.
That’s what competition could bring, price and innovation. They now develop faster, almost less than a year new generation cards will pop up because of this.
Same goes to processors. If you have the right mind you can think of that.
Only FinFET on the cache? Sounds like Intel is having sever problems with their 22nm process.
*crocodile tears*
Oh yes, I sure would love to pay $1000 for Core i3 level processors!
Just totally looking forward to that one.
Real funny. How old are you? Seven? Get real.
Bye-bye AMD, Ivy Bridge will be a final nail in AMD’s coffin. I expect in 2012, we will not see any AMD’s crap in the market.
Obviously where you live Maddoctor, the world is ending in 2012 anyway.
I sure hope it starts with you.
obviously, blu tossers are a.n.g.r.y.
Just let Doc die alone in a cold room of halucinations
This will be another -10% to +10% performance gain refresh like Wolfdale was, or like Sandy Bridge was, for that matter. ARM, Ubuntu and Android are going to swallow Intel’s laptop aspirations, their graphics and phone CPUs are going to continue to flop, and Fusion/GPGPU are going to overtake Intel in HPC.
Not that your mid-2000 era Celeron box can beat any current AMD CPUs anyways. Maybe when you turn 16, you can get a job at your local fast food joint and start saving up for some Intel craps.
You didn’t choose the best example to rebut maddoctor I have to say. I don’t actually know why you actually feel the need try. Well, anyway:
Wolfdale/Yorkfield kicked AMD down when it was already reeling. It was sweetly priced (3 GHz E8400 @ 150eur, soon followed by a 2.5 GHz E5200 @ 50 eur, with cheap s775 motherboards too), gave Intel an overwhelming performance advantage in 2007/08, and a serious power consumption advantage compared to 65nm Conroe (4 core Yorkfield ~= 2 core Conroe).
Intel would love for Ivy Bridge to have a Wolfdale-like advantage, but it seems they are dawdling a bit these days. Unlike their high point of 2007/08.
Yeah, so we got +/- 10% performance increase, slightly lower TDP, and a small increase in clockspeed, easily matched by a little OC on Conroe. Furthermore, there was no increase whatsoever in overclocking potential in that generation, or any generation since (unless you’re dumb enough to believe that any SB OCs to 4.5 to 5ghz, and will actually last at that frequency). Wolfdale and Conroe both maxed out at 4ghz OC, and had roughly identical performance, get over it.
Tick-Tock is more about marketing than actual progress, it’s meant to create the perception that performance is still progressing, even though ways of increasing IPC and performance-per-watt in x86 have just about dried up.
@mato: I still have an E5200 @4GHz on air. :D
@sp33d_bumpp: The difference in power consumption was substantial. Q6600 had a TDP of 105W. The Q9650, clocked 25% higher, and with 50% more L2 had a TDP of 95W.
Many boutiques are selling SB’s factory OC’d at 4.5GHz. They must be sure it’ll last at those clocks.
@Maddoctor…Thank god for AMD. If they weren’t around, you would be paying $500 for a low end Intel processor. Pull your head out of your arse and stop being a fanboy.
We can always count on Maddoctor to come along and quickly point out a quick “LOL@AMD UR GOIN DOWN” moment. I mean, it’s like he has TEH DOWNS SYNDRUM LOLOLOL.
Seriously man, you’re a broken record. Go fix yourself.
Bring back the drashek, or even amanfrommars. At least they’re mildly entertaining. This maddoctor is a one-trick-pony.
So if Intel is doing so well why did they release the head of their Atom products line? Most internet connectivity in five years will be tablets, netbooks and mobile phones and desktop cpus will be a small fraction of consumer sales. AMD c50 fusion APU’s draw the same power as an n550 Atom but quadruple the gpu chops. And Windows and Ubuntu are supporting ARM now and in the future so although Intel will remain relevant, claiming their supremacy and dooming AMD is beyond short sited and uniformed…
Bye-bye INTEL, BULLDOZER will be a final nail in INTEL’s coffin. I expect in 2012, we will not see any INTEL’s crap in the market.
BWAHAHAHAHA! Do you seriously think your magical Bulldozer (which you AMD fans have been clinging to for the past years) is going to beat Intel? You have been seriously misinformed. What makes you think AMD is going to magically defeat Intel with Bulldozer? Why do you think it has been delayed so much?
k, this is Intels “COME AT ME, BRO” moment. So. GF & AMD, show us what you got. Please? :3
This is a guaranteed mess. 22nm is where intel finally lost the one advantage they had left.
How is any sort of mess? I’m not understanding this. Are you a stupid AMD faithful? Still clinging to the hopes of Bulldozer? AMD sucks. Get over it. Intel won.