Slashdot Mirror
Why Intel Leads the World In Semiconductor Manufacturing
MrSeb writes "When Intel launched Ivy Bridge last week, it didn't just release a new CPU — it set a new record. By launching 22nm parts at a time when its competitors (TSMC and GlobalFoundries) are still ramping their own 32/28nm designs, Intel gave notice that it's now running a full process node ahead of the rest of the semiconductor industry. That's an unprecedented gap and a fairly recent development; the company only began pulling away from the rest of the industry in 2006, when it launched 65nm. With the help of Mark Bohr, Senior Intel Fellow and the Director of Process Architecture and Integration, this article explains how Intel has managed to pull so far ahead."
Andy Grove paid billions to get access to Area 51 alien technology back in 1998. What's so hard to understand?
Apple is a product company. It designs its products, and then someone else makes it.. Many components like the processor are third party, and companies like APPLE design a system around it.
After that the design goes to samsung, and its manufactured by samsung. I think samsung uses TSMC fab.
So if apple wanted to have a 22nm chip it could
1. Build a Fab(invest many billions)
2. Pay TSMC and partner with them in tech (invest some billions).
Return on investment may not justify the cost.
As you go smaller, you do gain an area and cost advantage, but you also run into lot of issues related to physics. So 28->22nm is not easy, and its really commendable Intel has done it.
My Aurora : http://www.youtube.com/watch?v=o91ZsGwJYyg
FB : https://www.facebook.com/TanveersPhotography
Intel, with their open-source graphics stack, makes for some of the easiest-to-maintain Linux boxes around. I'm typing this right now on Arch with Intel graphics. Sure, they don't have a lot of "gaming punch" but they are darn stable and just work with Linux.
My desktop right now has Windows and is running a first-generation Core i5 with an AMD Radeon 6870 added in. When that machine get's replaced with another gaming Windows machine in a year or two I'll be pulling the AMD graphics out of it and running on the i5 integrated Intel graphics. It will be super-low-maintenance in Linux. None of this rebuilding fglrx or nVidia modules every time you upgrade the kernel.
When I go looking for a Linux machine the very first thing I look to check-off is "Intel graphics"? Yup, then it's a buy.
Spying... on their competitors who are all years behind them?
You must be pretty high up in the CIA to have thought of such a genius spying scheme.
Let's see. How come GM doesn't have these tires factories that Firestone and Michelin has? I wonder... GM is making a heck of a lot more money than Firestone though...
Write boring code, not shiny code!
Is this true or just trolling?
Every benchmark I've seen so far has shown performance increases and power consumption decreases at around the same price.
If your statement is true then that suggests a lot of review sites out there are spoofing their results and that's very very bad.
Sure, if you have a sandy bridge chip there isn't a whole lot of reason to upgrade because for most users anything made in the last 5 years or so can handle everything you'd want to throw at it.
That's in part because they put the extra die space freed up to a new purpose: Graphics performance. If you just look at processor performance, Ivy is no better. Benchmark the inbuilt graphics and it's far ahead.
Of course, anyone who actually needs decent graphics wouldn't be using the on-chip graphics anyway, so I question just how useful this really is.
Traditionally, Intel has always been able to show lower power consumption and more than a tangible performance improvement when just doing a process shrink, but the Ivy Bridge does nothing extra in terms of performance and consumes not lower power than its older 32nm sibling
Is there any reason the parent is at +4, Interesting and not -1, Troll? Are the AMD fanbois really so desperate that they have to mod up blatant lies? Ivy Bridge uses 25-30W lower power at stock speed to deliver marginally better than SB CPU performance and considerably better (but still crappy) GPU performance. The only people that whine are those who want a 4.5+ GHz overclock. Anandtech called it quite possibly the strongest tick [Intel] has ever put forth, but I guess if you don't like reality you can invent your own.
Live today, because you never know what tomorrow brings
Most of the complaints are tiered towards the overclocking space; doesn't go as far as the old sandy bridge.
http://www.overclockers.com/intel-i7-3770k-ivy-bridge-cpu-review /
http://www.overclockers.com/ivy-bridge-temperatures
http://www.anandtech.com/show/5771/the-intel-ivy-bridge-core-i7-3770k-review /
http://www.anandtech.com/show/5763/undervolting-and-overclocking-on-ivy-bridge
etc.
It's true, more or less.
That's just for overclocking. Regular usage is improved.
Help I am stuck in a signature factory!
And you seem to have missed the part where "running hotter than SandyBridge" applies only to overclocking. Yes, IB is a worse overclocker than SB, but under normal conditions IvyBridge is faster and uses less power than SandyBridge. Remember that overclockers are a tiny portion of the market. IvyBridge isn't the amazing revolutionary chip some people were expecting but it is a successful, evolutionary step forward. Just like most processor generations.
The shrink from 22 to 32nm is a staggering size change - 33% finer lithography - and it uses their much-hyped 3D transistor technology on top of things. Yet, Ivy Bridge, being just a shrink of the older Sandy Bridge die, shows no improvements over the 32nm version. Traditionally, Intel has always been able to show lower power consumption and more than a tangible performance improvement when just doing a process shrink, but the Ivy Bridge does nothing extra in terms of performance and consumes not lower power than its older 32nm sibling - and let's not mention the inefficient heat packaging causing temperatures hotter than the 32nm Sandy Bridge. There's a problem here, Intel.
While I will accept you reversed some numbers (the shrink was from 32 to 22, not the other way around) and Intel is using tri-gate transistors, most everything else you describe is just flat out wrong. Ivy Bridge DOES show lower power consumption at stock voltages (TDPs of 77W vs 95W are a testament to that), and it is higher performance at the lower power consumption (though not by huge amounts, nor was it intended to be). Since it is lower power than Sandy Bridge at the same frequency, it is not having any issues related to thermals and packaging.
Now, if you want to rant about the fact that it doesn't handle overvoltage well for overclocking purposes, that is fine, but it is a separate discussion compared to stock. What you are seeing now is that Intel (probably extremely wisely for the market they are chasing most heavily) has tuned in their process node for stock voltages, but this is resulting in very leaky transistors at high voltages. Additionally, while the current packaging has the ability to remove heat just fine at stock voltages, when you start leaking too much the heat builds up too quickly- which certainly is a 22nm node issue and not actually a packaging issue. Quite possibly, though how far in the future I can't begin to guess, they will probably tweak the process for the Extreme Edition CPUS to make them handle an overclock without leaking so much, but that will take some time learning how they can play with the various knobs to get what they want without destroying what they need.
This leaves me with the feeling that the only problem here is your expectations of a CPU that was manufactured with the intent of taking the mobile market by storm (and they have tuned the process properly for that) when what you want is an overclocking king. Let's see how they tune the process technology for the Extreme Edition (and hopefully copy into other desktop-bound CPUS) before any decisions are made that they have screwed the pooch on being able to overclock.
why the hell did this get marked intersting?
1. if it was "just" a shrink over 32nm and had the same die size and power consumption, it would be no improvement. instead the die size is quite smaller than 32nm. past shrinks have held the die size only slightly smaller as they either added more cache or other logic in addition to the shirnk. in this turn things are mostly the same resulting in a smaller die. this means more dies per wafer and, eventually, lower cost (noting a move to 450mm wafers within the next 5 years).
2. ivy bridge does have lower power consumption than sandy bridge. same speed chips doing the same task can have the ivy bridge system using up to 30w less power at peak performance. where it hasn't really improved is in 'idle' conditions. why? ecause most tech to turn of various sections of the cpu were introduced in other process shrinks. as this is gennerally "just" a shrink, the idle power being roughly the same isn't about lack of improvement, but moreover less innovation in existing silicon power saving techniques. i mean the chip already slows it's overall speed down when idling, and will shutdown significant portions of it cache as well.
3. there has only been a performance improvement for various definitions of "performance". in other iterations intel has tweaked the chip. this is based on feedback from the chips being in the field. of course with sandy bridge running so effectively, there hasn't been much call to 'tweak' the design. still at the same speed, ivy bridge has better "power" performance than sandy bridge.
4. i don't think you've seen any benchmarks to be honest. graphics has improved significantly, cpu not so much. but it is still an improvement over snady bridge.
5. the packaging problems have only been encountered when people overclock the chip. that is, people running it out of spec. sure there's 1% of us out there that will push the limits. but for most people, they will never ever see those hotter temperatures. personally i go the opposite direction by undervolting to consumer less power and in turn producing less noise.
It is being punished as a troll, because it is wrong in annoyingly misleading ways. It probably would not have been punished as a troll if the GP had said "in the area of overclocking, 22nm is a fail because..." By making a sweeping comment that only applies to a very small subset of the market, potentially informative because troll. The mods got it right.
TI closed a lot of FABS. All TI designed stuff does not exclusively get manufactured in TI fabs. Much of it actually goes to TSMC
My Aurora : http://www.youtube.com/watch?v=o91ZsGwJYyg
FB : https://www.facebook.com/TanveersPhotography
Of course, anyone who actually needs decent graphics wouldn't be using the on-chip graphics anyway, so I question just how useful this really is.
There's a whole world of people who would quite like decent graphics, but who don't want to spring another hundred bucks or two to get something fancy. There's also the mobile market (laptops, tablets, etc.) where fitting an extra graphics card looks more like a liability than a good thing. Overall, it looks to me like a smart area for Intel to pitch their transistor budget at.
"Little does he know, but there is no 'I' in 'Idiot'!"
I think you've completely missed the point.
1. Ivybridge is a die shrink, nothing more nothing less. Everybody who really thought it would be lightyears ahead of Sandybridge in terms of performance was simply deluded. It's a new die size that anybody has yet to perfect, that will come in Haswell.
2. All this ire is unfairly directed at Intel. On the basis that AMD seems to have no idea what it's doing at the moment, Intel can relax and do as they please. If you want to be pissed at anybody, be pissed at AMD for not being anywhere near competitive and pushing Intel to continuously raise their game.
Since 1998 I've only ever used AMD CPUs in my builds. When I came to build a new rig in February, I simply couldn't justify buying AMD for my CPU again because they were so far behind and there was zero indication that Bulldozer would rectify that. It's sad watching them busily engage in killing themselves off as a serious desktop CPU manufacturer and leaving Intel to potentially become lazy and overpriced, but that isn't Intel's fault.
> but that isn't Intel's fault.
Actually it is, to at least some extent. Go back a few years, when Intel was making misstep after misstep, and AMD was coming on gangbusters with K8. At that point, Intel had missed the market so badly that had they been AMD they would have gone under. They weren't AMD, they were Chipzilla. AMD enjoyed a good product cycle with K8, until Intel managed to come back. But they didn't enjoy the great product cycle they should have. Their great product cycle was turned into a merely good product cycle because Chipzilla twisted a few arms and kept K8 out of key opportunities.
The other piece of reality is that Intel combines first-rate process technology with first-rate design capability. (I say "capability" because more than once they've shown themselves to be very capable of letting their eye off the ball, design-wise.)
AMDs biggest problems have always been financing and less-than-best process technology. Bulldozer is a misstep, agreed. But it's not a misstep of the degree of Netburst or IA64. Had K8 gotten the success it deserved, AMD would have been better able to properly fund their design shop. That wouldn't have helped their process problems, however.
The simple fact is that the way things are today, Intel can afford to screw up badly, and can recover. None of their competitors can.
The living have better things to do than to continue hating the dead.
pushing their own inferior architecture and holding everyone back
Oh brother, not this again.
On the low end, Intel will never beat ARM because of the large, expensive instruction decoder. That applies to the deep embedded stuff.
Cellphone chips aren't low-end any more. They're getting bigger and bigger and bigger. For big processors, intel does very well, as they have the best OoO scheduling and branch prediction which keeps the large, expensive ALUs busy, giving a very high IPC.
In 5 years time, ARM will need to have similar guts if they want anything but lots of weak cores. At that point, the main advantage that they have will be more or less down in the noise.
SJW n. One who posts facts.
Overrated I would think.
There needs to be a factually wrong mod. The lack of one causes a lot of problems.
Two words: Tick Tock.
Intel's development model is Tick (die shrink), Tock (new features). It's been this way for many years. Honestly, I'm not sure why you expected a Tick to have any new features. They did call Ivy Bridge a "tick plus," but even then I wouldn't expect any major overhaul in features or performance. Tick is a manufacturing process improvement, not an architecture improvement.
As far as heat packaging, I believe others have covered that sufficiently.
The road to tyranny has always been paved with claims of necessity.
I'm not sure about GM but I know Ford is no longer casting around here, they recently started tearing down the 60+ year old casting plant in Brooke Park. The reasons for the plants demise is that it's an ironworks and Ford basically doesn't use a cast iron block any longer and they don't view block casting as a core area and so there was no way they were going to invest the massive amount of capital it would have taken to move the plant over the casting aluminium.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
"Consider the Auto analogy."
No.
The Kruger Dunning explains most post on
"How the processor performs when overclocked (or really, not factory-underclocked) provides an extremely revealing look at business practices and yields."
It's an interesting look to be sure, but considering the IB chips perform fine and use less power than SB at their marked clock speeds, there's nothing you can derive from that look to back up "the Ivy Bridge does nothing extra in terms of performance and consumes not lower power than its older 32nm sibling" in the post you appear to be defending here...
0 1 - just my two bits
So the original Athlon was a shot out of the blue, it was the first AMD chip that really competed with Intel chips. Intel had to stop sadbagging and release faster P3 chips (it was capable of making them just wasn't because it didn't need to). AMD legitimately brought some serious competition. It was badly hamstrung by having horrible, horrible motherboard chipsets, but there you go.
Now the Athlon maintained competitiveness the next generation... Because Intel fucked up. Their Netburst architecture wasn't very good. I don't fault Intel on this, their research showed it would scale really well MHz wise, possibly up to 10GHz, so the slower IPC wouldn't matter. However it didn't, so they had a slower architecture compared to AMD. The problem? AMD wasn't updating. They just kept doing minor rehashed on the same thing.
Then, as you say, Intel dropped Core. They hadn't been standing still, they never do. They corrected the mistakes of Netburst and made a chip that was very fast per clock. AMD was still playing with old tech and Intel pulled way ahead. Then even worse as it continued, Intel kept revising their chip, AMD kept playing with the same basic thing. Their Bulldozer launch got pushed back and back. When it finally did happen recently, it was not at all competitive to Sandy Bridge, and of course Intel now just launched Ivy Bridge.
So AMD's initial competitiveness was no fluke, they dropped a good product. But the length it went on was kinda a fluke, since Intel screwed up, and AMD didn't do anything to work on improving their tech in a big way.
I work at TI. We do have our own fabs, but we also outsource manufacturing to foundries too. The new 65nm flash process I work on was developed at TSMC, and all manufacturing will be done there. I know that other processes run in TSMC as well, but I'm not sure which ones (we have a lot).
Visit the