Physics, Ivy Bridge, and the slow death of overclocking

In the weeks since Ivy Bridge launched, it’s come out that Intel used thermal paste between the CPU’s heat spreader and the actual die, rather than the fluxless solder it debuted with Prescott and adopted for subsequent CPUs. This, combined with evidence that IVB heats up very quickly when overclocked, has given rise to much wailing and gnashing of teeth from certain parts of the enthusiast community, despiteconflicting evidence on whether or not removing the heat spreader actually makes a difference.

Lurking behind the question of whether or not removing the heat spreader matters (and it’s perfectly reasonable to think that it at least could make a difference) is an unhappy truth: Overclocking is going away, and not because Intel chose goo over solder this time around. The problem is systemic; an outgrowth of the fact that while Moore’s law still works, Dennard scaling — the rule that said smaller transistors would use proportionally less power — began breaking down years ago.

To get an idea of the root of the problem, consider the transistor density of Nehalem, Sandy Bridge, and Ivy Bridge.

It speaks to Intel’s manufacturing prowess that the company has managed to scale transistor density the way it has while simultaneously reducing TDP at stock speeds, but increased density encourages the formation of hot spots across the die. The relationship is proportional — the smaller the die, the less surface area each component occupies. Smaller surfaces mean less area in contact with the heat spreader. There’s no simple way to “fix” the fact that hot spots are getting hotter as die surfaces shrink. The other factor working against Ivy Bridge is that, as process nodes shrink, the amount of resistance (heat) generated at a given voltage also rises. Increasing the voltage to reach higher clock speeds only exacerbates this trend. This drives core temperatures sharply upwards.

It’s a long-established fact that CPUs built on smaller processes require less voltage and respond more sharply to smaller increases, but the difference between Nehalem at 45nm and Ivy Bridge at 22nm is striking. Our original plan was to compare the relationship between CPU voltage, power consumption, and frequency across Nehalem (45nm), Sandy Bridge (32nm) and Ivy Bridge (22nm). Unfortunately, unanticipated technical problems intervened. As a result, we’ve been forced to merge our own Nehalem data with tests run by Anandtech (AT) and Tech Report (TR), and we’ve confined the comparison to Nehalem and IVB. While this means that our data is no longer strictly controlled, we have faith in the measurements of the other two sites, and the difference between the two isn’t subtle.

Our Nehalem system was built using MSI’s Big Bang motherboard; an enthusiast X58 design that featured lower power consumption and strong overclocking features. We used a low-end Radeon 5750 and just 2GB of RAM to minimize power consumption and reduce the impact of non-CPU components when comparing across product generations.

According to data from AT and TR, the (Ivy Bridge) Core i7-3770K draws ~120W at its stock speed of 3.5GHz. That’s a considerable improvement over our (Nehalem) Core i7-920, which drew 161W at full load. At 4.6GHz, IVB’s power consumption has nearly doubled, to 204W. At Tech Report’s high of 4.9GHz, the chip’s power consumption has risen to 236W.

Compare Ivy Bridge against Nehalem when we normalize the data sets to show proportional increases.

In the chart’s x-axis, the 40% refers to IVB, the 53% refers to Nehalem. This is less exact than we wanted, but the best-fit line we were able to build given disparate data sets. At 4GHz — an overclock just a bit above 50% — our i7-920 drew ~275W. At 4.9GHz, Tech Report’s Ivy Bridge drew 236W.

Focusing on wattage, rather than temperature, paints a clearer picture of how Ivy Bridge’s increased thermal density plays out in real life. Focusing on the chip’s thermal paste obscures the larger trends. With bus-based overclocking having largely gone the way of the dgodo and AMD unable to offer an enthusiast challenge to Intel, the days of buying a low-end chip and ramping the clock 30-50% to compensate are well and truly gone. Intel’s desktop products are now largely differentiated by core count, Hyper-Threading, and cache sizes rather than clock speed.

No turning back the clock

When we asked Intel for comment on this situation some weeks ago, the company sent us the following: “We are using a different package thermal technology on 3rd Generation Intel Core desktop processors (Ivy Bridge). Coupled with the higher thermal density of the 22nm process shrink, users may observe higher operating temperatures when overclocking.” Unofficially, company reps stated that the higher thermals really only affect air overclockers, and that super OCers are thrilled with IVB.

We note this last because it’s an unintentional hilarity. A little off-the-cuff research indicates that it takes a home-built, small-scale LN2 machine roughly 400W of power to create a single liter of LN2. I don’t know how many liters of LN2 an overclocker consumes per hour, but when you go to OC competitions, you’ll see liquid nitrogen being rolled in by the tankful — and the tanks aren’t small.

The good news, I suppose, is that the laws of physics still work both ways. If you’re willing to invest the enormous amounts of energy required to cool a CPU down to a healthy percentage of the temperature of deep space, you can still hit big overclocking numbers. Those of us without access to a thermal engine with a five-digit price tag, unfortunately, are out of luck.

The worse news, for overclocking enthusiasts, is that this isn’t going to change. True, removing the shim and reapplying better paste might win you a few hundred megahertz, but nothing you spread on top of the chip is going to fundamentally alter the slope of the voltage/power consumption curve. Moving from, say, a stable 4.8GHz to 5.2GHz, meanwhile, buys you just 8.3%. Sure, it’s faster — but it’s not the kind of jump OC dreams are made of.

Will Haswell change things? It’s not impossible, but it’s also not where Intel’s energy is focused. All of Intel’s manufacturing prowess is turned towards building smaller chips that draw less power, not ultra-high-performing chips that sacrifice efficiency to hit skyrocketing performance targets. Blade servers and virtualization have created a market where small, efficient chips are more important than single-threaded monsters. Multi-core optimizations and many-core pushes like OpenCL, CUDA, and GPGPU have changed the focus of the industry as well.

Single-threaded performance has gone from a cornucopia of plenty to a scarce resource, and Intel guards that resource jealously. Since the launch of Nehalem four years ago, it has parceled out the frequency gains slowly, knowing that to do otherwise risks the market perception that the company has run into fundamental limits that could damage its market value. Haswell’s new architecture and the slow maturation of 22nm might enable a few extra clock grades, but cutting the thermal cap off Ivy Bridge isn’t going to bring back the Good Old Days of overclocking when you could buy a Duron 600 or 0.13 micron Tualatin P3 and expect a 60-70% overclock as a matter of course. The frequency margins on cheap, air-based overclocking are shrinking and short of a manufacturing miracle, that’s not going to change.

Source

Deliberate excellence: Why Intel leads the world in semiconductor manufacturing

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.

We recently had an opportunity to talk with Mark Bohr, Senior Intel Fellow and the Director of Process Architecture and Integration at the company’s engineering fabs in Hillsboro, Oregon. We asked him to explain how Intel had managed to keep its two-year tick-tock cadence and why Santa Clara’s manufacturing is regarded as the best in the world.

Bohr attributes Intel’s success to several factors. First, Intel is virtually the only IDM (Integrated Device Manufacturer) left in the microprocessor business. Even companies like Samsung and IBM, which still handle a significant amount of their own product manufacturing, have teamed up with GlobalFoundries to jointly focus on R&D. The rest, like Qualcomm, Nvidia, Toshiba, and Texas Instruments, outsource their manufacturing to companies like TSMC, UMC, and GlobalFoundries.

Because it manufacturers all its own hardware, Intel avoids the conflict of interest that inherently exists in any foundry/customer relationship. The increasing difficulty and cost of transitioning to new nodes has amplified tensions between the two groups that were previously kept to a minimum thanks to the combined effect of Moore’s law and Dennard scaling. Moore’s law states that transistor density doubles every 18-24 months; Dennard scaling predicted a proportional, linear relationship between the size of a transistor and its voltage/current draw. Smaller transistors, in other words, draw less power.

Node transitions were still difficult and occasionally rocky, but the end results were fundamentally predictable. Higher short-term costs and greater defect densities would be more than offset as the new node came online and yields improved. That underlying predictability is what made the pure-play foundry model work. In its absence, we’ve seen customers like Nvidia pushing for new agreements in which custom IP design, R&D costs, and risk production expenses are shared more equitably between foundry and customer.

Bohr didn’t specifically distinguish between small, group-level collaboration and the large-scale sharing of information between different sections of the company, but his comments made it clear that design and implementation are treated as a joint effort at every level — including when things go wrong. This reduces the chance of an “us vs. them” mentality developing between groups of engineers and encourages further collaboration to solve the problems that do occur, rather than circling the wagons and going into CYA mode.

Copy Exactly! and the tick-tock cadence

“Copy Exactly!” is Intel’s method for duplicating successful chip designs across its various factories worldwide. The work begins at the company’s engineering fabs — D1C, D1D, and the upcoming D1X, currently scheduled to begin production in 2013. Copy Exactly!, which Intel began following in the late 1980s, is designed to control virtually every manufacturing variable that can be controlled. That starts with ensuring that every production line uses the same machinery, and extends to such seemingly irrelevant details as the ambient humidity, the precise degree of air filtering, the exact color temperature of the room lighting, and the barometric pressure.

The benefit of Copy Exactly! is shown in the following graph. Intel began working on CE! after it had trouble at the 0.5um (500nm) node and refined the process through each successive generation. The green line shows the initial product ramp at its first fab — after an early spike, yields cratered and only recovered over a period of months. Once Intel had Fab 1 working well on 0.5um, it started ramping Fab 2 only to run into new problems. As Copy Exactly! was developed and deployed, the company’s yields synchronized across the various fabs.

Copy Exactly! doesn’t solve any initial yield or ramp problems, but it helps ensure that such issues don’t reoccur at different factories. This, in turn, gives Intel a critical degree of manufacturing flexibility. One of the downsides to a factory with higher yields on Product A than any other facility is that the fab in question is effectively stuck making Product A until the engineering teams can figure out what’s causing the discrepancy.

Intel’s “tick-tock” model isn’t just a handy meme — it’s a way the company controls variation on an architectural level. Historically, Intel always tended to follow a tick-tock cadence in which new architectures debuted on established processes followed by die shrinks, but the gap between the two was often uneven and didn’t necessarily correspond to product names in a coherent fashion. The first iteration of the Pentium III (Katmai), for example, was basically a Pentium II with SSE support. When Intel moved to 180nm and launched Coppermine, the chip’s onboard L2 cache made it significantly faster, clock-for-clock, than its predecessor.

Tick-tock’s cadence organizes and deploys new technologies in a consistent fashion. This minimizes the chance that a feature added relatively late in the design cycle will create an unanticipated problem in final silicon and simplifies consumer product messaging and gives engineers a concrete timetable when it comes to determining which technologies are included in the next revision and which are put back in the oven to bake a little more.

Better practices, better processors

There are some commonly fielded explanations for Intel’s success that Bohr didn’t talk about. He didn’t point to R&D spending as the chief reason why Intel leads the pack, even though Intel typically spends more on R&D than any other semiconductor manufacturer. He didn’t claim Intel’s engineers were just smarter than those at TSMC or GlobalFoundries — in fact, he noted that both foundries have a number of extremely talented individuals. Intel’s secret is how well engineers in research, development, and manufacturing work together as a team.

If the problems facing future CMOS scaling could be solved by the application of additional funds, AMD and GlobalFoundries wouldn’t be at loggerheads with each other. ATIC, the investment company that owns a significant share of GlobalFoundries, has poured billions into the company since it was created. This fresh infusion of capital was precisely what AMD couldn’t afford and it was intended to help GF quickly close the process node gap with Intel. Instead, the company ran into significant trouble with its 32nm HKMG technology. Those issues, according to both AMD and GF, are now behind it — but full standardization will be a challenge. GF’s acquisition of Chartered Semiconductor in 2010 gave it a great deal of additional fab capacity, but those factories have to be synchronized with its existing facilities in Dresden and New York state.

Intel’s advantage is the result of close collaboration between CPU designers and process engineers, superb manufacturing controls, and robust, continuing investment into R&D. It’s by no means guaranteed that these practices will carry the company smoothly through 14nm, but their success thus far speaks for itself. Whether TSMC and GlobalFoundries can achieve similar results within the constraints of the foundry business model remains to be seen.

Watch out, Raspberry Pi: Intel unveils ultra-small Next Unit of Computing PC

Details of a new, ultra-compact computer form factor from Intel, called the Next Unit of Computing (NUC) are starting to emerge.

First demonstrated at PAX East at the beginning of April, and Intel’s Platinum Summit in London last week, NUC is a complete 10x10cm (4x4in) Sandy Ivy Bridge Core i3/i5 computer. On the back, there are Thunderbolt, HDMI, and USB 3.0 ports. On the motherboard itself (pictured below) there are two SO-DIMM (laptop) memory slots and two mini PCIe headers. On the flip side of the motherboard (pictured below-below), is a CPU socket that takes most mobile Core i3 and i5 processors, and a heatsink and fan assembly.

According to Fred Birang, a senior product marketing engineer at Intel, the NUC is primarily targeted at digital signage and kiosks — but I’m sure we can all agree that it would make an awesome set-top home theater PC (HTPC), or an introductory system for kids. The only real problem is the GPU:This is Sandy Bridge we’re talking about, so you only get Intel’s HD 3000. Presumably, though An Ivy Bridge version of the NUC, with the HD 4000 GPU, is on its way.

At 10x10cm, the NUC is actually one of the smallest complete PCs on the market. The only x86 competition comes from VIA, which has produced Nano-ITX (12cm), Pico-ITX (10cm), and Mobile-ITX (6cm) motherboards for a few years — but these motherboards only support slow, weak VIA CPUs, and are generally targeted at embedded, low-power installations. That Intel has managed to cram a mobile Core i5 processor into such a form factor is rather impressive.

Where does this leave Raspberry Pi? At 8.5×5.5cm, the Raspberry Pi is still a fair bit smaller than Intel’s NUC (and at 2cm deep, it’s probably thinner as well). The Raspberry Pi has more inputs and outputs, too, though the NUC’s mini PCIe connectors mean that almost any functionality could be added. Processing power-wise, though, the NUC’s Core i3 and i5 processors will utterly obliterate the Rasp Pi’s 700MHz ARM SoC — but of course it will consume a lot more power, too. There’s also the fact that NUC users will be able to use the vast x86 Windows software ecosystem — and likewise, developing for the NUC will be as easy as developing for a standard, Windows-based x86 PC; two perks the Raspberry Pi will not enjoy.

Price-wise, Birang, speaking to Just Press Start, says the NUC will “not be in the hundreds and thousands range,” and that Intel is still looking at “different kinds of SKUs.” It almost certainly won’t be as cheap as the $25 Raspberry Pi, but a price point around $100 would be realistic. Judging by the heatsink and fan assembly, the NUC will probably come with a CPU pre-installed — and hopefully some RAM, too. Intel certainly could produce a computer that competes with the Raspberry Pi on price, but it’s unlikely to do so (damn those profit margins). Availability-wise, Birang says we can expect the NUC to arrive in the second half of the year.

With a name like Next Unit of Computing, it would seem like Intel has grand designs for this mini form factor. The use of the word “unit” is particularly interesting — it suggests that the NUC might be stackable, in much the same way as my imaginary Apple iStack computer.

Read our Raspberry Pi explainer

Updated @ 21:00 We’ve heard from Intel and it turns out that the NUC will be Ivy Bridge-only. There will be no Sandy Bridge version available. Expect to see it some time in the second half of 2012.

[Image credit: Sweclockers]

Source

Does Ivy Bridge replace discrete video cards for gaming?

With Intel’s release of its new third-generation Core (aka “Ivy Bridge”) processing platform, the general consensus among tech reviewers seems to be that the processor company has finally made something of a breakthrough in how it handles its integrated graphics. AnandTech,Hot Hardware, Legit Reviews, Tom’s Hardware, and more — all are basically impressed. Heck, former ETer Loyd Case even pronounced that “Entry-level GPUs are dead.” Clearly Intel did something right this time, and people who buy Ivy Bridge CPUs will undoubtedly have more and better capabilities than were available on Sandy Bridge or its predecessors.

Which isn’t the same thing as saying they will be particularly good for gaming.

In the interests of full disclosure, I’m very anti–integrated graphics and have been for decades. Ever since I witnessed the wonder that EGA and VGA cards wrought on late-1980s/early-1990s computer games, I’ve insisted on having a discrete card in every system I buy or build for myself. (I felt, and feel, the same way about sound cards, but that’s a separate discussion.) But even if you believe that on-board video is somehow acceptable under specific circumstances, you should have your head examined if the idea crosses your mind — even for only a moment — that it’s somehow able to play games in a satisfying way.

In putting its Ivy Bridge chips out there (particularly the new flagship desktop model, the Core i7-3770K), Intel primarily touted their facility at handling games at a resolution most gamers wouldn’t be likely to consider: 1366×768. The most recent Steam Hardware Surveyshows that 15.54% most of that service’s users play at that resolution, compared to 25.36% for 1920×1080. So I couldn’t help but wonder: How well would the Core i7-3770K handle playing games at that higher and more popular resolution?

To find out, I pulled out six of the gaming benchmark tests I use on a regular basis — Aliens vs. Predator,Batman: Arkham City (that’s right, I sacrificed one of my precious installs for this project), Tom Clancy’s HAWX 2, Heaven Benchmark 3.0, Lost Planet 2, and Metro 2033 — and loaded them all onto a computer I’d thrown together specifically for Ivy Bridge testing (an Asus P8Z77-V Deluxe motherboard, 8GB of RAM, my go-to Western Digital VelociRaptor hard drive). I ran the benchmarks, with as many details as possible turned off (to give the Core i7-3770K CPU the best possible shot), at my 24-inch monitor’s native resolution (1920×1200, pretty close to my target).

Assuming a minimum threshold of 30 frames per second (fps) for basic playability, only one game made the cut: HAWX 2. This isn’t exactly surprising; it’s definitely one of the lighter tests I run, but I include it for purposes of balance between Nvidia and AMD GPUs (cards from the former company invariably get slightly better frame rates, as it’s a “The Way it’s Meant to Be Played” title). And its 54 fps result wasn’t too bad at all.

Wondering how much lowering the resolution further would help the other tests, I dialed them all down to 1680×1050 and reran the tests at the same settings. HAWX 2 was still the only one to get above 30 fps (61 fps), though Batman: Arkham City and Lost Planet 2 came extremely close (29 fps and 29.1 fps respectively). Things clicked considerably more at 1440×900, with only Heaven andMetro 2033 still holding out — which they also did at even the lowest resolution I tested, 1280×800.

But I wasn’t satisfied with the video quality at any of these resolutions; lots of things looked just a little too blobby for my tastes. So on every game that passed my first test, I upped the details just a tad to see whether it could hack a slightly more demanding challenge. HAWX 2 again flew at all four resolutions, Lost Planet 2 was fine up through 1440×900, and that was all. Nudging the settings up still further, HAWX 2 hit 33 fps and Lost Planet 2 made it to 35.2 fps at 1280×800, but neither could surpass 30 fps even at 1440×900, so that’s where I stopped.

Even with the minor detail improvements I implemented toward the end of the testing, you still have to forsake a fair amount in terms of video quality just to get these games to play — and then generally at frame rates most legitimate gamers would barely consider good. Don’t get me wrong: As far as I’m concerned, Ivy Bridge marks a significant leap forward for Intel, and definitely suggests some good things to come on the graphics side. But I can’t quite get myself to the point where I feel comfortable pretending that this is a major victory.

The main thing Intel has proven — or, if you prefer, proven again — with Ivy Bridge is that, if you want to play 3D games that both perform well and look good, even at lower (I would argue too low) resolutions, you absolutely need a discrete video card. There’s no way around this. It may not be news to most (okay, any) system builders, but the average consumer who buys an Ivy Bridge system and is attracted by the possibility of not having to shell out another three to five wallet-size portraits of Andrew Jackson just to get the newest titles to play is going to be disappointed.

Ultimately, the conclusion to draw here is exactly the one Loyd did: Intel has functionally obliterated any compelling reason to buy a $60 video card. I’ve been a bit skeptical of the need for them for years, to be honest, as I’ve generally questioned whether the gains you might see in tasks like video transcoding were worth the money if robust gaming remained elusive. But if Intel and AMD chips can handle all the basic stuff themselves and do it well — which they now do — the GPU guys will either need to devise a much better argument for the entry-level models, drop the prices across the board to edge the $100 cards (which I’ve found in most cases to be worth the money) closer to or even onto the lowest pricing tier, or give up on trying to court that segment at all. And once Ivy Bridge gets around, AMD and Nvidia will need to make that decision sooner rather than later.

So Intel definitely deserves to be congratulated on a solid release that will change our outlooks about processing and integrated video for at least the next year or so. Just don’t assume that Ivy Bridge’s strides translate into a revolutionary rethink of the graphics market for everyone. Those for whom gaming is, at most, a sometime thing will unquestionably notice some benefits. But everyone else should stick with at least a $100 standalone card (and preferably at least a $200 one if they can afford it) to ensure the games’ performance and appearance make them worth playing.

Source

Ivy Bridge: Intel’s killing blow on AMD

Reading through Joel’s review of the Ivy Bridge Core i7-3770K, I couldn’t help but wonder about its target audience. On the one hand, Ivy Bridge is better than Sandy Bridge in every way — you get better CPU and GPU performance at the same price point — but really, what’s a top-of-the-line CPU doing with an integrated GPU?

I’ve been sitting here, scratching my head, thinking of use cases for a Core i7-3770K. If you already have a Sandy Bridge machine — a Core i7-2700K (or 3820) — there’s very little reason to spend $315 on the 3770K, or any other Ivy Bridge launch chip, especially as some of the more compelling features (USB 3.0, PCIe 3.0) require a new motherboard as well. Even new computer buyers — why would they buy an (expensive) system with an uber CPU, but a GPU that can’t keep up?

And then I realized my mistake: I’ve been constraining my thoughts to the wrong form factor. Ivy Bridge is all about mobile performance, mobile screen resolutions, and mobile gaming. When you constrain your focus to laptops, Ivy Bridge simply blows the competition — AMD — out of the water. CPU-wise, the high-end Ivy Bridge mobile Core i7-3720QM is more than twice as fast as AMD’s mobile Llano parts. Even on the GPU front, a domain that AMD had no doubt hoped to remain ruler, Intel’s HD 4000 is faster than the “discrete integrated” mobile Radeon HD 6-series core. (The desktop Llano GPU parts are still faster than the desktop Ivy Bridge GPU, but not by much.) It’s hard to overstate the advances that Intel has made — The HD 4000 is some 50% faster than the HD 3000. It’s too early to be certain, but it would seem that Intel has finally worked out how to make a GPU.

Dominating the competition in terms of performance isn’t enough in today’s battery-oriented world, though — but fortunately, because Ivy Bridge is a 22nm die shrink “tick,” it’s incredibly power efficient. A complete Core i7-3770K system (without discrete graphics) idles at a power consumption of just 43 watts, and 102 watts under load — this is for a chip that has a TDP of 77W. We don’t have exact numbers for the mobile Ivy Bridge parts, and TDP isn’t the same as power consumption, but Intel lists the TDP of the Core i7-3720QM as 45W. This is comparable to the mobile AMD Llano parts — but remember, the Intel chip is significantly faster.

Ultimately, Ivy Bridge is the first example of Intel using its superior fabrication process to tighten its mobile platform thumbscrews. This is an issue for AMD, which has all but given up on competing on raw performance, instead focusing on more efficient heterogeneous computing — and now Intel rolls up with its new HD 4000 and 22nm process, all but decimating AMD’s lead. There’s no way that AMD will catch Intel on CPU performance, and while GlobalFoundries and TSMC are stuck at 28 and 32nm it’s unlikely that Sunnyvale will remain competitive in terms of power consumption either, unless it has some kind of miraculous response to Ivy Bridge and next year’s Haswell.

AMD still has an edge on GPU tech, and we’ve seen the potential speed-up that could be achieved with heterogeneous computing, but really we’re years away from developers, compilers, and operating systems supporting truly heterogeneous system architectures. Even with Trinity waiting in the wings, it doesn’t look good for AMD on either the desktop or laptop — and with Medfield making a strong debut on smartphones and tablets, Intel is in a very strong position indeed.

Read our full Ivy Bridge review

Source

Intel Core i7-3770K review: Ivy Bridge brings lower power, better performance

Intel’s Ivy Bridge (IVB) has been one of the hottest tech topics of the past 12 months — we haven’t seen this much interest in a CPU since Intel launched Nehalem. Ivy Bridge is the first 22nm processor at a time when die shrinks have become increasingly difficult, the first CPU to use FinFETs (Intel calls its specific implementation Tri-Gate), and it’s a major component of Intel’s ultrabook initiative. If all goes well, Ivy Bridge will usher in a new series of 15W ultra-mobile parts, though these won’t reach the market for a little while yet.

Ivy Bridge is a “tick” in Intel’s tick-tock model, but the company is referring to its latest architecture as a “tick+.” The reason for the change is the disparity of improvement between Ivy Bridge’s CPU and GPU components. IVB’s CPU core is a die-shrunk Sandy Bridge (SNB) with a few ultra-low-level efficiency improvements. Performance improvements on the CPU side are in the 5-10% range. Unlike Westmere (Nehalem’s “tick”), which offered 50% more cores, Ivy Bridge keeps Sandy Bridge’s quad-core configuration.

At 160mm² and 1.4 billion transistors, Ivy Bridge is just over half the size of Sandy Bridge, with 61% more transistors. It therefore follows that Ivy Bridge’s transistor density is substantially higher than anything Intel has previously built.

Intel has decided to keep Ivy Bridge focused on lower power, rather than ramping its clock speed. Rumors have flown recently that the TDP on the Core i7-3770K has been raised to 95W, but Intel’s press materials state 77W. It’s possible that the company is planning two different versions of the chip, or that the part may have two different steppings. Either way, 77W appears to still be on the table. Intel also isn’t changing its product differentiation; there’s no way for an enthusiast to buy an unlocked CPU that also offers access to technologies like vPro or VT-d.

If Ivy Bridge’s CPU is a bit boring, the new GPU more than makes up for it. Ivy Bridge’s integrated graphics core increases the total number of execution units by 33% (to 16, up from 12), and implements support for DirectX 11, OpenCL 1.1, and OpenGL 3.1. There are now two texture units instead of one, and the GPU can issue twice as many MADs (Multiply-Add) per clock. Ivy Bridge incorporates a small, dedicated L3 cache of its own, but retains its ability to share data across the high-bandwidth ring bus that connects it to the processor, if necessary.

Other improvements include better Z-culling, an improved anisotropic filter, and better image post-processing capabilities. Intel claims its new chip can boost performance by up to 60% compared to Sandy Bridge. The Quick Sync video technology that debuted with SNB last year also gets a performance boost from these new features, and IVB supports up to three displays (up from Sandy Bridge’s two).

Backwards compatibility*

One of Ivy Bridge’s other features is that it’s backwards compatible with Intel’s Series 6 chipsets — with a few strings attached.

Once your motherboard vendor makes the appropriate BIOS updates available, you’ll be able to drop an IVB into an old board, or re-use a Sandy Bridge part in a Series 7 system. Intel’s stance on PCI-Express 3.0 is that you’ll have to buy a Series 7 motherboard to take advantage of it; the company has no plans to support the new standard on Series 6 or X79-based motherboards.

These limitations remove most of the potential incentive of going the hybrid route, but this level of compatibility makes it easier to repurpose hardware or upgrade systems on a budget.

Benchmarks

We tested Intel’s DZ77GA-70K with both a Sandy Bridge 2600K (3.4GHz base, 3.8GHz turbo, quad-core, 95W TDP) and the Ivy Bridge 3770K (3.5GHz base, 3.9GHz turbo, quad-core, 77W TDP). We populated all four DIMM slots with 16GB of RAM and used Intel’s integrated graphics for the CPU comparison, in order to measure Ivy Bridge’s performance improvements. A 256GB OCZ Vertex 3 with AHCI served as storage for both systems.

When we covered the Ivy Bridge launch a few weeks ago, we detailed a number of problems we had with the motherboard. Intel sent us a new BIOS shortly thereafter and we’re pleased to note that the new chip solved the problems we encountered. The motherboard now properly detects USB 3.0 devices and can boot off SATA drives in the usual way.

3D Rendering

As expected, Ivy Bridge’s gains over Sandy Bridge are modest. Maxwell Render picks up a hair over 10%, Cinebench a bit less.

3D Gaming

Ivy Bridge’s synthetic 3D performance, on the other hand, is much stronger — IVB comes within a hairsbreadth of doubling SNB’s 3DMark Vantage score. We couldn’t compare using the newer 3DMark 11, as Sandy Bridge doesn’t support it, but the Vantage gains are impressive. The question is, do they hold up in real world titles? To test this, we measured both chips in three relatively new games — Civilization V, Battlefield: Bad Company 2, and The Elder Scrolls V: Skyrim.

Sandy Bridge couldn’t handle Civilization V‘s benchmarks very well, and Ivy Bridge has similar problems. Performance between the two was surprisingly static, with IVB squeezing out narrow victories. Dropping all of the game’s options to minimal levels substantially improved matters — IVB’s performance goes up to ~21 FPS in the Late Game View, but has the unfortunate side effect of rendering it pig-ugly. Switching to DX9 does no good — Civilization V is the rare game that actually runs faster in DX10/11 mode.

We’ve combined our Skyrim and Battlefield: Bad Company 2 graphs. Both games were run at 1366×768 at medium detail settings. We customized Skyrim‘s options by disabling the FSAA/FXAA that the game uses at “Medium” but left the other options at their defaults. Ivy Bridge’s performance makes a very noticeable difference in both cases; BF:BC2 frame rates jump 25%, while Skyrim‘s performance leaps 27%.

There’s no downside to these gains, but there is a caveat: Both games look pretty wretched. InSkyrim‘s case that’s easier to rectify; the FPS boost is sufficient to allow for an increased amount of detail without the frame rate plunging back into unplayable territory. With the faster-paced BF:BC2, however, that’s not an option. 20 FPS is only just this side of playable and the game still stutters.

One thing to keep in mind, however, is just how far we’ve come in a few short years. Three years ago, it wasn’t uncommon for Intel’s integrated graphics to stumble on four-year-old titles. AMD and Nvidia had much better solutions, but “better” is a relative term. None of them were up to playing the latest titles. Today, Intel’s on-die GPU can handle at least some titles released within the past 12 months.

Multimedia Encoding

Intel’s Quick Sync was a major component of Sandy Bridge’s unveil last year. It offers an unparalleled price/performance ratio when it comes to video encoding, easily outperforming GPU-assisted encode/decode solutions from both AMD and Nvidia. That debut was marred by Intel’s decision to split Quick Sync support between multiple chipsets — a problem the Z77 chipset happily avoids.

Intel isn’t giving Quick Sync a new label but Ivy Bridge’s version is quite a bit faster than Sandy Bridge’s already excellent performance.

Unfortunately, this graph doesn’t tell the whole story. We’re still investigating this situation, but here’s what we know for certain. Both Media Espresso 6.5 and Arcsoft Media Converter 7.5 offer an iPhone 4S preset. The quality settings of that preset change dramatically depending on which GPU is detected. The program doesn’t mention this, and while there is a “Higher Quality” option for Quick Sync users, it has no impact on the file size or quality of the final result.

This is the preset page ME displays regardless of GPU type.

Encoded with Intel Quick Sync, Stargate: Ark of Truth came to 1.86GB. Encoded with a Radeon 7950, the same film is 5.8GB. Our media analysis tool indicates that the Quick Sync version is encoded at 1.4Mbps, with a maximum bitrate of 2Mbps. The Radeon flavor, in contrast, is encoded at a constant 7Mbps. The image below shows the two outputs side-by-side; the Radeon 7950 is on the left. Click the image to enlarge it if you want a better look (large file warning).

Check the fire and the rocks at the lower-right-hand corner to see the difference

The quality difference is immediate and jumps out at you even when watching the movie, especially on a high-resolution screen inches away from your nose. Software-mode is virtually identical to the Radeon 7950 option, which leaves Quick Sync as the definitive odd man out. If you’re a blind person who enjoys listening to TV shows, Cyberlink’s implementation of Quick Sync is going to be right up your alley. Everyone else is likely to be disappointed with the final output.

Conclusion

If you own a Sandy Bridge-based system, you’ve got little reason to upgrade to Ivy Bridge, especially since some of the platform’s features require a new motherboard. Those with older equipment — particularly anyone still riding the Core 2 Duo/Quad bus — may finally have a reason to take the plunge. Ivy Bridge’s lower TDP will allow it to fit in smaller enclosures without compromising performance, while its enhanced 3D capabilities and better visual quality make it a more attractive as the basis for an HTPC.

Ivy Bridge isn’t just a tick on Intel’s roadmap — it’s a powerful example of how the company’s priorities have changed. It’s the first chip Intel has ever built that’s being marketed on the strength of its graphics hardware, and for good reason. While we didn’t have a Llano to compare against directly, a bit of research shows Ivy Bridge nipping at AMD’s heels when it comes to GPU performance. Long term, Ivy Bridge and its successors will have an impact on software development. In a year or two, developers will be able to assume OpenCL support will be standard across most x86 systems, no matter what hardware they use.

Intel’s focus on graphics with Ivy Bridge is an example of how the limits of semiconductor scaling have forced the company to re-examine its priorities and focus on improving performance on the areas most likely to benefit from it. IVB’s integrated GPU may not be good enough for us to recommend it to moderate gamers, but this is the first time we can fairly say we see Intel reachingthat goal in a generation or two. It’s a strong step forward on every front and it sets the stage for an impressive Haswell debut in 2013.

Source

Thunderbolt for PCs a reality as Ivy Bridge launch nears

If you’re a PC user dreaming of Thunderbolt, today’s news will be music to your ears. According to reports, Intel has begun shipping its second-generation Thunderbolt controller, which means motherboards supporting the technology (like Intel’s DZ77GA) and the company’s next generation CPU will be able to ship. Cactus Ridge is Intel’s first Thunderbolt chip that supports Windows, and is expected to launch alongside the chipmaker’s upcoming Ivy Bridge CPU.

Cnet reports that Ivy Bridge should land on April 23, and Intel is making a push among partners to include Thunderbolt support in motherboards built for their new CPUs. The company obviously has a vested interest in seeing this happen: it co-created the technology alongside Apple and, with USB 3.0 taking hold in the industry, Intel’s window is closing fast.

Ivy Bridge CPUs will have integrated support for USB 3.0 but not for Thunderbolt, so the additional controller is required. Acer, Asustek, and Lenovo are expected to be some of the first to debut the high-speed interface for Windows computers, but it is believed the option will be restricted to high-end laptops and desktops.

Why? Intel is fighting a battle of perception among PC manufacturers. HP has committed exclusively to USB 3.0 as it sees no “value proposition” in Thunderbolt. That is bound to hurt the specification overall, as HP is the world’s largest PC maker by a decent margin. Second place Dell has so far been quiet on its plans, further adding to the uncertainty.

Uncertainty is not the friend of business, and this is the biggest reason why manufacturers have so far steered clear of producing peripherals with Thunderbolt support. Save for a handful of drives and ultra high-end items such as Apple’s Thunderbolt Display, there isn’t much out there. Compare this with USB 3.0, which already has dozens of peripherals, and is widely available, including through retailers like Target and Walmart.

Thunderbolt will try its best to catch up, though. Intel’s PC Client group chief Kirk Skaugen said earlier this week at the Intel Developer Forum in Beijing that he expects at least 100 Thunderbolt-capable devices available at retail by the end of the year, increasing to the “hundreds” in 2013.

PC support is key. While HP may be a lost cause, Intel must focus on the rest of the market, especially Dell, to ensure that Thunderbolt does not go the same way that FireWire did. There too Apple had its hands in the pot, and created a standard that was too expensive to implement. Sounding like Thunderbolt yet? USB 2.0 was cheaper and easier (and also featured backward compatibility), and it won the day.

As ExtremeTech’s Sebastian Anthony pointed out last year, if Thunderbolt peripherals continue to cost $200-300 more, it doesn’t matter who supports it — nobody’s going to use it.

Source

Intel’s Ivy Bridge ultrabook is on a collision course with the iPad

At the CeBIT electronics show in Hanover, Germany Intel is reaffirming its commitment to the ultrabook segment. On display are reference designs for models based on the company’s Ivy Bridge architecture, on which the company has made great strides in power efficiency. In other words? A perfect fit for a device like the ultrabook.

The model on display sports a 13.3-inch capacitive touchscreen. The laptop’s enclosure is of similar sizing and thickness to other 13-inch ultrabooks, with a decent feature set: USB 3.0, SIM and SD card slots, and HDMI.

The inclusion of the touchscreen technology is where things get interesting. Intel’s ultrabook mimics the functionality of any decent tablets: the touchscreen is responsive, battery life will almost certainly be strong, and multitouch gestures work like you would expect. Simply put, it is a tablet with an attached keyboard.

Intel’s reference design carries a greater significance though. Whereas previously ultrabooks were seemingly marketed as a step above tablets, the trajectory of the sector puts it on a course that will encroach on tablet’s turf, especially when it comes to price. We have statements from the manufacturers themselves to support this argument.

Acer is one of these companies, and president Jianren Weng is repeating statements at CeBIT that he’s made to the media before: ultrabook prices are about to go way down. How much? To $499, and its obvious these companies have the iPad in their sights.

It has been said over and over again that the iPad is eating the notebook’s lunch. We’ve debated a similar point on the pages of ExtremeTech. The tablet has just about disposed of the netbook, and price has helped it to cannibalize the bottom end of the laptop sector, too.

There are many of us out there that still need the notebook experience, however. Why not strike back at the tablets (read: the iPad) directly, then? It’s not a bad business move, but the ultrabook manufacturers are running into cost concerns when trying to do so. In the same discussion as the $499 ultrabook, Weng admits it is making no money on its cheapest model in the market, the $799 Aspire S3.

As much as Weng wants to take on the iPad, it’s just not happening — at least for the moment.

Tabletizing the ultrabook adds a whole new level of cost, including the touchscreen display itself and the host of other parts necessary to make touch work. If Acer and other partners are having trouble getting the costs down now, it will be near impossible to get it down to that of a tablet without losing money.

(It should be mentioned that The Verge did a little snooping around at CeBIT, and Weng’s underlingsare much less optimistic about their boss’ grand plan to take on tablets.)

Circling back though, it appears from what’s out there that Acer is not alone in seeing a tabletized future for the sector, and the company is getting support (although indirectly) from Intel. These reference designs are a starting point from which all manufacturers can build off of. That said, Acer’s desire to compete with the iPad on price is a pipe dream at this point, but the ultrabook is moving the way of the tablet anyway.

With Windows 8 incoming combination devices like this are going to be interesting. From the ground up, Microsoft’s operating system is built for touch. Such a system will work beautifully on touch-enabled computers like this ultrabook, so regardless of the success of Acer or Ivy Bridge, get ready for a more touch-friendly Windows experience.

Source

The future of computers: 3D chip stacking

In a few weeks, Intel will release Ivy Bridge, the first mass-produced 22nm parts, and more importantly the first to use 3D “tri-gate” FinFET transistors. These CPUs will be incredibly fast and use very little power, but ultimately they are just another last-gasp effort to squeeze a little more life out of a material and process that will soon hit a wall. Computing is still predominantly single-threaded; throwing more transistors and more cores at a problem will only take you so far.

Fortunately, there’s another maturing technology that should provide a much-needed lease of life to the silicon industry: Chip stacking, or to give its formal name, 3D wafer-level chip packaging. Chip stacking is exactly what it sounds like: You take a completed computer chip (DRAM, say), and then place it on top of another chip (a CPU). As a result, two chips that used to be centimeters apart on a circuit board are now less than a millimeter apart. This reduces power consumption (transmitting data over copper wires is messy business), and also improves bandwidth by a huge amount.

Obviously, though, you can’t just take a DRAM chip and whack it on top of a CPU. The chips need to be designed with chip stacking in mind, and it takes specialized machinery to actually line the dies up and attach them. To this end, Applied Materials — the company that makes all of the machines used by Intel, TSMC, Samsung, GloFo, and every other semiconductor manufacturer — and A*STAR’s Institute of Microelectronics (IME) have announcedthe opening of a bleeding-edge 3D chip packaging lab in Singapore. Built with a combined investment of over $100 million, the Centre of Excellence in Advanced Packaging features a 14,000 square foot cleanroom containing a complete 300-millimeter production line and 3D packaging tools that are unique to A*STAR. The Centre isn’t a commercial fab, however: It’s actually designed as a facility for other companies, such as TSMC or Samsung, to come and experiment with 3D packaging. As far as Applied Materials is concerned, of course, this is an excellent way to demonstrate and sell its machines.

There are three main ways of stacking chips, all of which will be available at the new research center. The most basic technique (Bump + RDL) involves stacking two chips together, and then connecting them both to a flip chip at the bottom of the stack; the chips are physically close, which is a good step forward, but they can’t communicate directly with each other. This technique is already used in some SoCs to place DRAM on top of the CPU. The second technique, which is also the most complex, is called through-silicon via (TSV, pictured right). With TSV, vertical copper channels are built into each die, so that when they’re placed on top of each other the TSVs connect the chips together. This is the technique that IBM and 3M will use to stack hundreds of memory dies together to make super-density DRAM. So far, TSV has only really been used in camera CMOS sensors, but adoption will increase over the next few years as the technology matures.

The third technique, which isn’t technically stacking but still counts as “advanced packaging,” uses a silicon transposer (pictured above, below the stacked chips). A transposer is effectively a piece of silicon that acts like a “mini motherboard,” connecting two or more chips together (if you rememberbreadboard from your days as a budding electronic engineer, it’s the same kind of thing, but on a much smaller scale). The advantage of this technique is that you can reap the benefits of shorter wiring (higher bandwidth, lower power consumption), but the constituent chips don’t have to be changed at all. Transposers are expected to be used in upcoming multi-GPU Nvidia and AMD graphics cards.

In theory, there’s almost no limit to how many dies can be stacked in this way. Applied Materials, Micron, and Samsung have been mooting the idea of an eight-layer DIMM, but in an interview, Applied Materials tells us more layers should be possible. The only real restriction is heat generation and dissipation, which will limit the number of CPUs that you can have in a stack, but there’s no reason that an entire SoC — CPU, DRAM, NAND flash, radios, power management IC, and GPU — couldn’t be built into a single through-silicon via chip. According to Applied Materials, this would allow for packages that are around 35% smaller, consume 50% less power, and perform significantly faster — desirable traits when it comes to smartphones and tablets. Moving forward, TSV is likely to dominate any space that puts a premium on power efficiency, such as mobile and server.

Finally, chip stacking obviously works in synergy with Intel’s 3D FinFETs — though curiously there is no sign of TSV on Intel’s roadmap, while TSMC is all over it. Perhaps the most important thing to remember is that new production and packaging processes take a long time to roll out: It has taken Intel 10 years to iron out mass production of FinFETs, and likewise, chip stacking has been touted as the next great thing for almost as long. Applied Materials and IME’s new 3D packaging lab is definitely a step in the right direction, but don’t expect your next desktop CPU to have DRAM stacked on top of it; we’re still a couple of years out at least.

Source

Intel i7 3820 released, spiritual successor to i7 920; Ivy Bridge delayed until June

The Lord giveth and the Lord hath taken away: Intel has quietly released the Sandy Bridge-E Core i7 3820, a quad-core, mid-range, socket LGA2011 part that’s the spiritual successor to the i7 920 and 930 — but according to a report from Digitimes, mass roll out of Ivy Bridge has been postponed until summer.

The Core i7 3820 is the first sub-$500 chip to be released for the LGA2011 socket, the same socket used by Core i7 Extreme 3980X and 3860X, the i7 3930K, and the Xeon E5 server processors. Compared to its 6- and 8-core compatriots, the 3820 is “only” a quad-core, but it retains the same on-die 4x 64-bit DDR3 controller, 40 PCIe v3 channels, and dual-link QPI. There’s 10MB of L3 cache, too, and the base clock rate is 3.6GHz. The price? A miraculous $285. Yet again, it doesn’t look good forAMD’s top-of-the-line FX-8150, which is priced around the same point and yet significantly slower than Intel’s offerings.

According to VR-Zone, Intel achieved this price by creating an entirely new die. Where the Core i7 Extreme, 3930K, and Xeon E5 processors are all based on an octo-core LGA2011 die, the 3820 uses a brand new quad-core die that’s roughly 30% smaller, and thus cheaper. It is expected that this same die will be used in the upcoming Xeon E5 1620, which will also be a very cheap chip ($290).

In other news, “mass shipments” of Ivy Bridge processors have been delayed until after June.According to Digitimes, a weak global economy has caused a build-up of Sandy Bridge inventory both at Intel and OEMs. If Intel went ahead and mass released Ivy Bridge in April, these Sandy Bridge parts would have to be thrown away or sold at much lower prices. Now the plan is to release someIvy Bridge chips in April, but postpone mass shipments (presumably of consumer-oriented parts) until after June. System builders should still be able to get their hands on some mid- and high-end Ivy Bridge chips in April.

Source