ARM has been making waves over the past two years with plenty of processor and graphics IP announcements, but they are not alone in the game. MIPS Technologies, almost as old as ARM itself, also licenses RISC processors. With licensees like Broadcom and Sigma Designs, they have undoubtedly held the upper hand in the home entertainment / set-top-box arena as well as the networking space. However, success in the fast-growing mobile / tablet space has been hard for MIPS to come by, thanks to ARM being well-entrenched in that market.

Today, MIPS is introducing a range of new processor IP cores in the Aptiv lineup, similar to ARM's Cortex. The members of this lineup range from small microcontroller cores to triple dispatch superscalar ones. By introducing a member at each performance level to compete directly with offerings from ARM, MIPS has made its move in the processor IP battle. Read on for our analysis of the announcement and the processor  architectures.

At the 2012 Global Influencer Summit in Shanghai, HP announced a full slate of new thin-and-light and ultrabook systems, and we've gotten to go hands on with all of them.

The most important products of the lot are the new Envy ultrabooks and sleekbooks. Sleekbook is just a marketing term to describe notebooks that don't meet all of Intel's criteria for ultrabook classification due to CPU and storage component selection, but are otherwise identical to the ultrabook line, including sharing the same 19.8mm thick chassis.

Both of the model lines are available in 14" and 15.6" sizes, and feature Beats audio, glass trackpads with multitouch and gesture support, and optional backlit keyboards. The lid and palmrest are brushed aluminum, while the bottom of the notebook is a soft-touch plastic material. There are two color options - silver metal/black plastic, or a much more visually arresting black metal/red plastic model. The ultrabook lines are based on Intel's 3rd generation Core ultra-low voltage processors, similar to other ultrabooks, while the sleekbooks come with normal Ivy Bridge processors in the 14" or AMD"s Fusion APUs in the 15.6". The entire lineup has battery life quoted in the 8-9 hour range (depending on screen size and CPU choice). I suspect that the ultrabook and sleekbook versions of the will be mixed up very often (I saw some HP product managers confuse the different demo units more than once here in Shanghai), but what gets lost is that effectively, they're all just different flavors of the same notebook.

HP has always chosen to design its volume platforms as highly modular systems, with consumers and retailers able to chose from a variety of screen sizes, Intel or AMD processors, and a variety of storage/memory/graphics options. For the first time, we're seeing that mentality hit the ultrabook-class of device. HP has designed the new Envy as a single platform with two screen sizes, a range of Intel and AMD processors, a choice of mechanical, solid-state, or hybrid storage options, and optional AMD dedicated graphics (for the Intel models - the AMD models have onboard graphics that are deemed good enough to not merit a dGPU option). If you tick the right combination of boxes (Intel ULV CPUs and hybrid or solid-state storage), the Envy can meet Intel's spec to be classified as an ultrabook. If not, HP calls it a sleekbook. It's a bit confusing, especially in HP's relatively vague press blast, but upon explanation, the new term makes sense.

The Envy devices are pretty nice, with an attractive design (particularly the red and black one) and solid build quality. THe interesting thing here is that the AMD-based Envy 6 (the 15.6" Envy) starts at $599, making it a pretty great value for that price point considering the premium design and features. The base Envy 4 starts at $699, while the ultrabook-spec Envy 4 (essentially the same system, except with a 32GB mSATA caching drive to suplement the mechanical storage) goes for $749. The Intel-based Envy 6 starts at $799 and comes standard with the caching drive, thus meeting the ultrabook spec as well. The AMD-based Envy 6 will be available on June 20, while the rest of the Envy systems are available now.

HP is also expanding the premium Envy Spectre line with the Envy Spectre XT. Unlike the Spectre 14, the Spectre XT has a 13.3" display and is made entirely of anodized brushed aluminum. The very attractive all-metal design goes in a different direction than the Spectre 14, which was the first notebook to be made primarily of glass. While innovative, the Gorilla Glass chassis led to the Spectre 14 being a bit expensive, as well as thicker and heavier than most ultrabooks tended to be. The Spectre XT, on the other hand, has much more petite dimensions and measures 14.5mm thick at the thinnest point, along with a 3.07lb weight (1.395kg). The Spectre XT certainly feels like a step up in design and build from the Envy, but it's not as unique or as immediately stunning as the Spectre 14. The Spectre XT will hit market on June 8th, with a starting price of $999 with Intel's Ivy Bridge ULV processors and solid state storage. The Spectre XT is aimed squarely at the Samsung Series 9 ($1399) and other premium ultrabooks, so HP seems to be pricing their new portables relatively aggresively. It's a good sign, one that means we'll likely see the entire ultrabook market go more mainstream in terms of price point going forward.

In other ultrabook news, HP announced their first ultrabook meant specifically for the corporate world - the EliteBook Folio 9470m. The 9470m has a 14" screen and weighs a scant 3.6lbs. It squeezes a number of enterprise-centric ports and features into its 19mm thick frame, including VGA, DisplayPort, a full-sized Ethernet port, Smart card reader, fingerprint scanner, embedded TPM security chip, and Intel's vPro security technology. It will come with Ivy Bridge mobile processors and an optional SSD, and is expected to be available sometime in October 2012.

Apple is a bit infamous for its tight control over new products, particularly iPhones. Development mules are often updated internals stuffed into nondescript previous generation designs. And talk of new products is forbidden, until the official reveal. Samsung, seems to have taken a different tack with their latest flagship device, the Galaxy S III. In the weeks leading up to today's announcement there were positively dozens of leaked images and specifications lists. And each one seemed so pointedly different than the last to leave the tech press exhausted with confusion and anticipation. All those leaks are put to the test, now, though; as Samsung has just revealed their most advanced phone yet, the Galaxy S III. 

Dubbed the Galaxy S III, the latest Samsung flagship device brings updated internals and many software additions to the Galaxy line. The 8.6 mm thick slate features a 4.8" HD SAMOLED display pushing 1280x720 pixels, on an RGBG stripe. The body is curvy and smooth and comes in Pebble Blue or Marble white, and appears to be metal in appearance (Ed. note: the images seem a bit ambiguous, we'll update when we get hands-on). There's more to unpack than was even included in the announcement, so let's get to it. 

The display is going to undoubtedly be a point of contention for some of you. Indeed, before any of you have read down this far, I suspect the first comment has appeared below bemoaning the lack of a '+' at the end of the display's nomenclature. Yes, the 4.8" 1280x720 AMOLED display is of the Super variety, but lacks the RGB stripe of the Plus variety. We'll crunch some numbers and consider the likelihood that anyone will be able to suss out individual subpixels later. Aside from that we'll reserve judgment till we have a review unit in hand to sort out display quality. The 4.8" display only stretches the width of the device another 1.5 mm or so, so users comfortable with these larger phones should have no issues. Those of us still skeptical about this screen size might hesitate. 

The design is oddly reminiscent of the iPod touch, but with gently curved surfaces across the front and back of the device. There is a familiarity to the design and doesn't step boldly away from the language first seen in the original Galaxy S. We'll know more about the tacticle experience after our hands-on. For now, peep the gallery and stay-tuned for a hands-on and an overview of the software aspects being introduced today. 

Some of what we know today wasn't leaked, but was formally announced by Samsung earlier. Samsung Semiconductor, designers and fabricators of their Exynos family of ARM silicon, anounced that the next Galaxy phone would include their Exynos 4 Quad, previously known as the Exynos 4412. Like NVIDIA's Tegra 3, the Exynos 4 Quad features four ARM Cortex-A9 cores, though no companion core is put to use for powersavings. Each core can be power gated individually, just like Tegra 3. The big news here is the Exynos 4 Quad is built on Samsung's 32nm high-k + metal gate process, which should provide for a sizeable decrease in leakage and an overall improvement in power consumption compared to previous 45nm desgins.

That's not the only SoC we'll be seeing in the latest Galaxy devices, though. Like the Galaxy S II devices, LTE is limited to devices running Qualcomm SoC's, so US variants on Verizon, AT&T and Sprint will likely be sporting our latest favorite SoC, the Snapdragon S4. We can expect a lot of this, as it's not just necessary to move the performance bar with each generation, users expect battery life to see a respectable improvement as well. In our HTC One X (AT&T) review we noted just how much better battery life is with Qualcomm's 28nm radios, this is the kind of generational leap we want to see. Now, we're not sure why we're not seeing their 28nm radio-only parts, the MDM9x15. These basebands will include the same LTE Cat. 3 performance, but at much lower power consumption. We've never seen its predecessor the MDM9x00 paired in a phone with anything other than Qualcomm silicon, but we have seen it in data-only situations paired with the likes of the Apple A5X in the iPad 3 and within mobile hotspots and data cards. 

None of this was mentioned during the event, but our man Brian Klug is hard at work pressing Samsung for confirmation on these details and we'll update as we learn more. Notably, is this statement from Samsung: 
Samsung Mobile is planning a U.S. version of Galaxy S III, optimized for the fastest LTE and HSPA+ networks in the U.S., which will be available in the summer of 2012.  Exact timing and retail channel availability is not being announced at this time. We believe the Galaxy S III is the most anticipated product in the 20-year history of Samsung Mobile; therefore, we will continue to share information as it becomes available. 
So, without committing to anything, they're admitting that additional work will go into the S III before it appears on these shores.  

While NVIDIA doesn’t publically announce most of their OEM desktop graphics cards, they do update their website with the specifications of these cards, which is how we usually find out about them. Today has been no exception, and after NVIDIA's latest site update a bit of digging has unearthed the fact that NVIDIA has released their first Kepler cards for the desktop market. There are 5 new OEM desktop cards, composing a mix of both Kepler and Fermi: the GT 645, the GT 640, and the GT 630.

If this product stack looks familiar, it should. It’s generally the same product stack as the GeForce 600 series Mobile lineup, except with higher clockspeeds. As with their mobile parts, NVIDIA is going to be mixing 40nm Fermi parts and 28nm Kepler parts into their desktop product stack, leading to a hilariously frustrating selection of video cards.

At the top of the new product stack we have the GT 645, which is a GF114 Fermi rehash. GT 645 has 288 CUDA cores enabled and paired with what’s listed as a very crippled 128bit memory bus. However considering the memory bandwidth NVIDIA lists for the card (91.9GB/sec) and the fact that they already have a very similar card in the GTX 560 SE, we’re confident that the 128bit bus in NVIDIA’s specs is a typo and that it’s actually a 192bit bus, and we are listing it in our charts accordingly. In any case you’re still looking at significantly less memory bandwidth the GTX 560 is typically paired with.

The next card is the GT 640, the GT 640, and the GT 640. Just like the GT 640M LE, NVIDIA is mixing Fermi and Kepler here in a very odd manner. We have a GT 640 that’s a full GK107 (384 cores) with GDDR5 memory and a fairly high clockspeed, a GT 640 that’s a binned GF116 (144 cores) with DDR3 memory, and a GT 640 that’s a full GK107 (384 cores) with DDR3 memory and lower clockspeeds. Not even the TDP or form factor is consistent among these cards; the GK107 DDR3 card is a low-profile 50W card, while the other two are full-profile 75W cards.

The final card is the GT 630, which is another GK107 part. This is also a full GK107 (384 cores), paired with DDR3 memory and a mid-range clockspeed, with a TDP of 50W. The most interesting part? It’s clocked 10% higher than the equivalent GT 640 and should have better performance as a result, though memory bandwidth is the same between the two.

It’s safe to say that at this point the OEM desktop video card market has turned into a similar mess as the OEM laptop market, and this latest round of video cards serves to cement that fact. As with the laptop market we’ve reached a point where it’s nearly impossible to tell which video card a product actually uses based on computer specs alone, and that’s worrisome. Accordingly, our best advice for buying an OEM desktop is the same as buying an OEM laptop: make sure you research what you're getting if you want faster GPU performance. It may not be possible to tell what video card is in use until a product has been reviewed.

Oh a final note, it’s interesting though not surprising that NVIDIA is releasing desktop GK107 cards to OEMs first. They did the same thing with the GT 200 series, which were NVIDIA’s first 40nm cards, and while these GT 600 cards don’t have the same distinction, the root cause – a lack of sufficient GPU supply – is the same. On a positive note however, this launch means that retail GK107 desktop cards – particularly a retail version of the GDDR5 + GK107 based GT 640 – can’t be too far away; we’d speculate a few months at the most. So budget desktop users shouldn’t be waiting too much longer for the 28nm generation to hit their market segment.

Source: SH SOTN, NVIDIA

We recently had a chance to take a meeting with Toshiba representatives in San Francisco, California, where we were previewed their upcoming releases for 3Q12. Toshiba is essentially targeting the back-to-school crowd with their refreshed notebook lines, while their new tablets aim to take both a more aggressive and a more out-of-the-box approach to the market.

Read on to see what Toshiba plans to bring to market for the back to school season.

HTC and Sprint teamed up today to announce the latest in the EVO line: the EVO 4G LTE. So, not the most novel name, but it gets to the point. The 4.7" device is carved out of an aluminum space frame, anodized to an all black finish with red accents and an exposed silver edge. And along that prominent red band across the back? That's right, the kickstand's back. But don't call this part of the One series. This is a Sprint EVO device through and through. 

Intel had a few updates about its Android SoC strategy for us at MWC this week. The first is a spec revision. The Atom Z2460 we talked about in great detail at CES was originally specced to run at a max of 1.3GHz but it could burst up to 1.6GHz if the thermal conditions allowed it (ala turbo boost). Z2460 yields on Intel's 32nm LP process are apparently better than expected so the Atom core will be able to turbo up to 2.0GHz instead of 1.6GHz. The default max frequency remains unchanged at 1.3GHz.

Next, Intel announced two new Atom SoCs for the smartphone market: the Z2580 and the Z2000.

The Z2580 is a higher end part, back from the fab now but shipping in devices in the first half of next year. It features the same architecture as the Atom Z2460 but instead of a single Saltwell core it has two, with Hyper Threading enabled (2 cores, 4 threads). The CPU cores can burst at up to 1.8GHz, while the default max CPU frequency remains at 1.3GHz.

The CPU isn't the only part of the SoC to get an upgrade: Intel equipped the Z2580 with a PowerVR SGX 544MP2 GPU running at 533MHz. The SGX 544 is similar to the 543 used in Apple's A5, however it adds support for Direct3D 9_3. At 533MHz you can expect roughly twice the shader/compute performance of the 543MP2 that's in the A5. Granted by early next year we'll likely see competitive, if not faster GPUs implemented in SoCs (quite possibly long before then).

The Z2580 will be paired with Intel's XMM 7160 LTE baseband. The 7160 is an upgraded version of the XMM 7060 that adds support for 3GPP Release 9. The full specs of the solution are below:

The Z2580 also received upgrades on the modem interface side to cope with the increased bandwidth from the LTE baseband. Remember the Z2460 launched with a HSPA+ baseband pair (XMM6260).

Just like with the Z2460, Intel has produced a FFRD (Form Factor Reference Design) based around the Z2580/XMM7160. I don't have photos (nor have I seen) the new high-end FFRD, but I'm told it's a better looking design than the current Medfield design. The only detail I have about the new reference design is it comes with a larger battery: 6.771Whr.

The Z2580/XMM7160 FFRD will begin sampling in the second half of this year, with customer units shipping in the first half of next year.

The Z2000 is Intel's new low-end Atom SoC SKU. While the Z2580 has two Atom cores, the Z2000 only has one. Max clock speed is limited to 1GHz and there is no support for Hyper Threading. Intel is planning on putting this core up against the ARM11 based SoCs that still sell into the low end of the smartphone space.

The GPU remains unchanged from the Z2460 (PowerVR SGX 540) however the max clock speed is limited to 320MHz (down from 400MHz). 1080p video decode is supported but video encode is limited to 720p.

The Z2000 will be paired with Intel's XMM6265 HSPA+ baseband:

The Z2000 has its own form factor reference design, which is different from both the Z2460 and Z2580. Once again I only have a single piece of information Reference designs are especially important at the low end as they can save customers quite a bit of money. With narrow margins there's not a whole lot of room to spend on industrial design or hardware customizations. A turnkey solution that's well built and reliable will be very useful for this market.

HTC begins 2012 with the reveal of a new unified brand strategy. Although HTC as a company has made significant progress in attaining mindshare, its devices lack a single focus to compete with the likes of Apple’s iPhone or Samsung’s Galaxy S brands. HTC, like many of its competitors, chose to spread its brand equity across multiple device brands like EVO, Sensation, Thunderbolt, Desire, and so forth. Moving forward, HTC is hoping to change that with the introduction of a new unified brand to do battle with these other brands: the HTC One.

The goal is that you’ll be able to walk into any mobile operator store, in any region, in any part of the world and ask for the HTC One. There will still obviously be variants of the One, but the brand will remain constant across them. Read on for some impressions and benchmarks.

We've known that Microsoft has been planning an ARM-compatible version of Windows since well before we knew anything else about Windows 8, but the particulars have often been obscured both by unclear signals from Microsoft itself and subsequent coverage of those unclear signals by journalists. Steven Sinofsky has taken to the Building Windows blog today to clear up some of this ambiguity, and in doing so has drawn a clearer line between the version of Windows that will run on ARM, and the version of Windows that will run on x86 processors.

Up until now, we've operated under the assumption that a new version of Windows called Windows 8 would be released this year, and that it would run on both x86 (32-bit and 64-bit - throughout this article I'll use x86 to refer to both architectures) and ARM processors - Sinofsky's post makes it clear that the ARM version of Windows, officially referred to as Windows on ARM (WOA), is considered to be a separate product from Windows 8, the same way that products like Windows Server and Windows Embedded share a foundation with but are distinct from Windows 7. Windows on ARM has a "high degree of commonality" and "very significant shared code" with Windows 8 - much of the user's interaction with the OS will be the same on either platform, and much of the underlying technology we've seen in our Windows 8 coverage so far will be present in both versions, but they're distinct products that will be treated differently by Microsoft.

This post is quite lengthy and represents what is likely to be our best look at WOA for at least a little while - we'll get to see some of its features in the Windows 8 Consumer Preview when it is released at the end of the month, but for now Windows on ARM is only being tested internally, and on customized hardware that will be sent to some developers and hardware partners at about the same time. It will be a little while before we see anything remotely similar to shipping hardware.

The Windows Desktop, Office, and x86 Apps

One of the biggest recurring questions I've seen about Windows on ARM is whether the standard Windows desktop would be available for use on those devices as it will be on Windows 8 machines - the answer is yes, it definitely will be. The desktop can be invoked from the Start screen, and once there users can perform standard Windows Explorer operations, launch the desktop version of Internet Explorer, and other tasks either via touch (for which Explorer has apparently been optimized) or via keyboard and mouse input. The desktop will only consume resources when it is launched, meaning that there are no performance or battery life implications for users who stick with the Metro interface for everything - the desktop is there if you want it, but one of Microsoft's stated goals with the Metro interface is to make it so that you don't need to use the desktop as a fallback.

Microsoft will also be bundling versions of Microsoft Word, Excel, PowerPoint, and OneNote with Windows on ARM systems. These Office apps will be a part of the new Office 15 family of products (suggesting, but not guaranteeing, that we may see a full x86 version of that suite before the end of the year as well), but will be touch-optimized versions of the applications rather than ports of the standard suite. The Windows on ARM products will "maintain fidelity" with their x86 counterparts (meaning that a file created in Word or Powerpoint on an x86 machine will look the same on an ARM machine), but will otherwise be redesigned to fit the platform - an early version of Excel is shown above.

That said, Microsoft is firm in its insistence that it will not support running, emulating, or porting existing x86 apps on the Windows on ARM desktop - apps can only be downloaded and installed through the Windows Store, and only apps written to target the new WinRT APIs can be distributed through the store (however, the store will be able to distribute and update both ARM and x86 versions of apps in the event that the app uses any code native to either architecture). Microsoft suggests that current Windows developers should be able to take significant bits of their existing code and wrap them in a Metro layer, but acknowledges that bringing over existing apps will require a bit of work - WinRT is clearly the wave of the future where Windows is concerned, but it'll be up to individual developers to decide how, when, and if to bring their programs over.

Supported Devices and Release Date

Windows on ARM is being written to run on ARM SoCs from NVIDIA, Qualcomm, and Texas Instruments, and it will only be available on devices designed to run it - you won't be able to buy a license for Windows on ARM and install it on an existing tablet, or a tablet designed to run Android. Microsoft is working with partners to deliver compatible hardware, and the company's goal is to start shipping devices running Windows on ARM at the same time as x86 devices running Windows 8 (currently slated for late this year).

In addition to SoC type, Microsoft will have a set of broad guidelines for Windows on ARM tablets that are similar to those for current Windows phones (the "chassis specification," in Microsoft parlance) - likely a set of supported screen resolutions and a list of required hardware devices designed to provide a middle ground between the uniformity of the iPad and the diversity-to-the-point-of-insanity of Android tablets. On Windows phones, these requirements are in place to give consumers some choice while also limiting developer headaches and ensuring a standardized look and feel across different devices from different manufacturers - the requirements for Windows on ARM will have the same aims, and we'll talk a little bit more about some of the hardware that will be common to WOA devices later on in this post.

In treating Windows on ARM as a separate product, Microsoft has left itself some wiggle room to let its release date slip without holding up Windows 8 (wiggle room is very important to the post-Vista Windows team, and they generally don't give hard dates unless they expect to be able to hit them). Microsoft obviously wants to ship before the end of the year because, let's face it, they don't want to give Apple, Google, Amazon, and the rest another holiday season all to themselves, but at this point in the game a botched or half-baked release in time for Christmas could actually be worse for Microsoft's market and mindshare than a well-executed release a few months later. Expect a concurrent release with Windows 8, but know that Microsoft hasn't yet completely committed to it.

When it is shipped, Windows on ARM should come as a single edition of Windows from a feature standpoint (though the company notes that no decisions regarding new Windows product editions have been finalized) - Microsoft promises to "adjust the features ... such that [WOA] is competitive in the marketplace and offers a compelling value proposition to customers of all types." That doesn't tell us much, but I think we should expect the consumer-oriented features that you'd find in a Home Premium version of Windows along with business-minded features (like domain joining and device encryption) thrown in to increase WOA's appeal to enterprises. Whether the decision to ship a single Windows on ARM SKU will have any effect on the x86 version's army of different editions remains to be seen.

Drivers, Updates, and Hardware

So, since Windows on ARM will only be available on devices designed specifically for it, Microsoft can actually keep track of what hardware WOA devices are guaranteed to be using. This means that all software, from OS patches to device firmware to specific drivers, can and will be distributed using Windows Update. Apple has achieved something similar in OS X - Macs are many and subtly varied, especially when you take multiple model years into consideration, but ultimately there is a finite set of hardware in the field, and Apple can keep every Mac in use up-to-date with drivers, firmware, and OS updates through Software Update, rather than the broad array of different first and third-party updaters required to patch those separate elements on an x86 Windows box (and I promise that I'm just comparing the two to give you a frame of reference, not because I consider one system to be inherently superior to the other).

To reduce the number of drivers it will have to keep up to date, Microsoft is relying heavily on "class drivers" to support hardware in both WOA and Windows 8 - for those of you just tuning in, a class driver is designed to support all hardware manufactured to certain standards, rather than targeting specific devices. They're why you can freely plug in different USB keyboards and flash drives to a Windows computer and have them recognized by the machine without needing to pop in a driver disk first.

A lot of the work Microsoft is putting into class drivers is also applicable to Windows 8 - we've already looked at new class drivers for USB 3.0 controllers, mobile broadband chips, and motion sensors, and we should also see class drivers for printers, Bluetooth, Embedded MultiMediaCard (eMMC) storage, and drivers for different busses and input devices (like the Windows, power, and volume buttons).

Where Microsoft can't create class drivers, it's trying to enforce some common specifications - WOA devices will all have DirectX-capable GPUs and drivers, which will power Metro apps, the Windows UI, and GPU acceleration in Internet Explorer among other things. This baseline has enabled Microsoft to improve on the fallback software GPU driver to enable a nicer-looking display on devices without a specific driver (and also for system diagnostic and information screens). This new soft GPU driver will also be available in Windows 8, where it will replace the standard VGA driver that has been a part of Windows for just about as long as Windows has been around.

WOA systems will also require UEFI firmware and Trusted Platform Module (TPM) hardware across the board to support its secure boot and data encyption features, both of which will also be available to Windows 8 devices with the correct hardware (TPMs have been used to encrypt hard drives with BitLocker since Windows Vista and UEFI is slowly replacing BIOS in OEM PCs, but Windows 8 should push the adoption of both in a wider range of computers).

Conclusions

To see what Microsoft is trying to do with Windows on ARM, the most applicable template to examine is the one the company followed with Windows Phone 7. In both cases, Microsoft is entering an established market where competitors have established footholds through very different strategies (in each case, Apple and its tighty-controlled iOS on one end, Google with its infinitely malleable Android on the other, and a few other competitors fighting for scraps in between) and has tried to forge a middle path. Windows Phone 7 has been a bit of a slow starter because of Microsoft's low profile in the smartphone field and because of some lackluster handsets, but the platform has some very vocal fans - if the company can achieve a similar balance in Windows on ARM and get it to market on competitive hardware by the end of the year, that (combined with Android's relative weakness in the tablet market so far) might just be enough to establish Windows as a major player in the tablet space.

As usual with these Building Windows post summaries, I've relayed and distilled the most pertinent information for Windows users and enthusiasts here. If you'd like to read the full post, which also includes some details about how Microsoft is testing Windows on ARM in its labs and some of the more technical details involved in "porting" Windows from x86 to ARM, it is linked below for your convenience.

Source: Building Windows 8 blog

With the emphasis on smartphones and tablets at this year’s CES, it should come as no surprise that the various SoC IP developers are focusing their announcements around the show, and Imagination Technologies is among them. In 2011 Imagination announced their next generation of SoC PowerVR GPUs, the Series 6 family, based on the PowerVR Rogue architecture.  Now just under a year later Imagination has announced that they’ve officially released their first GPU designs to licensing for inclusion into SoCs.

Shedding more light on feature support for the first time, Imagination has announced that the baseline graphics feature set for Series 6 will include support for OpenGL ES “Halti”, the current working name for ES 3.0, itself derived from OpenGL 3.x. In terms of DirectX generations, this would make Series 6 a DirectX 10 part, analogous to the GeForce 8/9/200 series, the Radeon 2000-4000 series, and Intel’s HD2000/3000 iGPUs. Interestingly enough Imagination will also be offering designs that are DirectX 11.1/OpenGL 4.x compliant, which would bring them to parity with the very latest GPUs from AMD and NVIDIA.

Meanwhile on the compute side OpenCL will also be supported, and while Imagination doesn’t list the specific version we believe they will be conformant up to version 1.1. Microsoft’s DirectCompute is not specifically mentioned, however at a minimum the DX11.1 parts would need to support it.

Unfortunately at this time Imagination is still playing their cards close to their chest, so while we know what APIs Series 6 supports, we don’t know much about the configuration of the first two GPUs: G6200 and G6400. G6200 features two “compute clusters”, while G6400 features four of them, though beyond shader blocks we don’t know what a compute cluster entails.

PowerVR Series 5XT Architecture Diagram

Most likely Imagination is configuring their GPUs in a method similar to the SGX543 series, where a fixed frontend is coupled with a specific number of shader blocks and ROPs, along with several fixed function DSPs. The biggest question perhaps is whether Series 6’s geometry performance will once again be fixed; SGX543 only scaled the number of USSE2 pipelines, so while Imagination could grow the number of pixels they could deal with geometry performance was solely a function of a given GPU’s clockspeed.

Long term Imagination is planning to have designs that offer up to 1 TFLOP of shader performance, which would be nearly 10 times the theoretical shader performance of the SGX543MP16. The initial G6200 and G6400 will be much more conservative, though we don’t have specific performance estimates for them yet.

Finally, as Imagination is an IP vendor, there isn’t any kind of timeline on availability as this is up to their customers. The only SoC announced to use Series 6 so far is ST-Ericsson’s Nova A9600, which is not scheduled to arrive until sometime in 2013. Given the fabrication ramp-up schedule for most SoCs, any Series 6 equipped SoC is still a year out if not more; in the meantime there are still a number of ARM A15 + SGX543/544 scheduled for later this year. And for larger, more capable GPUs such as the SGX545 the release gap has been closer to 2 years, so DirectX 11.1 SoCs in particular are almost certainly 2014 products assuming Imagination gets a DX11.1 GPU design out this year.

We’ll have much more on Series 6 later this year as further designs are announced and Imagination publishes more details about the underpinnings of their PowerVR Rogue architecture, so stay tuned.

It wouldn't be far off the mark to call Google TV as one of the unmitigated disasters of 2010 - 2011. Through the failure of the Logitech Revue, it was responsible for Logitech's below-par performance last year, and also for the stepping down of its CEO. Anand covered Intel's winding down of the Digital Home Group and it could be said that Google TV / Intel's concept of Smart TV not taking off as expected was one of the reasons.

However, Google doesn't give up on its efforts without a fight. With access to the Android market and an upgrade to Honeycomb, Google TV received some life support last October. However, pricing and device power consumption were the two other prime factors which needed to get addressed. In order to take care of these factors, it was inevitable that Google and its partners would end up moving to an ARM based platform. Given that ARM has remained the architecture of choice for Android smartphones, this was also a move predicted by many.

We covered Marvell's foray into the DMA (Digital Media Adapter) market with their ARMADA 1000 platform. Today, Marvell is officially launching the next generation ARMADA 1500 (88DE3010) SoC. They also announced their team up with Google and indicated that all the Google TV boxes at the 2012 CES would be powered by Marvell silicon. Read on for our analysis.
 

Later today AMD will be releasing the first preview for their Catalyst 12.1 driver set. AMD has been going through preview/beta drivers at a rapid pace in the last couple of months – we’ve seen 3 different 11.11 preview drivers in as many weeks – and as 11.12 nears, AMD is preparing for what 2012 and the Catalyst 12.x series will bring. It may sound like hyperbole to say that 2012 will be the biggest year yet for AMD’s Catalyst driver team, but it’s the truth. Graphics Core Next will be the biggest GPU architecture change for the company since R600 (2900XT) nearly 5 years ago, bringing with it a great deal of backend driver work that needs to be done, while the frontend team has their own goals and aspirations.

At the same time it’s going to give AMD the chance to close the book on 2011. 2011 brought with it some great developments out of the Catalyst team such as significant performance boosts for both Cayman (6900 series) GPUs and CrossFire across the board, while other groups delivered on more consumer-facing features such as SteadyVideo to go along with the launch of the Llano APU.

But 2011 also brought with it some technical debt and some reputational debt, all of which needs to be paid in 2012. AMD outright blew the launch of Rage by posting a faulty driver, Battlefield 3 in CrossFire mode did not work out of the box (i.e. without microstutter) even with nearly a month-long public beta and AMD’s close relationship with DICE, and CrossFire support for The Elder Scrolls V: Skyrim took the better part of a month to reach Radeon HD 5000 series owners. Not to kick the Catalyst team while they’re down, but for all that went well for them in 2011 they failed in other areas where they could least afford it. Thus 2012 becomes all the more important for AMD as they need to erase their debts from 2011.

Erasing those debts starts today for AMD, with the release of the Catalyst 12.1 preview driver. 12.1 won’t be the driver that buys AMD redemption – I think 11.11c is more important in that respect – but it is the driver that sets the pace for the year. And quite frankly it’s the driver that’s going to buy AMD a lot of goodwill, even if it only brings with it a few changes.

Custom Application Profiles As Implemented By NVIDIA

Of those few features I’m going to immediately dive in with what I think is the headline feature: custom application profiles. Ever since NVIDIA introduced custom application profiles so many years ago I have been a firm believer in their importance for GPU enthusiasts. While most games have been good about implementing anti-aliasing and anisotropic filtering controls, that’s about as much progress as they’ve made. With the introduction of driver enhancements like Adaptive/Transparancy anti-aliasing, coverage sample/EQ anti-aliasing, tessellation clamping, and the widespread use of multi-GPU, the idea that you can set & forget your drivers on a global level has become antiquated. These features deserve to be used, and custom application profiles are the most efficient way of using them.

For more than 4 years now I’ve asked AMD for this feature – in meetings and in articles – but it hasn’t been something where we’ve seen eye-to-eye. AMD made some progress in 2010 with the introduction of Catalyst Application Profiles (CAP) to distribute out of band profile updates, and while CAP was a big step forward for AMD, the C I was looking for was custom. Tools like Radeon Pro have filled the gap in the meantime, but it’s never the same thing as having such functionality built into the driver itself, especially when 3^rd party tools will never have the reach of 1^st party tools.

With Catalyst 12.1 AMD is finally taking application profiles to their logical extension by allowing for custom application profiles, and I couldn’t be happier. As is the case with NVIDIA, AMD is allowing users to create new application profiles and to modify the application profiles distributed through drivers and CAP updates. This not only includes settings traditionally available through the driver, but for the first time AMD is opening up CrossFire – you can now force various CrossFire modes by using a custom profile.

Breaking things down a bit, if you have used NVIDIA’s custom profiles in the past then you should find the functionality nearly the same. All of AMD’s control panel settings can be saved to a custom profile which will then be used alongside the game the profile is for. For example this allows for forcing MSAA in Starcraft II or clamping tessellation factors in HAWX 2 without the need to set (and then unset) these features at a global level. If you’re an image quality purist, and particularly if you’ve spent a significant amount of money on GPUs to achieve this, then the value of custom profiles cannot be understated.

As for multi-GPU users, they will be the other significant group to benefit from custom profiles. Previously if you wished to force CrossFire on an unsupported application you needed to rename the executable to match a game AMD had a profile for, and then hope that specific CF mode worked. With custom profiles AMD is enabling several different CF modes: default (which uses whatever CF profile AMD has defined for the game), AFR Friendly (forced AFR), Optimize 1x1, and Use AMD Pre-define Profile, which allows a custom profile to have a CF mode from another game mapped to it (similar to NVIDIA’s SLI compatibility bits). Even disabling CF on a per-profile basis is an option here, though we found that it suffers from the same quirk that NVIDIA’s implementation does: the second GPU is decoupled but CF isn’t actually disabled, so games that can detect CF (such as Crysis) will follow their AFR friendly render paths as they still see CF enabled.

At this point all of the necessary functionality is present and accounted for, and in our tests we’ve found it to work without any hitches. AMD is finally at parity with NVIDIA in providing this small but crucial feature.

With that said, while AMD has done a great job implementing the functionality of custom application profiles the interface could use some further work. The whole implementation still feels like it’s been shoehorned into AMD’s existing 3D Applications Setting panel; AMD doesn’t sufficiently separate the concept of global and custom profile settings, as you use the same control panel to make changes to both types of settings. It’s possible (and likely) that you’ll accidentally set your global settings at least once when trying to save a custom profile.

Furthermore whereas NVIDIA uses application detection to pre-populate a list of profiles, AMD has no such detection. In order to create a profile you need to first select your settings in the 3D Application Settings panel and then save those settings to a new profile, a process that involves hunting down the executable of the game. Of course NVIDIA’s detection system isn’t perfect and you’ll have to follow a similar process at times, but if you have a large Steam library you’ll appreciate not having to drill down through several directories to find the right executable for each game.

Once a custom profile has been set however, AMD actually has a second panel that lists all of the custom profiles and their settings, and allows you to delete them. Note that this is just a listing of custom profiles, so pre-defined profiles continue to remain hidden. Custom and pre-defined profiles play well together for the most part, although if you create a custom profile for a game that already has a pre-defined profile AMD will warn you that the custom profile will override the pre-defined profile.

Overall if you’re a previous NVIDIA user who has missed custom application profiles you should be quite content with AMD’s latest addition. Otherwise if you’ve never had the opportunity to use custom application profiles before then you’re in for a treat.

Rounding out the changes to the Catalyst Control Center, along with the addition of custom application profiles AMD has also made some minor tweaks to the Video Color and Video Quality control panels. There’s no new functionality to speak of, but they have been tweaked to simplify their use.

Finally, outside of the CCC AMD has also added a couple new features to their driver, along with some specific performance enhancements. 3D users will find that quad-buffer (gaming) 3D finally works in conjunction with CrossFire, while TV users will find that AMD now supports frame-packed 3D over HDMI at 1080p30, on top of their existing support for 1080p24 and 720p60. A quick check of the HDMI specification lists frame-packed 1080p30 as an optional (secondary) resolution, but it’s there for the TVs that support it. Meanwhile for performance AMD is still hard at work on Skyrim; 6900 series users can look forward to up to 10% better performance in Skyrim when using MSAA alongside CrossFire.
 

Update: Released. http://support.amd.com/us/kbarticles/Pages/Catalyst121Previewdriver.aspx

In two deals announced late last week, Verizon Wireless has expanded its spectrum holdings through deals that will give it control over various frequencies, almost all in the AWS space. This in the same week that AT&T had the door nearly entirely shut on them in their efforts to merge with T-Mobile. Though AT&T touted its desire to increase competition, produce more jobs in the United States and provide more and better services for its customers, spectrum was the real target of this acquisition. Ever hungry mobile data customers are the drive for bandwidth, and falling short of meeting your customers needs can be costly to a provider (just ask AT&T). There are several techniques that can be deployed for encoding more bits into each wave, but ask a firefighter what he wants to help fight a fire. More hoses will be the answer everytime, and in wireless that means more spectrum.

Where We Are, Where We're Going

Remember this guy? When Brian spent some time clarifying the HSPA+ state of the HP Veer 4G, he discussed modulation schemes used in current mobile data networks. The most efficient scheme currently in use is 64QAM, which provides enough density to encode a 6 digit binary number into a single wave. The ability to discern similar points in a QAM constellation becomes harder as they become more dense, and with HSPA+ and LTE we are approaching the limits of how spectrally efficient we can be. Going back to our firefighter analogy, firehose comes in various diameters. The larger the diameter, the harder the hose is to control. Implementing 128QAM would add another bit to each Hz, but increase the burden on each radio to discern between increasingly dense points on the plot. 

What's next then? LTE-Advanced will fulfill the true potential promised by 4G mobile broadband networks, and will do so through a more crowded and complex arrangement than we have now. Right now, your phone is picking up signals from, most likely, more than one cell tower owned by or servicing your network provider. These towers serve many customers over a wide geographic area, and are referred to as macrocells. Your phone utilizes the tower that will provide it the fastest and most reliable connection, and this determination is made using several factors including interference from other nearby towers. If you are lucky, your phone will have (if supported) a MIMO connection with low SNR resulting in a high speed, low latency connection. 

In Qualcomm's LTE-Advanced scheme, users would be served by a heterogenous network of relay basestations and pico- and femtocells, all operating together to provide fast mobile data that can service ever more customers at once. Unlike the macrocells, these small cells are distributed in an ad hoc manner, being placed in hotspots or coverage gaps. We've seen some of this in the deployment of picocells within AT&T and Apple stores in congested areas. It isn't enough to deploy these cells, though; it also becomes necessary to coordinate their operation with the macrocells, and this is the next step. 

Verizon currently operates their network in a 22MHz chunk of the 700MHz spectrum, this is divvied up in a frequency duplexed scheme with 10MHz for downstream, 10MHz for upstream, and two 1MHz chunks at the sides. This all fits nicely with LTE's ability to use varying sized channels, in this case 10MHz serves upstream and downstream fine. Lower frequencies propagate through buildings better, so 700MHz is beneficial in urban areas where customers might not get any signal from higher frequency towers.

Now, add in some lower power small cells throughout the range of a single macrocell. Many of these smaller cells will be within buildings, or covering outdoor hotspots (think Times Square and Madison Square Garden). Those smaller cells don't necessarily need to worry about building penetration as they are either outdoors or already within the building they're trying to serve. As a result,  that 700MHz frequency becomes less important, though they may still operate on them. Now, add in some of the channels that Verizon is acquiring; particularly a 20MHz chunk of AWS in Minneapolis, MN. Imagine that the whole channel is added for downstream operation. You now have multiple cells, with several channels to choose from, all operating in one geographic area. And that's just for LTE, the PCS spectrum they acquired could be added to their 3G and telephony networks for enhanced voice service and fallback data. One more picture to look at. 

Those 20MHz chunks are where LTE really shines. By moving to 20MHz channels, current generation (Cat. 2) devices could more regularly hit their bandwidth limits of 50Mbps and next gen devices (Cat. 3/4) could jump to 100-150Mbps. 

The New Land Grab, Spectrum Acquisitions

Now let's breakdown the two deals Verizon's made, and hope will get approved. Cricket Wireless is a regional operator that provides cellular services on spectrum it owns or leases within a given market. This is similar to ClearWire, though Cricket provides telephony services along with data services. Cricket, like ClearWire, intends to move to LTE and wants some of that coveted 700MHz spectrum. It just so happens, Verizon has 12MHz it isn't using in Block A of the 700MHz spectrum covering the Chicago area. So, Cricket gets 12MHz of building penetrating frequencies in the 3rd largest metropolitan region in the US, and Verizon gets 23 PCS and 13 AWS (Advanced Wireless Services) licenses in markets spread across the US. Since each chunk is regional, this won't mean that Verizon will have dozens of 20MHz channels blanketing the whole country, but with this deal alone they will have many 20MHz channels over many markets.

And what's so good about AWS? Mainly, it's available. AWS exists in the microwave spectrum, with downstream bandwidth provided between 2110MHz and 2155MHz, and upstream bandwidth provided between 1710MHz and 1755MHz. It was first put up for auction in 2006, and was almost entirely scooped up by T-Mobile for its 3G network. 

The key to this deal is that Cricket isn't actually doing anything with these channels, nor is Verizon doing anything with their 12MHz. The FCC generally frowns on deals that will adversely affect customers of a service. So if Cricket were giving up their only operating frequency in Fresno, CA, there might be some push-back. As it is, the FCC shouldn't have anything to complain about. Indeed, by exchanging 700MHz spectrum they are basically creating an honest to goodness competitor in the Chicago area. And just to bear out how important these AWS acquistions are, Verizon is also throwing in $100 million to help build out Cricket's LTE network. 

Spectrum Co. is a joint venture between Comcast, Time Warner Cable and Bright House, and was formed to manage a large portfolio of spectrum that could be used by the wired telcos to branch out into wireless services. The venture went nowhere, and it seems the trio is ready to cash out. So, for $3.6 billion Verizon will be purchasing 122 AWS licenses that undoubtedly cover some of the largest markets in the US. This won't be a straight sale, either. Verizon is essentially entering a partnership with Spectrum Co. that will enable them to directly sell Verizon services, or purchase them wholesale for use under their own branding. 

Since all 122 of these licenses were being unused, the FCC won't be concerned about a loss of service to customers. What might strike some alarms is the notion that the largest cable telcos are going to be locked into offering only one wireless provider's services. This is obviously a strategic win for Verizon, but might be looked down on by regulators. It will be curious to see whether this is the end of Verizon's buying spree, there's only so much spectrum out there. 

And AT&T? AT&T is awaiting approval from the FCC on their purchase of Qualcomm's MediaFLO 700 MHz spectrum. Qualcomm having bowed out of the mobile TV business is letting this slice of spectrum sit idle, so this deal should be approved any day now. The purchase of T-Mobile is a different story altogether. Though becoming the number one wireless provider in America has its appeal, the acquisition of spectrum is the driving force behind this deal. Combining AT&T's block of 700 MHz spectrum with T-Mobile's nationwide AWS spectrum would boost their LTE competitiveness. Mergers of this type receive thorough evaluation from technical, legal, consumer and anti-competitive perspectives. In their recent Staff Report, the FCC opined that while AT&T's competitiveness will improve, the merger will negatively affect the public and lead to higher prices. The release of this report succeeded AT&T and T-Mobile withdrawing their application with the FCC, after Chairman Genachowski requested that the merger be put through a hearing to determine approval. AT&T will continue their merger efforts with the Department of Justice, in the hopes that a positive result there could sway the FCC in a future submission. So overall, it looks a little grim at AT&T right now.

Wrap-up

LTE isn't going to get too much better at using the spectrum we feed it. To that end, if more and more customers are going to be demanding more and more mobile data, the only solution is to feed the beast. Verizon had a headstart with the first US LTE network, which celebrated its first birthday the other day. Now Verizon is taking the next step and, should the AT&T/T-Mobile merger fall through, may end up with a commanding lead in LTE spectrum. What remains to be seen is how this will play out for the consumer. If wireless providers follow the lead of their wired kin, we could end up with high prices to take advantage of this high performance. That sounds logical, until you see how poorly US broadband speeds scale with their price; $200 for 105 Mbps just isn't balanced against $42 for 1.5 Mbps.

We discussed the availability of AMD branded memory modules earlier this month, but today AMD is officially unveiling information on their memory platform. There are a few major questions many will have: why is AMD entering the memory market at all, and what do they hope to offer that we can’t already get from other vendors? Let’s take those in turns.

The reason for AMD’s entry into the memory market comes from two areas. First, AMD’s APUs are now shipping in large volumes and can definitely benefit from higher bandwidth memory modules. We’ve already shown the sort of performance scaling you can get from an A8-3850 with higher clocked DRAM, but many people buy A-series APUs as part of a prebuilt system, and right now lots of OEMs are still cutting corners on the RAM and using DDR3-1333. That’s the second aspect of the move: AMD wants to enable a [buzzword alert!] “holistic customer platform experience”, and they may be able to help drive down costs for AMD platforms. A final element AMD mentions is a desire to drive and enable future memory product developments.

The other item to discuss is what AMD offers that we may not already have. Here the distinction between AMD branded memory and other options isn’t quite so clear, but AMD will be doing testing and validation in their labs using AMD platforms. AMD also notes that they will not be using any ETT (Effectively TesTed) or gray market RAM. The latter is used as a term to collectively group hardware that may be less desirable; as an example, Intel unboxed CPUs are “gray market” because they are intended for OEM use but can still end up being sold at retail. Basically, gray market parts would cut out some of the supply channel (in the example just cited, gray box processors typically cut out AMD/Intel and only have a short warranty from the seller). ETT parts on the other hand are a way of cutting costs by skipping branding; the RAM is still tested and is supposed to be high quality, but without branding it’s one small way to reduce costs. Generally speaking, ETT memory is destined for value RAM modules, so basically AMD is saying is that their AMD RAM will start out a step above value RAM. AMD also states that they will take end-to-end ownership of the AMD Memory ecosystem, working with module manufacturers, memory partners, IC partners, distributors, and VARs (value added resellers).

With that out of the way, let’s discuss the specifics of what AMD Memory will be available and the target markets. Here’s a slide from AMD’s presentation summarizing things:

 As you would expect from any memory, the AMD RAM will work with both AMD and Intel platforms; the main difference between the tiers will be the speed and packaging. Entertainment Edition memory will target the mainstream/value segment, come in single 2GB and 4GB DIMM packages, and is rated for CL9 operation at DDR3-1333 and/or DDR3-1600; Entertainment Edition memory is already available, starting in October. The Performance Edition memory should start shipping this month, and it will come in 2GB, 4GB, and 8GB capacities (these are presumably two-DIMM kits with 4GB, 8GB, and 16GB total capacities; Bulldozer could potentially use four-DIMM kits). The main difference with Performance Edition memory is that it is rated for CL8 operation at DDR3-1333/1600 speeds. Last is the Radeon Edition memory, which will come in 4GB and 8GB kits and offer DDR3-1866 and up to DDR3-2133 support with CL9 operation (and presumably CL7/8 operation at lower speeds). The Radeon Edition parts will also have support for overclocking via AMD OverDrive software; availability is expected in Jan/Feb 2012.

So what does all of this really mean? That’s the difficult part. If all AMD memory supported speeds of at least DDR3-1600, that would be a clear break from the current offerings, but the press release indicates that there will be both DDR3-1333 and DDR3-1600 parts. The upgrade to DDR3-1600 provides a significant performance increase; we linked our Llano A8-3850 article above showing some of our own results, but here are some charts of our testing along with AMD’s results:

AMD shows up to a 20% performance increase in their testing by upgrading from DDR3-1333 to DDR3-1600, while our own results show an average increase in performance of around 14% across seven tested games (with a range of improvement of around 8% to 41%). Should you choose to spring for faster DDR3-1866 memory (or just overclock some decent DDR3-1600 RAM), the average performance increase is around 20% and up to 40% in some cases (or as low as 8% in Civ5). This isn’t too surprising as the AMD Fusion GPUs are significantly faster than competing solutions and the combination of shared memory bandwidth with the rest of the platform along with generally slower memory speeds (compared to dedicated GPUs) is a double-whammy. So why would AMD continue to sell anything less than DDR3-1600? Your guess is as good as mine.

Several of us have chatted about the AMD Memory announcement, and really we’re not quite sure if this is necessary or useful. If it means systems with better quality and higher performance RAM at the same price, that would be a good thing, but the persistence of DDR3-1333 for desktop parts doesn’t jive with that goal. What’s more, RAM prices are already incredibly low, so AMD entering a commodity market doesn’t appear to be a good way to improve the bottom line.

AMD’s first partners for their branded memory initiative are Patriot Memory and VisionTek, with Patriot being a familiar name to memory shoppers and VisionTek known for their graphics products. There’s nothing inherently wrong with AMD branded memory, but unless the price is lower than existing options (e.g. AMD mentions bundles as something we’re likely to see), there’s also not much that it adds to the market. For now, we’ll stick with recommending you buy RAM that will supports at least DDR3-1600 speeds if you’re buying a Llano (or future APU) system; whether that memory is AMD branded or otherwise will likely be far less important than how much the memory costs for the desired level of performance.

Passively cooled high performance GPUs are quite popular with the HTPC community. NVIDIA GPUs are preferred by many HTPC users because of good software support (LAV CUVID, for example) and the ability to use custom renderers like madVR without losing out on hardware decode acceleration. I have already covered this in detail in a previous piece.

A look at the list of passively cooled GPUs on Newegg reveals that higher end NVIDIA GPUs are not represented well. In fact, we have a number of GT 430 and GT 520 passive models, but only one GT 440 model. On the other hand, AMD's GPUs seem to be quite popular in this space. We have a large number of 6450s. There are two models each of the 6570 and 6670. The 6750, 6770 and even the 6850 have one passively cooled model each.

Zotac is trying to level the playing field here with the introduction of a passively cooled GTS 450.

The GTS 450 Zone Edition comes with a GTS 450 GPU (192 shaders) underclocked to 600 MHz / 1200 MHz. The 128-bit 1 GB DDR3 memory runs at 1333 MHz. Unlike other GTS 450 units, this one will not require a PCI-E power connector. The GTS 450 Zone Edition will have a MSRP of 99 Euros in the EU (with the pricing in the NA market yet to be determined). The unit is currently shipping to retailers and is expected to be out on sale in time for the Christmas shopping season.

Given that even the NVIDIA GT 5xx models seem to be looking a bit dated right now, we asked Zotac as to why this cooling mechanism wasn't put on one of the more recent NVIDIA GPUs. It appears that the thermal limitations of passive cooling required underclocking which NVIDIA wouldn't allow on the 500 series.  We are sure this will turn out to be better than the GT 430 models we have been recommending for HTPC use so far (particularly if you want to use madVR with 1080i60 streams). Will the lower speed DDR3 memory and core clock speeds hurt it when compared to the passively cooled GT440 (for HTPC purposes) ? We will know as soon as the card hits the market.

We’re here on the USS Hornet attending NVIDIA’s GeForce LAN 6 event, where NVIDIA has just finished a kick-off keynote and product announcement between rounds of gaming. While NVIDIA has held LAN parties in the past, they don’t traditionally use them to announce new products. But the reality of the GPU product cycle is that with Kepler due in 2012 NVIDIA won’t be launching any major new consumer GPUs this fall, so instead the fall is being dedicated to their ecosystem products and GeForce LAN 6 is being used as the launch event for those products. So while today’s announcement isn’t a new GPU, it is still quite relevant to gaming.

This evening at GeForce LAN 6 NVIDIA has announced a pair of technologies: 3D Vision 2 glasses, and 3D LightBoost. The 3D Vision 2 glasses are the long awaited replacement for NVIDIA’s earlier 3D Vision wireless glasses, and feature a new fit and larger lenses. Meanwhile 3D LightBoost is an interesting monitor adaptation that increases the amount of light that get through NVIDIA's 3D Vision glasses by turning a monitor's backlight off and on to allow the glasses to stay open longer. The two are being introduced tonight as complementary technologies.

A while after the iPhone 4S announcement, I posted a piece on the iPhone 4S' cellular architecture and talked about the new inclusion of HSPA+ alongside CDMA2000 1x/EVDO and some specifications. Since then, there's been some ongoing confusion about whether the device supports HSPA+, whether it's "4G," and just what all that really means. 

It's funny because this issue came up with the HP Veer 4G on AT&T a while back, and with that particular device I did a similar explanatory article in the context of a This Just In post. I wager the somewhat limited cross shopping (and commercial success of the Veer) resulted in many iOS users missing that discussion, and it's cropping up again for the iPhone 4S. The confusion is really twofold. First is a misunderstanding about 3GPP releases and what features are optional or mandatory. Second is a misunderstanding about HSDPA, HSUPA, and modulation coding schemes. Let's break it down. 

UMTS standards are defined by an organization known as the 3rd Generation Partnership Project (3GPP), which is a body comprised of partners like telecommunications companies, wireless carriers, and cellular hardware manufacturers. Standards are important, and their formation requires collaboration and discussion. Major revisions are finalized, synchronized, and become "releases." For example the first releases define GSM, release 99 defines UMTS, release 5 defines HSDPA, 6 defines HSUPA, 7 - HSPA+, 8 - LTE, and 10 - LTE Advanced.

This is a huge oversimplification, but you can see how major milestones are bundled up and become releases, and thus when baseband manufacturers like Infineon, Qualcomm, or ST-E release hardware, they're said to be compliant with a given release, and inherit the big features from releases before it. 

Where confusion arises is that you don't need to include every feature from some release to say that your hardware is compatible with a given release, and really only features from release 99 are mandatory if you're out to make some UMTS hardware. So when a baseband is "HSPA+" (which really is a colloquial name for the improvements to HSPA made in 3GPP release 7), it doesn't need to include every last feature. Features included in release 7 include things like reduced signaling, faster call setup, continuous packet connectivity, reworked RCC states (idle, DCH, FACH, e.t.c), MIMO, and higher order modulation support. On the downlink, 64QAM or MIMO gets added, and on the uplink, 16QAM gets added. 

The confusion in the case of many HSDPA 14.4 (Category 10) devices seems to center on whether it is or isn't HSPA+ if it doesn't include 64QAM support on the downlink (which starts in category 13). This is one of those few times that looking at the Wikipedia table for HSDPA UE categories can be misleading. UE categories with 64QAM are introduced in 3GPP release 7, however it's up to manufacturers to include or not include that feature when actually making hardware. It's interesting to me that all of this misinformation comes up over HSDPA UE category, and somehow HSUPA UE category never gets brought up - I haven't seen anyone claiming that some baseband isn't HSPA+ because it excludes 16QAM on the uplink, yet the overwhelming majority of devices are just using QPSK. 

Apple hasn't stated what baseband is in the iPhone 4S (and I guess we won't know absolutely for certain until it gets opened up), but it is almost without doubt MDM6600 just like the Verizon iPhone 4. The specs all line up, and while that part doesn't include 64QAM on the downlink, it does include a number of the other HSPA+ features I've mentioned, and is thus 3GPP Rel.7 and therefore "HSPA+." 

Actually using these higher orders of modulation requires favorable radio conditions, of course, and the result is that 64QAM on the downlink or 16QAM on the uplink only gets used in very well tuned systems or cell center. There have been some drive tests done which have shown that 64QAM only really gets used 10% of the time on average. In the iPhone 4S' case here in the USA, it's silly to argue about whether or not the phone can decode 64QAM if AT&T doesn't even use it. A few Samsung devices so far have been based on ST-E's Thor 5730 HSDPA 21.1/Category 14 baseband, and include an engineering menu which shows the percentage of frames encoded using QPSK, 16QAM, or 64QAM. 

Note that 64Q is 0. I have run many other tests where 16Q is 100.

I've shown a few other interested people how to see this menu on the Samsung Infuse 4G on AT&T, and thus far none of them has seen any 64QAM frames show up. I've never seen 64QAM pop up either, even with line of sight to the tower in the middle of the night. Those markets include my own in Tucson, AZ, and New York City, which are both AT&T "4G" markets (connoting HSPA+ with improved backhaul) on the coverage viewer. Two data points isn't enough to say conclusively that AT&T isn't using 64QAM, but I haven't seen it yet. You can also see this on video from an Infuse in NYC. Not all AT&T WCDMA "3G" markets are HSPA+, but if you're in a market that AT&T has marked with dark blue (and calls "4G") you can be pretty much assured it's HSPA+.

So the next question is when HSPA+ is and isn't real 4G, and obviously the bar right now is LTE. We've gone over all the real ITU decisions and such in our LTE piece, (whose definition does include DC-HSPA+ and LTE instead of previously just LTE Advanced) but what matters is actual throughput. There's credibility to HSPA+ being comparable to LTE speed as soon as DC-HSPA comes around which aggregates together two 5 MHz WCDMA channels. With all the same features (MIMO, 64QAM, and wider channel bandwidth), DC-HSPA+ gets you approximately to the level of theoretical maximum throughput you can get on LTE. It's easy to see how things are arguably fairly comparable when you consider Verizon uses 10 MHz FDD, and DC-HSPA+ aggregates together two 5 MHz WCDMA channels. Throw in MIMO in addition to just dual-carrier, and maximum theoretical throughput lines up pretty closely as shown in this table by Qualcomm:

That's saying nothing about network latency, however, where LTE will continue to outshine HSPA+. In that case, it does arguably make sense to associate DC-HSPA+ with real LTE and call it real 4G, but until DC-HSPA+ or at least HSPA+ with MIMO it is almost without doubt still 3G. Obviously USA carriers' attempts to pass off HSPA+ 14.4 or even 21.1 as "4G" is immediately laughable marketing, but DC-HSPA+ is arguably real 4G.

Lastly, rumor has it that AT&T and Apple are "working together" to change the 3G status indicator to "4G" for AT&T in the US, a move that would mark the first carrier intrusion into the otherwise completely carrier-untouched iOS. I don't have issue with the status indicators showing H or H+ (for HSPA and HSPA+ when appropriate), but misleading indicators do nobody justice and just serve to reinforce these kinds of misconceptions. 

Western Digital jumped on to the local media streaming bandwagon quite early in the game. With better codec compatibility and a more stable platform compared to other vendors, the WDTV lineup caught the imagination of the consumers. Over the last few years (coinciding with the rise of Netflix), US consumers have tended to prefer over the top streaming solutions. Recognizing this trend, Western Digital has been strengthening their media streamers portfolio with the appropriate features.

Today, Western Digital is introducing the latest addition to their WDTV lineup, the new WD TV Live Streaming Media Player. Its roots (in terms of both hardware and firmware) seem to lie in the WDTV Live Hub which received praise from us last year. The industrial design has been updated, and the main hardware addition seems to be the integrated wireless network capability. The internal hard drive has been removed. Local media streaming compatibility seems to be retained.

Like the Roku 2 XS, the new WD TV Live has support for single band (2.4 GHz) 802.11n wireless only. However, the network port is Gigabit. Also, Western Digital PR claims that the new WD TV Live (as well as the WDTV Live Hub) supports 1080p video along with Dolby Digital Plus audio for selected Netflix titles. This is something we hope to evaluate and report back soon.

The new premium service making an appearance in the WD TV Live (as well as the Hub) is Spotify. As the PR reproduced below shows, the unit offers the full Spotify experience including account management and the social features (such as sending songs to your friend's inbox). In addition to YouTube, the new media player (as well as the WDTV Live Hub) gains access to DailyMotion, a service quite popular in Europe. The last few firmware releases for the WDTV Live Hub have also brought some simple games like Sudoku and Texas Hold'Em.

Priced at $99.99, this unit seems to cover all the media streaming aspects missed by the Roku 2 XS. It probably doesn't have the fan following of the Roku or the developer support that Roku provides. We also don't have much hope of it solving the DTS-HD MA bitstreaming issue. However, for the general consumer, Western Digital seems to have lined up a winner at a sweet price point (they could have chosen a better name, considering that the WD TV Live was the name given to the second generation player in the WDTV family). Stay tuned for our review of the new WD TV Live Streaming Media Player.

We have covered the powerline networking battle between HomePlug and HomeGrid in a lot of earlier pieces. With demonstration of working silicon at CES 2011, G.hn finally emerged from vaporware territory. Today, Marvell is announcing their first G.hn transceiver chipset, the 88LX3142. Read on for our analysis.

Last week NVIDIA provided an update on its Tegra SoC roadmap. Kal-El, its third generation SoC (likely to launch as Tegra 3) has been delayed by a couple of months. NVIDIA originally expected the first Kal-El tablets would arrive in August, but now it's looking like sometime in Q4. Kal-El's successor, Wayne, has also been pushed back until late 2012/early 2013. In between these two SoCs is a new part dubbed Kal-El+. It's unclear if Kal-El+ will be a process shrink or just higher clocks/larger die on 40nm.

In the smartphone spirit, NVIDIA is letting small tidbits of information out about Kal-El as it gets closer to launch. In February we learned Kal-El would be NVIDIA's first quad-core SoC design, featuring four ARM Cortex A9s (with MPE) behind a 1MB shared L2 cache. Kal-El's GPU would also see a boost to 12 "cores" (up from 8 in Tegra 2), but through architectural improvements would deliver up to 3x the GPU performance of T2. Unfortunately the increase in GPU size and CPU core count doesn't come with a wider memory bus. Kal-El is still stuck with a single 32-bit LPDDR2 memory interface, although max supported data rate increases to 800MHz.

We also learned that NVIDIA was targeting somewhere around an 80mm^2 die, more than 60% bigger than Tegra 2 but over 30% smaller than the A5 in Apple's iPad 2. NVIDIA told us that although the iPad 2 made it easier for it to sell a big SoC to OEMs, it's still not all that easy to convince manufacturers to spend more on a big SoC.

Clock speeds are up in the air but NVIDIA is expecting Kal-El to run faster than Tegra 2. Based on competing A9 designs, I'd expect Kal-El to launch somewhere around 1.3 - 1.4GHz.

Now for the new information. Power consumption was a major concern with the move to Kal-El but NVIDIA addressed that by allowing each A9 in the SoC to be power gated when idle. When a core is power gated it is effectively off, burning no dynamic power and leaking very little. Tegra 2 by comparison couldn't power gate individual cores, only the entire CPU island itself.

In lightly threaded situations where you aren't using all of Kal-El's cores, the idle ones should simply shut off (if NVIDIA has done its power management properly of course). Kal-El is built on the same 40nm process as Tegra 2, so when doing the same amount of work the quad-core chip shouldn't consume any more power.

Power gating idle cores allows Kal-El to increase frequency to remaining active cores resulting in turbo boost-like operation (e.g. 4-cores active at 1.2GHz or 2-cores at 1.5GHz, these are hypothetical numbers of course). Again, NVIDIA isn't talking about final clocks for Kal-El or dynamic frequency ranges.

Five Cores, Not Four

Courtesy NVIDIA

Finally we get to the big news. There are actually five ARM Cortex A9s with MPE on a single Kal-El die: four built using TSMC's 40nm general purpose (G) process and one on 40nm low power (LP). If you remember back to our Tegra 2 review you'll know that T2 was built using a similar combination of transistors; G for the CPU cores and LP for the GPU and everything else. LP transistors have very low leakage but can't run at super high frequencies, G transistors on the other hand are leaky but can switch very fast. Update: To clarify, TSMC offers a 40nm LPG process that allows for an island of G transistors in a sea of LP transistors. This is what NVIDIA appears to be using in Kal-El, and what NV used in Tegra 2 prior.

The five A9s can't all be active at once, you either get 1 - 4 of the GP cores or the lone LP core. The GP cores and the LP core are on separate power planes.

NVIDIA tells us that the sole point of the LP Cortex A9 is to provide lower power operation when your device is in active standby (e.g. screen is off but the device is actively downloading new emails, tweets, FB updates, etc... as they come in). The LP core runs at a lower voltage than the GP cores and can only clock at up to 500MHz. As long as the performance state requested by the OS/apps isn't higher than a predetermined threshold, the LP core will service those needs. Even with your display on it's possible for the LP core to be active, so long as the performance state requested by the OS/apps isn't too high.

Courtesy NVIDIA

Once it crosses that threshold however, the LP core is power gated and state is moved over to the array of GP cores. As I mentioned earlier, both CPU islands can't be active at the same time - you only get one or the other. All five cores share the same 1MB L2 cache so memory coherency shouldn't be difficult to work out.

Android isn't aware of the fifth core, it only sees up to 4 at any given time. NVIDIA accomplishes this by hotplugging the cores into the scheduler. The core OS doesn't have to be modified or aware of NVIDIA's 4+1 arrangement (which it calls vSMP). NVIDIA's CPU governor code defines the specific conditions that trigger activating cores. For example, under a certain level of CPU demand the scheduler will be told there's only a single core available (the companion core). As the workload increases, the governor will sleep the companion core and enable the first GP core. If the workload continues to increase, subsequent cores will be made available to the scheduler. Similarly if the workload decreases, the cores will be removed from the scheduling pool one by one.

Courtesy NVIDIA

NVIDIA can switch between the companion and main cores in under 2ms. There's also logic to prevent wasting time flip flopping between the LP and GP cores for workloads that reside on the trigger threshold.

NVIDIA expects pretty much all active work to be done on the quad-core GP array, it's really only when your phone is idle and dealing with background tasks that the LP core will be in use. As a result of this process dichotomy NVIDIA is claiming significant power improvements over Tegra 2, despite an increase in transistor count:

Courtesy NVIDIA

NVIDIA isn't talking about GPU performance today but it did reveal a few numbers in a new white paper:

Courtesy NVIDIA

We don't have access to the benchmarks here but everything was run on Android 3.2 at 1366 x 768 with identical game settings. The performance gains are what NVIDIA has been promising, in the 2 - 3x range. Obviously we didn't run any of these tests ourselves so approach with caution.

Final Words

What sold NVIDIA's Tegra 2 wasn't necessarily its architecture, but timing and the fact that it was Google's launch platform for Honeycomb. If the rumors are correct, NVIDIA isn't the launch partner for Ice Cream Sandwich, which means Kal-El has to stand on its own as a convincing platform.

Courtesy NVIDIA

The vSMP/companion core architecture is a unique solution to the problem of increasing SoC performance while improving battery life. This is a step towards heterogenous multiprocessing, despite the homogenous implementation in Kal-El. It remains to be seen how tangible is the companion core's impact on real world battery life.