WHY LESS IS MORE The high-k dielectric metal gate transistors used in Intel’s Penryn mark a radical advance in semiconductor design. With this, chips have far less power leakage, run cooler, and can thus scale higher.

Of these three components, the CPU continues to get the most attention, and deservedly so. The latest mobile Core 2 Duo SKUs are based on Intel’s 45nm (Penryn) architecture. We and others have covered Penryn for months, so suffice it to say here that the microarchitecture’s principal benefits boil down to four things: 1) Smaller transistor and other feature sizes mean more ability to cram in millions more circuits, which is why you see L2 cache sizes leaping on 45nm chips. 2) Intel’s hafnium-based high-k dielectric metal gate transistors significantly reduce power leakage. Most of us blow this off as an “uh huh” item, but Intel touts it as the biggest breakthrough in transistor technology since the late ‘60s. The smaller the circuits, the greater the leakage risk, so the new technology is sorely needed. Specifically, Penryn delivers over 5X less source-drain power leakage and over 10X less gate oxide leakage. 3) New SSE4.1 support adds 47 instructions on top of prior-gen SSE3 for improved performance in multimedia and other tasks. 4) “Deep Power Down Technology” drops power consumption so low during idle core periods that leakage becomes negligible, thus helping battery life in notebooks.

Naturally, there are other benefits. A little-mentioned feature called HD Boost in Penryn chips can allegedly improve HD multimedia applications by up to 50%, although we’ve never seen this independently benchmarked. Single-threaded apps can benefit from Penryn’s Intel Dynamic Acceleration Technology, which essentially overclocks one core when the adjacent core is inactive. Virtualization Technology (VT) is improved. Frequencies are higher. Front side bus speeds are higher. And so on. Best of all, thanks to Intel’s habitual and periodic price drops, there’s now little to no end-user benefit to buying 65nm processors. Penryn is now the only sensible Intel architecture option.

We got a first dose of mobile Penryn at the top of 2008 with the Santa Rosa refresh. With Montevina, a second 45nm chip generation arrives, bringing with it a front side bus increase from 800 MHz to 1066 MHz. The “power-optimized performance” SKUs fall into the P-series, such as the P8600 and P9500 at 2.4 GHz and 2.53 GHz, respectively. (The 8000-series models feature 3MB of L2 cache while the 9000-series parts double to 6MB.) According to Intel, P-series chips feature a 25W TDP, which is extraordinarily low considering the performance from these dual-core products. Moving into T-series CPUs bumps the power profile back up to 35W, but it also allows for higher frequencies, as with the 2.8 GHz T9600.

As of the Montevina launch, there is only one mobile Core 2 Extreme processor with a 1066 MHz bus, namely the 3.06 GHz X9100. This part sports a 45W TDP. However, we don’t expect many X9100s to fly off of shelves because the Web is rife with news of the pending Q3 launch of the X9300, still of 1066 MHz FSB, but with a quad-core design and 12MB of L2 cache. This will mark Intel’s first quad-core mobile chip, and, like other Extreme Edition quad-core models, Intel states that it will be overclockable.

As gaming becomes ever more popular on high-end notebooks (there were already two “gaming” models listed on Intel’s April Montevina barebones catalog) and Intel continues to try ingratiating itself with the gamer community, we can’t help but wonder if enthusiast notebooks will soon have the same level of BIOS options and GUI-based tools, such as Intel Desktop Control Center, found on desktop motherboards. If so, this could be a new class of value-add that resellers can pitch to consumers. In any case, the Montevina platform supports the X9300 and subsequent quad-core designs, but resellers will have to keep an eye on how well the ODM supports the quad-core guidelines in Intel’s Montevina Platform Design Guide. All quad-core Montevina designs will support dual-core, but the opposite is not true.

For those who want power savings and battery life over raw performance, keep an eye out for subsequent mobile Penryn releases in the third and fourth quarters. The SL-series (sometimes noted as the traditional “LV” part) will have a 17W TDP. ULV Core 2 Duo models will hum along at only 10W TDP. There’s even a 1.2 GHz Core 2 Duo U3300 rumored to consume only 5.5 watts, but the lowest ultra-low-voltage (ULV) spec Intel shows in its pre-launch literature is 10 watts. Expect these ULV SKUs to show up in a wide range of small form factor mobile designs.

CHIPSET

As on the desktop, Montevina sports a number of 4-series chipset options. The flagship among these at launch time is the GM45, featuring Intel’s GMA X4500 integrated graphics processor (IGP). This is Intel’s fifth-gen IGP core, now with 10 unified pixel/vertex shaders, DirectX 10 and OpenGL 2.0 support, and a 533 MHz clock. The GM47 variant boosts this clock to 640 MHz, although it seems sensible that buyers wanting more graphics performance would opt for the IGP-free PM45 paired with a discrete GPU card. According to Intel, users should expect a 30% to 40% 3DMark06 improvement with the GM45 relative to the GME965.

While the X4500 is suitable for light to moderate 3D gaming, the GM45's visual story is really about media output. This starts with 1600 x 1200 dual independent display capability and hot plug support for all displays. In addition to all the usual analog and DVI ports, the GM45 also handles HDMI and the DisplayPort format now gaining steam throughout the computing and consumer electronics worlds. While you shouldn’t expect DisplayPort to rule the market by this Christmas, expect the format to be mainstream by the end of 2009.

Whatever advantage ATI and NVIDIA might have previously enjoyed as notebook HD playback platforms, Intel now levels the field. The GM45 bakes in HDCP keys for security decryption and adds hardware-based acceleration for H.264, VC-1, and MPEG-2, which covers the major high-def bases. In particular, Intel now adds inverse transform and variable length decode capabilities for VC-1 and AVC, both of which were absent under the GME965. Now users can add a Blu-ray drive to their notebooks and not worry about swamping the CPU. Better yet, with dual display, users can have that movie playing on one screen while keeping other apps active on another. The chipset uses DVMT 4.0 to share up to 384MB of system memory with the video subsystem.

Mobile Makeover Montevina’s GM45 chipset bumps the frontside bus to 1066 MHz, the memory bus scales to 800 or 1067 MHz (DDR2/3), and the X4500 IGP delivers loads of new graphics functionality.

The GM45 and PM45 both feature support for either DDR2 (up to 800 MHz) or DDR3 (up to 1067 MHz). In both cases, a maximum of 8GB of system memory is supported. Of the 23 Montevina models noted in Intel’s April design list, only one opts for DDR3: Clevo’s 17" M570TU, which predictably pairs with the PM45 and is aimed at performance gaming. Given that DDR3 is only just now grabbing a foothold on the desktop, we don’t expect widespread DDR3 adoption on the mobile side until next year.

We should point out some of the fine details in Intel’s memory update. This isn’t just about faster bus speeds; Intel also weaves in a power angle. For example, when the CPU is in states C2 to C6, there’s no longer a need to access system memory, so the memory enters a self-refresh mode to save power. Moreover, individual ranks of DRAM can be powered down when idle. And don’t forget that DDR3 consumes about 25% less power than DDR2 at comparable speeds.

Often, the southbridge is sort of the afterthought in chipset discussions, but Montevina’s options are worth exploring. There are three mobile ICH9 variants: the ICH9M, ICH9 Enhanced, and the ICH9M Small FF (due by Q4). While it’s not worth dwelling on here in August, the SFF model differs from the ICH9 Enhanced mostly in its size—31x31 mm versus 16x16 mm, with nearly all other specs being identical.

The ICH9M is the budget section of the family. All mobile ICH9 models dispense with PATA support, but the vanilla ICH9M also omits RAID, VT-d, Matrix Storage, and Intel Active Management Technology (AMT). The ICH9M Enhanced has all of these, specifically supporting AMT 4.0 and RAID levels 0, 1, 5, and 10. This is interesting because the ICH8M predecessor only offered RAID levels 0 and 1 (again, only on the Enhanced version). Why provide for triple-drive RAID 5 and quad-drive spanning with the new chip? Perhaps because the ICH9M and ICH9M Enhanced increase the SATA/eSATA port support from three to four, opening the door to more robust storage arrays comprising both internal and external drives. This should definitely interest commercial purchasers and desktop replacement-minded power users. We’re also intrigued by the prospect of AMT support, and thus Centrino 2 with vPro badging, on SFF mobile devices, a development that could give a firm boost to business netbook sales next year.

NETWORKING

Few people get excited about updates to Ethernet chips, but the Intel 82567 (Boazman) Gigabit controller found in Montevina has a few noteworthy perks. First off, the package anticipates the move to SFF designs by shrinking 35% over the prior LAN chip, down to a mere 8x8 mm. The PHY boasts only a 602 mW TDP—a 45% power drop from its predecessor—but it has other power perks, such as the ability to power down when there’s no link detected. Not least of all, the 82567 supports AMT 4.0, making this the de facto Ethernet chip for vPro notebooks.

Interestingly, wireless networking doesn’t show up on a lot of Intel’s Montevina engineering docs, but the updates to this component are probably more anticipated than any other aspect of Centrino 2. Santa Rosa brought us the WiFi Link 4965AGN mini-PCIe card, the first instance of Draft N wireless support for Centrino, in addition to backward compatibility with 802.11 a, b, and g. Ever since the Wi-Fi Alliance started certifying Draft N 2.0 solutions, the industry has more or less accepted Draft N as being equivalent to 802.11n in advance of that standard’s final IEEE approval.

QUAD FOR THE REEL Expect Montevina to welcome quad-core mobile processors as the year continues and maybe even (gasp!) become overclockingfriendly in order to attract high-end consumers and provide more value.

With Montevina, Draft N 2.0 support is updated with four new adapters. Apart from a form factor shrink, the new WiFi Link 5100AGN has little new to offer over the 4965AGN. Both cards use a 2x1 (two receiving, one sending) antenna arrangement and spec out with the conventional 300 Mb/s theoretical throughput rate. The 5300AGN gets interesting. This card matches the older 4965AGN in size but now moves to a 3x3 antenna array, propelling the throughput spec up 50% to 450 Mb/s. Naturally, this will require a compatible access point or router, but Linksys, D-Link, and others are already lined up and ready to supply these.

Later in the year, expect matching follow-up adapters, the 5150 and 5350, to add WiMAX support on top of Wireless-N in the same card. This isn’t the time or place to detail WiMAX’s long and arduous road to launch, but all signs seem to indicate that we’ll be there by the end of 2008. Suffice it to say for now that WiMAX aims to be a last-mile wireless broadband alternative to cable, DSL, and fiber optics. In practice, WiMAX will likely deliver 802.11b-like performance, but how it becomes a viable channel technology will largely depend on how Sprint/Clearwire, the primary WiMAX carrier in the United States, rolls out service. No one expects WiMAX to be a compelling technology in 2008, but simply having support for the technology in Montevina may be a key buying consideration, especially for business customers.

WIRELESS-N BREAKTHROUGH Standard 802.11n specifi es the use of multiple antennas in order to reach 300 Mb/s theoretical speeds. Intel’s 5300AGN wireless adapter uses three antennas to hit 450 Mb/s when used with a compatible access point.

“I think there’s enough in Montevina to persuade Santa Rosa owners to upgrade,” says Steve Maser, vice president of product development and marketing at Seneca Data. “There are some great new graphics pieces in there. You’ve got WiMAX becoming a reality and new capabilities in Wi-Fi. Montevina is greener on power savings. Even the Turbo Memory support is better. Is it going to happen the day it launches? Probably not. But over the next year or so, you’re not going to be able to stretch out that Santa Rosa as well.”

Try as we might, there haven’t been a lot of opportunities in the last couple of years to pump up AMD’s mobile offerings in print. Turion has been something of a non-entity in the whitebook space, even if it did pull together some OEM wins, particularly in the budget segment. The one thing AMD had going for it last year was the ATI Mobility Radeon brand. Given that ATI has historically enjoyed about 80% of the discrete notebook GPU market, this wasn’t a bad card for AMD to have up its sleeve, even if that card didn’t see much play on the company’s bottom line.

TIME FOR A MEMORY MOVE Just when we were getting settled into DDR2 SO-DIMMs, Montevina will see us start the migration to higher-performance DDR3 modules, such as this mainstream-oriented SKU from Kingston.

AMD has been doing a fair Cheshire Cat impersonation of late, smiling coyly and telling everyone just to wait and see what’s around the corner. Well, here we are on the cusp of Q3, and you’ll be seeing AMD’s mobile surprise on shelves by the time you read this. The surprise is a notebook platform code-named Puma, and it may just put AMD back on the mobility map.

AMD isn’t striving for end-to-end mobility coverage like Intel. Team Green won’t hesitate to tell you that it only wants the middle 80% of the notebook market. Ultra-portables? Enterprise notebooks? Forget it. Those are single-digit segments. According to AMD’s research, the sweetest spots are thin and light consumer and SMB notebooks, the 13" to 15" and 4.0- to 6.5 pound units that together constitute 72% of the notebook market. AMD numbers also reveal that roughly 60% of all notebook owners currently or plan to enjoy multimedia on their mobile PCs. Another 43% use the machines for playing video games. Keep these numbers in mind as you read on and consider how such demographics might play into your own sales efforts going forward.

Depending on which side of the fence you want to watch, Puma is either a good news or bad news development. Your first sign of this ambiguity might be AMD’s recent propensity to downplay mobile CPU performance and talk up total platform performance, which typically means graphics/video and battery life. Sure enough, we now know that Puma includes the Turion X2 Ultra processor, formerly code-named Griffin.

At first glance, Griffin looks like a higher-clocked respin of existing Turion chips. As with Intel’s Centrino 2, there’s no mobile quad-core option in sight at launch time. Cynics might argue that you don’t really need quad-core in a notebook yet, and they’d probably be right. Serious multitasking usually means multiple monitors, and notebooks only do multi-mon under duress; it’s not a paradigm suited to the form factor. Games are still in the process of becoming many-threaded, and we’d argue that mobile graphics are typically more of a performance bottleneck to gaming than mobile processors. Would quad-core help with Office, Quicken, Web surfing, and the other things that most people do with notebooks? Not really.

So what we have with the Turion X2 Ultra is a K9 architecture (same as Athlon 64 X2) with several cool elements from the new K10 (think quad-core Opteron and Phenom) baked in. On the surface, most of the bullet points seem pretty routine. At launch, we’ll have three Turion X2 Ultra SKUs: ZM-80 (2.10GHz, 2 x 512KB L2 cache), ZM-82 (2.20GHz, 2 x 512KB L2 cache) and ZM-86 (2.40GHz, 2 x 1MB L2 cache). Each chip is based on 65 nm fabrication and sports a thermal design power (TDP) of 35W. Although, as we’ve seen from AMD and Intel before, actual sustained usage is typically far less than the stated TDP.

Power savings makes up half of Puma’s appeal, and how this savings is achieved proves that AMD isn’t taking the Core 2 threat idly. For starters, you probably know that AMD uses point-to-point HyperTransport (HT) links instead of a front side bus, and the first-gen HT technology used in AMD’s prior mobile generation now gives way to HT 3.0. This update propels maximum link bandwidth from 12.8 GB/s to 41.6 GB/s. More importantly, HT 3.0 adds a self-configuring power management mode.

“Our previous platform had HT1 support, which was 800 MHz,” says David Rooney, AMD’s Puma product manager. “We more than doubled the frequency capability of Puma’s HyperTransport bus to 1.8 GHz. But in terms of power control, some of the revolution here is that we are able to change the link width of that communication bus. Today, it runs at x16 upstream and downstream. It’s bi-directional, as opposed to the front side bus, which is uni-directional upstream or downstream. So with Puma, we can run the HT bus at a higher bandwidth when plugged in, shrink the bandwidth down to x8 on battery, and even do a full disconnect. The trick to saving battery life in notebooks is all about shutting off different parts of logic and subsystems when the PC is in idle and then turning them back on when you need performance. We don’t want to pay for performance when we don’t need it. And while there are some more coarse levers for toggling performance on a notebook, such as the OS choosing P-states, this represents a step further in terms of controlling subsystems like the HT link between the chipset and CPU.”

Griffin Jumps Into Flight The 65nm die at the heart of AMD’s Turion X2 Ultra may still be dual-core, but it delivers plenty of performance for low- to midlevel buyers while emphasizing low power consumption.

Another Griffin advantage is the use of different power planes within the processor enabling independent throttling of frequency and power for each core, something AMD calls Independent Dynamic Core Technology. Based on dynamic user demand, the system can throttle each core higher or lower or even send it into a “C1e” deep sleep. Consider an application such as video playback, which may require very little interaction with the CPU if the GPU is handling all of the decoding. The system could put the core to sleep and only keep the power plane that has the memory controller and HyperTransport controller activated, resulting in a relatively significant power savings.

Again, AMD doesn’t spend a lot of time discussing the Turion X2 Ultra. The chip is obviously an integral part of the platform, but when you’re talking about innovations like the DRAM prefetching embedded within the DDR2-800 memory controller, the eye naturally drifts to where there’s better eye candy...namely the graphics.

Puma leverages the M780G northbridge, best known for its support of ATI Hybrid Graphics. In fact, it’s hard to separate the chipset and graphics with Puma because the two always seem to appear in the same breath. There are other aspects to the M780 family, such as support for hybrid hard drives, the addition of DisplayPort support, official support for the DASH 1.0 remote management spec (don’t expect to hear much about this until Puma refreshes), and the ability to use HyperFlash, AMD’s own spin on Intel Turbo Memory. These are all worthy items, but the bunch seems to pale beside AMD’s next-gen mobile graphics.

The M780G integrates an ATI Radeon HD 3200 graphics core, complete with 205 million transistors, 55nm fab process, DirectX 10.0 support, and enough oomph to outperform NVIDIA’s low-end discrete cards. According to AMD, the 3200 core delivers 3X the graphics performance and 5X the HD image quality (as measured in HQV HD) as Intel. Benchmark scores taken in Half-Life 2, played at 1024 resolution on battery power, show Puma pulling down 49 FPS as opposed to 15 FPS on a competing Intel platform. Mind you, every AMD-to-Intel comparison we’ve seen pits Puma against the GM965 chipset in Santa Rosa, which is no longer an appropriate comparison, so watch the fine print.

Still, the HD 3200 is plenty impressive in its own right. The IGP comes outfitted with AVIVO and the ATI Unified Video Decoder (UVD) engine, providing cleaner digital video playback alongside hardware-based decoding of the main HD video codecs. According to AMD, between UVD and Puma’s power savings, users can watch an entire Blu-ray movie on a single battery charge—a feat that normally would have been impossible a year ago.

Perhaps Puma’s single biggest attraction is its use of hybrid graphics, meaning the CrossFire pairing of a discrete GPU with the northbridge’s IGP. While AMD expects finished configs featuring hybrid graphics to have $100 to $200 price premiums over their non-hybrid counterparts, much of this delta derives from non-graphics component upgrades. The ATI Mobility Radeon HD 3650 card itself represents only about $30, which seems an extremely low cost for performance benefits that AMD says will reach up to 70% greater than running with the discrete card only. To work, the two graphics engines have to be roughly equivalent in capabilities. You couldn’t mismatch it and pair the HD 3200 in the IGP with a significantly higher-class discrete GPU. AMD is looking at modifying this restriction in the future, but today this is the hand that Puma deals.

PIECES OF PUMA AMD’s Puma platform differs markedly from Montevina. Here, AMD gives us the mobile CPU plus a northbridge and southbridge. There’s no sign of a mandated NIC, but youdo have discrete graphics.

“If you look at what we can do with integrated graphics, we still expect to have 3D superiority, even over Montevina,” says AMD’s Rooney. “But if you add a discrete card to an Intel solution, it’s no better than the discrete card by itself. With the AMD platform, you not only get the benefit of the discrete part, but you get more than that. Are you going to see 1.7X in every application? Of course not. But if you look at benefits and price from the AMD platform versus the competition, it’s compelling for many users today.”

Alongside its hybrid graphics, AMD implements a power-saving technology set called PowerXpress, a mobile-only feature (presently) that works in conjunction with Windows Vista. PowerXpress dynamically switches between the IGP and GPU adapters as conditions demand.

“Discrete notebooks suck up battery power really fast,” says Scott Shutter, notebook division brand manager for AMD. “When you’ve got an AMD-based system with PowerXpress, you can either press a button on the system to power down the discrete card or have it operate dynamically so it’ll power down when you unplug. A little dialog box will pop up and say you’re about to go into integrated graphics only, please close all your applications and restart. With the ATI Mobility Radeon HD 3450 card, it’s a 2.4X increase in performance over integrated when it’s plugged in, and when you unplug you can get up to 90 minutes of extra battery life.”

PowerXpress works alongside other AMD power-saving features, such as ATI’s traditional PowerPlay technology, which throttles the GPU into low, medium, or high power states based on demand from applications. Also note that Puma carries forward with a little-known ATI technology called Vari-Bright. Vari-Bright tracks display content and optimizes backlight intensity for different areas. There’s no point in keeping maximum brightness on an area that isn’t being watched. The feature makes sure that backlight transitions are smooth enough that the eye doesn’t note the zone differences, but the net effect is a bit of battery life savings.

As a business play, we also like that having two graphics cores in a system, each with dual independent display capability, means that a Puma notebook can offer quad-head output. Today, there’s no way to do this on an Intel-based notebook using strictly internal components.

“If you think about notebooks really taking hold in terms of replacing desktops in homes and SMBs,” says Rooney, “you really start to see that a notebook has no compromises anymore. You can load it up with desktop-level performance, both CPU and graphics, and yet still get a very acceptable battery life out of it. With things like hybrid graphics, you can put both GPUs together and get workstation-like capabilities. At the same time, you can get all kinds of power savings. It’s the best of both worlds without having to make sacrifices in battery life or graphics performance.”

As a final note on Puma, AMD has made a minor hullabaloo about its Draft N wireless implementation, noting that its open, non-proprietary design allows Puma to use whichever wireless adapter best suits the OEM’s and end-users’ needs. In contrast, if you buy Intel, you get Intel, period, and Intel allegedly isn’t a wireless networking specialist. In comparing the Atheros AR9280 Draft N adapter in some Puma designs with the Intel 4965AGN, AMD found the Atheros to offer 38% faster transfer times at moderate-range distances. That’s all fine, but if we’re going to talk about an optimal wireless experience, then we should be comparing against the 450 Mb/s Intel 5300AGN.