IATH News

Reseller Advocate Magazine features Professor Bernard Frischer

SFF Rises Again: Small Form Factor Just May Save the Desktop
By William Van Winkle
Cover Story (See original article)

Size Matters

Anybody who doubts that mainstream PC sizes are a function of underlying form factors should harken back to the AT and Baby AT designs of the late-'80s through mid-'90s. These 386-era behemoths were immensely popular but saddled designers with certain problems, such as a large chunk of the motherboard resting under the drive cage. The smaller you tried to build the system, the less convenient the machine became to work with. Killing boards by not inserting the power cables correctly was another minor problem.

In 1995, Intel produced the ATX form factor specification, still in widespread use today for performance-oriented systems. However, at 12" x 9.6", ATX isn't a contender for the SFF space. The starting point here is microATX, which shrinks the PCB size to 9.6" x 9.6". Most microATX boards use three expansion slots, although the spec allows for a maximum of four, but otherwise sacrifice very little of ATX's capabilities. Given that small form factor buyers tend not to care as much about expansion but still value out-of-the-box functionality and features, microATX has proven to be an immensely successful form factor in the mini-tower and media center arena.

When one talks about media center PCs, the assumption is that one means a mini-tower in an office setting and a CE-friendly desktop design in the living room. Strictly speaking, these aren't really small form factors, although the sleek styling from case manufacturers like Antec and SilverStone sure make you think twice. Fortunately, microATX does allow for some size shavings beyond these dimensions.

"Given that it's the dominant form factor in the industry, how small can you build a system on microATX?" posits Intel's Peter Brandenberger, marketing manager, channel products group. "If you're only willing to use desktop componentsdesktop hard drive, desktop power supply, a half-height add-in card, and so onthe answer is about 10 liters, and we have systems in beta to support that. Then you have to ask how much you gain by going with a slim optical drive; that'll take you down to about the 7L space. If you're willing to go to an external power supply and maybe a slimmed down heatsink-fan, like around 10mm, you can get about 5.8 liters. But getting that small will be hard for the channel because of external power supplies, custom solutions, etc. From an OEM perspective, it will be an option, but the channel will do best in that 7L to 10L range. So as an industry, what trade-offs are we willing to make? A lot of people focus small form factor just on the motherboard, and I keep trying to push people on whether they're willing to accept a slim optical drive or a riser cardthose kinds of things. Because if the answer is no, which a lot of times it is, then I don't care what board you use. You're in a fairly big system. You're up in the 10L space, even on a mini-ITX board."

For a few points of reference, a hulking beast like Zalman's TNN500A fanless tower is about 100 liters. A conventional mini-tower hits around 40L, and a set-top style ATX system from the likes of U.K.-based Hush Technologies measures about 17 liters. Shuttle XPC cubes hover around 10L, and slim mini-ITX systems often slip under 8 liters.

Intel saw early on that small form factor would be a popular trend and that microATX likely wouldn't hold the industry forever. Thus 1999 saw the arrival of FlexATX as an adjunct to the microATX spec. Outfitted with the same mounting hole and I/O plate arrangement as microATX but measuring only 9.0" x 7.5" (229 x 191 mm), FlexATX should have been all the rage among SFF manufacturers. But it wasn't. FlexATX volumes never grew large enough to give the form factor a fighting chance against microATX, and this lost cousin on the ATX family tree has been wilting into obscurity ever since. Truth be told, there are a couple dozen open form factors for vendors to develop on, but the dictates of economics and market momentum can have more to do with which thrives than technical superiority.

[BTX, Begone!]

No better example exists of this disparity between market conditions and top engineering than BTX, which, if it isn't officially decapitated, at least went into indefinite exile when Intel decided to stop developing on the platform last September. The various BTX designs arrived in 2004, and with them came a complete revamp of mountings and component placement, all with an eye toward smarter airflow, cooling, acoustics, and smaller size. BTX was seen as the saving grace for the thermal problems set to overtake desktops at a time when CPUs were edging past 130W and quad-GPU setups would swallow at least another 300 watts. Despite the added costs tied to transitioning the whole industry to a new form factor, everyone from the tier-one OEMs on down was getting onboard...and then Core microarchitecture arrived.

With that, the biggest thermal concerns vanished like a puff of ozone, and BTX, already much maligned by critics as being a work-around for poor processor designing, was finished. There were lingering thoughts that picoBTX, with only four mounting holes and one expansion slot, would still survive in the ultrasmall and tiny desktop spaces, but that plan also hit the skids.

"Large multinationals and OEMs are still on BTX and planning on staying there," notes Intel's Peter Brandenberger. "Therefore, we'll continue to support BTX. From a channel perspective, the benefits of BTX weren't sufficient to overcome the hurdles of adoption. As a result of that, our channel roadmap is no longer offering BTX solutions for next year. We expect adoption to be very slow over the next several years unless something significant happens. " Or, as Antec senior vice president Scott Richards puts it, "Yeah, BTX was a big disappointmentand if anybody out there still needs BTX cases, I've got 1,100 pieces!"

The logical question is: Well, we've gone from the Pentium M at 40W back up to the quad-core Core 2 Extreme's 130W, so aren't we back where we left off with the Pentium Extreme Edition? Well...yes and no. Manufacturers repeatedly tell us in vague terms that the industry has found "better ways" of coping with thermal problems in the last couple of years, making BTX no longer necessary. Even Brandenberger admits that ATX should remain sufficient for the low-end and mainstream segments for "the foreseeable future."

Below the microATX form factor, Intel is largely quiet these days. When pressed, representatives will mutter something about ITX being a solid solution, and we recall seeing at least a couple of ITX-based reference designs when the MoDT concept systems were floating about. However, given that VIA was the originator of the ITX specifications, not Intel, embracing the form factor must be a bitter pill for the blue team to swallow.

[ Deciding on DTX ]

In January, AMD released its alternative to ATX and BTX: DTX, measuring 200 x 244 mm, and Mini-DTX, measuring 200 x 170 mm. DTX is completely backward compatible with ATX cases. According to AMD, market research showed that most microATX users were not using more than two of the four slots on their boards, and often they used only one or none. Obviously, a smaller design could help cut costs.

"We've heard from partners and the industry at large that there is a growing demand for small form factor," says AMD's Daryl Sartin, director of platform initiatives, "but in the absence of a standard, proprietary solutions are all that's available. Those are servicing the industry to the extent they can, but the volume of demand behind that is not being serviced. Our solution to that problem is DTX. It's an open industry standard. We drafted the spec, and in many ways it's a follow-on to ATX. We went as far as we could in terms of compatibilitystand-offs are the same as ATX or microATX, the I/O block is positioned the same, add-in cards are the same. The SFF market, if it's going to be realized in high volumes and equal costs to current desktops, needs something that's easily available and a low cost platform. DTX is royalty-free."

Given the size of conventional PCB blanks used in motherboard manufacturing, shrinking the motherboard width to 200mm yields four motherboards per planar instead of two. Likewise with Mini-DTX, reducing the depth of the board from 244mm to 170mm yields exactly six motherboards per blank. On this basis, reducing down to mini-ITX's 170 x 170 mm delivered little to no added benefit. In theory, assuming equivalent volumes and scales of production, Mini-DTX should be cheaper to manufacture than microATX. "My take on DTX is that from an idea perspective, it sounds good," says IDC's Richard Shim. "It sounds like that's what the industry wants. They want something compatible with ATX that isn't going to cost a premium and that'll get them to smaller sizes. In some cases, they're saying that DTX will even be less than ATX. Fewer components, smaller board, and everything. I'm still a little reserved because I want to know exactly what the specification is going to be. My main concern is that the volume sizes they're targeting are pretty aggressive, and they're under the small form factor, according to my definition, volume size. They're actually closer to the ultrasmall category, and the unit sales potential there is even smaller."

AMD is careful to point out that it doesn't see DTX as a replacement for ATX, although it's hard not to see it as such from a motherboard perspective. (ATX power supplies will continue to service DTX systems.) According to Daryl Sartin, interest is coming in across all segments, from enterprises wanting smaller footprint desktops to price-sensitive emerging markets. With a target of mid-2007 for functional prototypes, AMD plans on enabling developers to ship DTX designs in 2008. Ultimately, AMD envisions DTX servicing the industry for 10 to 15 years. On some points, even Intel admits that DTX makes good sense...albeit with predictable caveats. "Having an ATX derivative is probably the right next step," says Intel's Brandenberger. "That makes sense if the market is ready for it. With DTX vs. ATX, the only difference really is going from four slots to two slots. That makes sense to me in a lot of applications, especially for small form factor. The two slots are really only required in a consumer application for graphics and tuner. However, the size you save in the board is about five percent. So if you had a 10L system, you go to 9.5. It's not that compelling, and when you put the two systems next to each other, you're going to be looking at something else—the bezel aesthetics or whatever. I like the idea of two slots, but I don't think it's compelling enough to create or sustain a new standard in the industry. From a panel savings perspective, I'm not sold on that value proposition. Some board people think it's compelling; some want to use that board space for other things. We looked at panel utilization with BTX, and we came to understand that there is no one panel utilization for the whole market. Mini-ITX has one slot instead of two, but in both of those you're looking at higher layer count, more board complexity, and potentially double-sided assembly. If you do that, the cost of the board goes up significantly."

[ Inching Toward ITX ]

In 2001, VIA demonstrated the ITX form factor as a diminutive platform for its recently acquired Cyrix C3 processor line. ITX was a good idea but a little too close to FlexATX to sway manufacturers. Thus followed the mini-ITX (170 x 170 mm) form factor in 2001, which gradually became a favorite in two groups: modders, who would use the toy-sized motherboards to convert things like toasters and Millennium Falcon models into full-blown x86 computers, and resellers (Logic Supply and Mini-Box.com are two excellent examples) targeting niche applications, such as in-car PCs and embedded systems.

VIA has not been able to keep pace with Intel and AMD when it comes to manufacturing processes and architectures, and VIA's technology is now roughly two to three years old in comparison. But such comparisons are a bit unjust since VIA positions itself in a different space. Whereas most Intel and AMD chips now run in the 60W to 100W range, VIA stays around 20W with speeds ranging up to 2 GHz. You're not going to see a modern C7 chip run 3D games under Windows Vista's Aero UI, but neither will you see top-end Core 2 Duo and Athlon 64 chips in a 5L case. Cutting-edge processors quickly overheat in such tight spaces because there isn't enough room to fit an effective cooling solution. That's why VIA excels in the ultrasmall and tiny segments, even though the system volumes there remain fairly low. Knowing this is where its advantage lies, VIA plans to press its love of the Lilliputian even further. "For many applications, we see even mini-ITX as being too big, which is why we're coming out with smaller and smaller solutions for systems," says Alp Sezen, director of sales, embedded division at VIA. "There's nano-ITX now, and soon we'll have pico-ITX. Mini-ITX is 17x17. Nano is 12x12 centimeters. Pico will be 10x7 centimeters. The reason we picked this is it's almost an identical size to a 2.5" hard drive. You can place it right on top of the drive and have a C7 CPU. You can go to 2 GHz, but that would require a fan. You could do up to 1.2 GHz in a fanless solution, though, with four USB ports, SATA, PATA, 10/100 or Gigabit, DDR laptop memory, LVDS output so you could drive two flat panels, and a VGA connection with a DVI version as well. The cost is fairly low to be able to do a solution with this almost off-the-shelf. It could be used in signage or embedded in the walls for automation. Our expertise in Linux also plays here. We support all of the kernels."

Also keep in mind that VIA does its own chipsets, audio codec chips, IDE controllers, and so on. One motherboard can be practically wall-to-wall VIA. This may make driver support more attractive for resellers. A unified parts source also lets VIA offer a 5-year longevity for its embedded line motherboards, much like AMD's and Intel's stable image programs. For VARs and system builders looking to leverage VIA's advantages in industrial applications, a 5-year parts support and replacement guarantee could be vital vital to securing corporate interest.

Players and Places

A lot of manufacturers still count themselves as active in the SFF world, even if that activity isn't quite as noticeable in recent years, and it's worth taking a deeper dive to examine some of the most prominent options. Our vision on who the key names and products are may be skewed by geography, as developed Asian and European countries tend to be more in tune with space savings than we expansive Americans. By and large, though, most channel manufacturers seem still to be targeting the consumer space with their SFF designs, and HTPC applications for the living room remain in everyone's focus. The question is whether these manufacturers are actively promoting such products and helping resellers to market them. For example, Gigabyte was one of the most recognizable Viiv builders last year, but we haven't noticed much channel activity around the company's Entertainment PC lineup. This is unfortunate, because models like the 945G/Pentium D-based H663 are fine pieces of set-top work, and expected SKUs like the G965/Core 2 Duo-based H971 should improve the end-user experience even more. However, Gigabyte is doubtless asking itself how much economic sense it makes to push a Viiv set-top given the prevailing sentiment on Intel's digital home platform.

MSI is another of the companies that has a very solid SFF lineup, particularly in its MEGA line, but now finds itself wondering where to carve out success in the States. The company is going to make a stab for the living room with its new Media Live set-top, based on NVIDIA's GeForce 6150 northbridge for AMD's Athlon 64 AM2 socket. This is actually one of the most exciting designs we've seen in the HTPC space for quite a while. The 11L Media Live is a highly integrated microATX box outfitted with a slot-loading optical drive, VFD display, 7-in-1 flash card reader, Gigabit Ethernet, 7.1 HD Audio, one riser-based PCI slot, and all of the back panel audio outputs you could ask for. You'll also find a SCART connector (widely used in Europe), S-Video, and HDMI, but there's no VGA or DVI output to be seen. It's the living room home theater or bust with this unit, which hopefully won't limit its prospects. Interestingly, MSI doesn't have a set-top-based Viiv box on the burner, although MSI headquarters are in discussions on this point. For now, the company is quite happy with AMD's platform.

"AMD also promotes the media center PC," says Philips Tsai, channel marketing manager for MSI Computer. "They actually outsell the Intel solution. According to my information, LIVE! outsells Intel Viiv by over 20 percent. Because the average cost matters. For Viiv certification, you have to pay a lot of money, and the total ownership cost is much higher than AMD's solution. The average on a complete Viiv solution is over $2,000 while AMD is only about $1,000."

Obviously, there's a message to take away from thisbut what? Few who have spent the time to grasp all of Viiv's features and functionality would assert that today's AMD LIVE! platform is 20% better than its Intel rival. (How LIVE! Ready set-tops and possible coordination of Athlon and Xilleon chips will improve LIVE!'s prospects remains to be seen.) Intel has higher performance-per-watt, Viiv-specific content and application optimizations, a far more robust networking and content organization infrastructure, on and onyet LIVE! outsells it by 20 percent.

So do we infer that you'll sell more SFF media center machines on AMD's platform? Or should we deduce that you'll sell 20% fewer SFF Viiv machines but you'll have double the average price per machine? In reality, the issue probably comes down to configuration and positioning. Taking Newegg.com as a reference and supposing we were building a small microATX system, one can grab a 2.13 GHz Core 2 Duo E6400 and an Intel D945GBOLKR Viiv motherboard for $324. An ASUS M2NPV-VM (based on the NVIDIA GeForce 6150) with a 65W Athlon 64 X2 4600+ (2.4 GHz) runs for $283. All other parts stay identical, meaning that the multimedia platforms need only differ by less than $50 before platform certification fees.

Perhaps what MSI is seeing is that AMD is being configured as a budget media center platform while system builders are gravitating to planting Intel in higher-end configurations. In truth, the two platforms are roughly equivalent today. You shouldn't pigeonhole Viiv and LIVE! into separate system styles. Both will make equally excellent platforms in a closet server or a living room SFF, and both platforms are set to evolve in the coming months.