Last month, we spotlighted Intel’s LGA775 CPU innovations and the new 915/925X chipsets. While these few chip changes are significant, though, they only tell half of the revolution’s story. This month, we’re back to fill in the blanks.
The entire layout of desktop motherboards is about to get an extreme makeover as the Big Water tide rolls in. Second-generation DDR memory is quickly coming to market. The most significant change of all, PCI Express, is so important that we made it into a special pull-out insert for this issue. And of course all of these are just ideas until we actually see some implementations in real-life products, so we’re going to bring you some highlights to start stocking your shelves with today.

The Slow Birth of BTX

The ATX form factor has been with us for about a decade, and it has served quite well. (Some of us can remember feeling vexed at the switch from AT to PS/2 connectors and single-plug power connectors, although such moves now seem obvious and necessary in retrospect.) The form factor has seen us all the way from ISA slots to AGP 8X and highly integrated motherboards, not to mention all the years of Pentium updates.

But the fact is that ATX had several built-in drawbacks that didn’t become apparent until relatively recently. The first and most obvious is increasing difficulty with handling rising component temperatures. One need only study a chassis such as Cooler Master’s CM Stacker (see below for our review) to see the convoluted and ingenious measures vendors must go to in order to keep case temperatures under control in the face of modern processors and drives. The ways in which board components are arranged and how air is made to flow past them has a tremendous impact on cooling efficiency. Moreover, more cooling almost always means more noise.

Tied to this is power. You don’t need a degree in electrical engineering to recognize that power demands are swiftly rising in modern PCs. Gone are the days when high-capacity power supplies were sneered at as excessive indulgences for “mine’s bigger than yours” gamers and one could point to Tier 1 OEM designs to prove that 180W or 250W power supplies were all anyone really needed. Now everyday configurations have powerful CPUs, high-performance chipsets and GPUs, multiple hard and optical drives, on and on, all of which only grow more power hungry as specifications rise. ATX’s continuing inability to cope with these issues translates into either rising costs or poorer performance.

Balanced Technology Extended (BTX), formerly code named Big Water, seeks to remedy these problems. After many months in development, Intel released version 1.0 of the BTX specification on September 16, 2003. Now that almost a year has gone by, you might wonder that the market isn’t full of BTX designs. Well, word on the street is that manufacturers have remained gun shy about developing for the platform because the specification continues to be tweaked. In reality, a quick glance down the dates at formfactors.org shows that Errata A followed the 1.0 spec in February, and version 1.0a arrived that same month. If anyone is keeping track, ATX is now in version 2.2, and there are also offshoot motherboard specs for MicroATX and FlexATX.

We’ve spoken with some vendors who are skeptical of BTX, but none have been willing to go on the record with us in an Intel bashing session. For our part, we can see few things wrong in the BTX plan, although we’ll add that this opinion comes before building any BTX systems. For example, one of our pet peeves with ATX is how some mobo manufacturers place their RAM slots too close to the plane of the AGP card. If you insert and lock an AGP card before installing the memory, you can’t close the DIMM clips because they’re trapped under the video card. We’re fairly sure nobody spotted this flaw until after parts became available. Perhaps we’ll uncover similar pitfalls over the coming months with BTX.

The core idea in BTX seems unassailable, though. There is one case fan, and it sits at the front of the case drawing in cool air. Immediately behind this in the CPU and its attendant power circuitry, all of which is encompassed in a plastic tunnel known as the thermal module. This way, the hottest components in the system are the first to get the coolest air. Behind the CPU heatsink is the North Bridge, then the South Bridge, followed by the I/O ports on the backplane. Above the I/O ports is an grille with an optional exhaust fan. Thus the the fan(s) and thermal module help to create a sort of wind tunnel through the center of the system. The memory modules sit just to the left of this air stream and the x16 graphics card slot just to the right, allowing both of them to be secondary beneficiaries of the air current.

Card slots on the right? Yes. BTX slots rest where the CPU is today. How many slots there are depends on the BTX type. Three layouts have been specified: picoBTX (8.0” x 10.5”), microBTX (up to 10.4 x 10.5), and standard BTX (up to 12.8 x 10.5). As you can see, the formats simply get wider, not longer. Moreover the core layout from the memory slots along the left edge to the x16 connector on the right stays constant. Larger designs simply add more slots to the increasing PCB real estate. This design commonality should help to reduce design and production costs significantly.

The picoBTX design should prove popular in SFF and set-top systems while standard BTX is for larger desktops and towers. Many people expect microBTX to be the de facto selection for mainstream mini towers and desktops. Additionally, there are two BTX case heights. The more common Type II requires at least 3.98” in chassis height and can use standard height expansion cards while Type I cases are a slim 3.0” tall and may need half-height cards. Hopefully, half-height PCIe cards won’t be as scarce in the future as half-height PCI and AGP cards are today.

BTX proponents like to tout that BTX cases will be backward-compatible with ATX power supplies. Well, yes and no. It’s true that the 4-wire Molex connectors and their smaller counterparts for floppy drives will stick around. Beyond that, things get tricky. The minimum power spec for BTX is expected to be 350W, which automatically knocks away many if not most of today’s factory PSUs. Conventional ATX power supplies will not fit in microBTX and picoBTX chasses. There is a new 3.3V SATA power connector in development. The main power connector will migrate from 20 to 24 pins, and the 12V power connector is expected to go from four pins to six. In other words, be very careful if you’re telling customers that the pricey PSU they’re buying today will be ready for BTX tomorrow, because most of them won’t be.

At the recent Computex event in Taiwan, considered the bellwether of what to expect for the coming year in PC technology, there were precious few BTX designs. If Intel decides that the BTX design is final, then we should see products trickle in during the third quarter and hit their stride in the fourth. If tweaking continues, it’ll be sometime in 2005 before manufacturers dedicate themselves to the new standard. “It’s too early to say that BTX will flop,” says Scott Richards, worldwide vice president of sales and marketing for Antec, “but I do hear that Intel is pushing the big OEMs to adopt it. The soonest I see it happening at the SI/VAR level is the first quarter of next year, though. We have a couple products on our roadmap, but most people are in the wait and see stage. Intel wants more air to the CPUs, but now that they’ve abandoned the high megahertz roadmap, it may not be as critical—even to them.”

All of this has the overtones of being grim news, but it’s really not. Plain and simple, BTX is a better design. It’s quieter, more scalable, and more cost efficient. As with all major format changes, there will be a few bumps in the beginning to iron out, and manufacturers are now busy ironing. Intel projects that BTX adoption will remain shy of 10% in 2004 but jump to 20% in 2005 and 70% by 2007.

The bottom line is that BTX is going to happen soon enough, and when it does it will likely mean the end of the beige box for good. Critics sometimes argue that BTX is a high-level ploy to spark equipment sales. Yes, BTX will either inspire or force a large wave of upgrades throughout the industry, but in return customers will get fewer fans, less expensive heatsinks, and less frequent upgrading required when they place higher thermal loads on their systems. So long as you demonstrate to buyers that they’re buying into a long-term architecture with more tangible benefits than ATX, then the upgrade costs will be justified.

BTX Cat Fight

You may not be chummy with the 800-pound gorilla, but nobody wants to upset the beast, either. In the process of preparing this article, I heard rumblings here and there that vendors weren’t necessarily embracing BTX with open arms, and some outright think it’s a terrible idea. However, when I’d ask for specific reasons why, the response was usually something like, “Um...are you recording this?”

After a couple months of probing, I finally got one vendor to slip me some anti-BTX comments under an oath of anonymity. So we’ll call him John Doe from Company X. After reading over the remarks, I thought, “Hm, interesting! I wonder what Intel has to say about this.” So I found out. Craig Randleman is Intel’s BTX program manager, which makes him perhaps the world’s most qualified voice on the subject. We have our contestants. Let’s get ready to rumble!

JD: Here are some board size FACTS: MicroBTX is fixed at 10.4” wide by 10.5” deep (front to rear). BTX is fixed at 12.8” wide by 10.5” deep (front to rear). In comparison, microATX has a fixed width at 9.6” by worst case depth of 9.6”. (Many are built that are well under 8.6” deep.) ATX has a fixed width at 12.0” by worst case depth of 9.6”, and some are built as narrow as 7.0”, such as the ASUS P4B533-X. All other things remaining equal, larger motherboard size results in: a) higher motherboard cost, b) higher chassis cost, c) higher chassis packing materials cost (foam and carton), d) possibly reduced chassis pallet quantity, and e) higher chassis and system shipment costs.

CR: It is true that microBTX and BTX both increase board real estate relative to microATX and ATX, respectively (17 square inches for microBTX and 19 square inches for BTX). While this represents an increase in board costs, other elements of board cost are lower with microBTX and BTX boards due largely to improvement in the thermal environment. For instance, processor power delivery costs are lower because lower BTX power delivery temperature allows cheaper power delivery components and/or the elimination of power delivery phases. Chipset heatsink component costs are lower because better BTX airflow and air temperature allows cheaper heatsink technologies and/or elimination of a heatsink altogether. Total board cost estimates have been compiled and show BTX at parity with ATX or with considerable cost savings. The savings are proportional to system power—the higher the total system power, the more a BTX board will save you. The primary justification for increasing board real estate, however, is saturation. In the core area (processor, chipset, and memory), a 4-layer ATX board has placement and routing that are 98% saturated. As a new form factor standard, BTX increased available board real estate in this critical area in anticipation of future increases in socket and package sizes and the strong desire to stay with 4-layer desktop boards. In response to the concern regarding chassis cost, a BTX chassis may require more sheet metal and be slightly larger volume but this is very specific to the individual system design. For existing ATX implementations, it appears that most can be replicated at the same total system volume in BTX, at least according to the industry designs we’ve reviewed and case studies we’ve done ourselves. Importantly, BTX allows the design of smaller systems than ATX allows today—under five liters with the use of an external PSU and under seven liters with an internal PSU. And at the lowest possible microATX system volume (approximately 13 liters), the total microBTX system BOM [bill of materials] cost is considerably cheaper in the mainstream and performance categories. As is the case with boards, system savings increase as total system power increases.

JD: BTX envisions driving air flow under the motherboard. The motherboard is “raised” in the chassis to accommodate this. All other things remaining equal, the chassis will be wider, the packing foam will be larger, the carton will be larger, potentially there will be reduced chassis pallet quantity, and potentially there will be higher chassis and system shipping costs.

CR: The greatest benefit of the under-board airflow is the improvement in board surface mount component temperatures. Socket temperatures are approximately 25C better in BTX, substantially improving socket reliability. Part of the improvement in power delivery component temperatures mentioned above comes from the under-board airflow. A portion of the heat generated by power delivery components is transferred through the board into this under-board airflow stream. This heat flow path also exists for surface mounted chipset components. As noted above, this leads directly to power delivery component and chipset heatsink cost savings. Finally, and as noted above, you can design smaller systems with BTX than with ATX, even with the additional volume under the board.

JD: BTX specifies adding several new components, including a Support and Retention Module (SRM). As currently proposed in the spec, this is an 11.4 oz (almost 3/4 pound) plate that needs to “clip” into the chassis.

CR: Actually, the BTX Interface Specification does not define the SRM. SRM guidance is offered in the Platform Design Guide associated with each Intel processor socket and chipset. (Perhaps this is not an important distinction, but I find it’s often best to be precise.) I will concede that virtually all BTX chasses will ship with an SRM and that the weight of this component may add to their shipping cost.

JD: There is also the CPU cooler shroud (thermal module) to be provided with boxed processors or provided by the system builder if they use tray processors. Then there’s the CPU cooler shroud interface tunnel (thermal module interface). Since the CPU cooler shroud does not mount directly to the front of the chassis, a CPU cooler shroud interface tunnel is required to span the gap between the front of the chassis and the front of the CPU cooler shroud.

CR: Clarification: The Thermal Module Assembly (TMA) may mount directly against the chassis front panel. In fact, two available Intel reference design systems (a 12.9 liter system using a microBTX board and Type I TMA and a 6.9 liter system using a picoBTX board and a Type II TMA) illustrate this TMA-to-chassis attach method. However, in the event the front panel is not concurrent with the TMA inlet, the chassis must provide the duct noted by Mr. Doe, the cost for which is estimated to be $0.20-0.25. This additional duct assures that no recirculation of warm air from the TMA exhaust can re-enter the TMA inlet. I do not have an estimate of the TMA heatsink, power delivery, graphics card heatsink, or chipset heatink cost increases that would be required if recirculation were allowed, but I suspect that the sum of these would exceed the cost of the additional $0.25 duct.

JD: Not least of all is the CPU cooler stator to flatten out air flow coming from the CPU cooler fan.

CR: A stator is not required in the TMA. Many thermal solution providers have TMA solutions in development that do not use a stator. The stator does condition the flow in a way that assures it is most efficiently moved through the processor heatsink; thereby increasing the pressure capability of the fan. This allows the fan to operate at lower RPM at any given airflow and therefore offers an acoustic improvement.

JD: Taken all together, these added components result in higher cost.

CR: It is true that a BTX TMA will often cost more than an ATX processor heatsink. After all, the TMA heatsink is allowed to be bigger and there is a TMA duct. It is important, however, to remember two things: (1) the total system BOM cost will be equivalent or, in most cases, lower for a BTX system, and (2) the TMA fan replaces the processor heatsink fan, the ATX rear panel fan, and the fan required on a high power graphics card heatsink. Since it is also anticipated that chipset power will continue to increase, the memory controller chipset heatsink will soon require a fan in an ATX system, but not in a BTX system. That’s four to five ATX fans versus two BTX fans—obviously an improvement in total system cost and in the system’s acoustic behavior.

JD: Let’s talk about the transition period, which is estimated by Intel to be from three to seven years. You’ll have boxed processor SKUs (non-BTX and BTX versions, potentially doubling the SKUs stocked) as well as non-BTX and BTX chasses versions, likewise potentially doubling the SKUs stocked. All other things remaining equal, the added number of SKUs will result in higher purchasing, warehousing, and inventory management costs (more items to buy, stock, and manage) along with higher unit or carrying costs until 100% transition is reached. Volume price breaks are potentially affected. Lower volume purchased may result in higher unit cost. Or the system builder may purchase more units than needed to reach the price break quantity, resulting in additional carrying costs.

CR: RPG will ship both ATX and BTX thermal solutions with their boxed processors. The mix at any given time will be based largely on demand and forecasted demand. RPG is still working through shipping and warehousing related costs in preparation for their early-Q404 BTX boxed processor launch. I believe that inventory and pricing management is an issue with technology transitions in every portion of the computing industry. Recent transitions in memory, graphics, connectors, cables, etc., have all come with pricing and inventory management concerns. Not to be insensitive to this issue, but companies that will do well in a technology transition are ones that innovate to capture or create market share and manage their inventory and supplier pricing aggressively during their product transitions.

JD: There are alternatives to the BTX transition. After all, during most transitions, the industry develops various alternative “solutions” for any given problem. In the case of BTX, it is reasonable to conclude that the industry will devise improved thermal solutions applicable to non-BTX chassis, e.g., improved “thermally advantaged” chassis features such as heat pipe/cooler solutions, etc. The market may also move to less thermally demanding solutions (processors) in current chassis. Bottom Line: If there are two alternatives and both are Intel brand and Intel-supported as thermally effective, is it reasonable to conclude, all other things remaining equal, that the more expensive solution will be selected?

CR: When you consider that BTX cost savings increase with system power and that system power shows no signs of decreasing in any desktop segment (after all, there are 100W+ PCI-Express x16 graphics cards that will ship in 2004), the costs of doing business in ATX will become too burdensome. ATX thermally advantaged chassis and heatpipe thermal solutions are not cheap and there aren’t a plethora of design or technology options on the ATX horizon. Those that exist are not cheap. Therefore, while we expect there to be innovation in an attempt to extend ATX longevity, to most of the industry a BTX transition will eventually make good financial sense. Nonetheless, we expect the BTX transition to be led by those that offer performance computing ingredients or systems in their portfolio simply because the cost and acoustic advantages are more obvious in that segment.

Doubling the Debate With DDR2

DDR2 memory has more in common with BTX than you might suspect. Both have had official specs in place for almost a year. Both allow for far greater performance scaling. And both are flanked by legions of critics even before the first production SKUs hit store shelves.

Like DDR, DDR2 transmits data on both the rising and falling edges of a clock cycle, which is why a 200 MHz module yields an effective rate of 400 MHz. (For convenience, we’ll refer to all DDR-based modules by their effective data rates.) The initial speeds of DDR2 modules are 400 MHz and 533 MHz. However, relatively few 400 MHz parts are even in production. The reason is DDR2’s increased latency times.

Whereas mainstream DDR modules features a CAS latency (CL) of 3 and high-end modules even get down to CL2, DDR2 modules chime in at CL4. Thus a battle between clock speeds and latency times ensues when trying to judge which technology is faster. (Of course, we’ve already seen this trade-off when using high-end DDR modules in the PC4000 to PC4400 range.) The upshot is that a mainstream DDR2 module performs about on par with a mainstream DDR module. So if a 400 MHz DDR module performs slower than 400 MHz DDR2 and costs significantly less, would you want to buy the DDR2? Probably not.

In fact, this price vs. performance comparison explains why many of the new 915- and 925X-based motherboards you now see coming to market still use DDR rather than DDR2. Some vendors figure that in the first generation of DDR2-compatible chipsets and motherboards, end-users will be loathe to pay the extra cost for the newer memory format . How much of a premium is there? We hopped on MonarchComputer.com to see the pricing deltas from a large online system builder.
We searched for mainstream Corsair modules since Monarch offers these in both DDR and DDR2. One 256MB stick of DDR400 (sans heatsink) costs $54, and one 512MB of DDR400 costs $82. On the DDR2 (533) side, a 256MB stick costs $85 and a 512MB module is $153, so you can see that the newer format does carry a hefty premium to compensate for its low initial volumes. In all fairness, though, some Web sources are starting to release information showing that many of today’s 533MHz DDR2 modules can overclock quite easily to 667MHz once Intel’s overclock lock is circumvented. To keep the playing field more level, then, perhaps it would be more appropriate to judge against performance XMS 512MB modules, which sport prices of $129 for 500 MHz or $189 for 550MHz. Keep in mind, too, that there are only two or three motherboards in existence capable of reliably running DDR at 550MHz. So all told, the case against DDR2 on price doesn’t look quite so bad once you step beyond mainstream SKUs.

“DDR2 is a new technology that, just like the transition between Synchronous RAM and DDR, will have a crossover period where performance is comparable with the older technology,” says Mark Tekunoff, senior technology manager for Kingston. “At the initial speeds of 400 MHz for server systems, there is a good performance improvement over 333 MHz and 266 MHz designs. In the case of 533 MHz memory for desktops, you’ll typically find performance to be comparable or in some cases better. As pricing trends continue to go down for DDR2 memory from the semiconductor vendors, the DDR2 products will become a good value. If you build a system that you want to use for a long time and might want to upgrade in a year from now, consider DDR2.”

“We are in a technology transition period and it’s fairly typical for some of the initial buzz to be somewhat disappointing,” echoes Crucial Technology engineer Andy Heidelberg. “The introduction of DDR2 is no different. New products (memory, motherboards, BIOS, etc.) usually end up being refined a few times prior to mass consumption levels. But the performance gains we’d hoped for/expected with DDR2 are there. DDR2 offers several significant performance improvements over DDR, especially in terms of lower power and higher speeds. When there are mainstream DDR2-based systems available, the performance increase should be clear.
When you look beyond pricing and speed at how DDR2 actually works, the long-term benefits of the new format are undeniable. One of the reasons you see such a spike in high-speed DDR module pricing above 400 MHz is that PC3200 is the top speed at which DDR can be reliably and cost-effectively produced under JEDEC specifications. Most of these high-speed modules are obtained through speed binning of DDR400 parts. (Each module is tested and sorted into “bins” according to the maximum speed at which it will operate without failure.) In contrast, DDR2 is still at the beginning of its scaling range. Current roadmaps are paved to 800 MHz and no one is yet counting out speeds significantly higher than that.

The next big gain for DDR2 is a drop of nearly 50% in power consumption, down from 2.5V in DDR to 1.8V. This will be a strong selling point in mobile systems. DDR2 memory chips will incorporate On-Die Termination (ODT) to minimize memory signal reflections at high speeds, thus improving timing margins. Additionally, DDR2 memory chips will come in capacities up to 4 gigabits, so an eight-chip module could offer up to 4GB—and even that’s not the limit on DDR2 sizes. Unlike DDR, the new memory type supports die stacking, an advanced technique in which one chip is built atop another with a gap for air cooling left between them.

Note that DDR and DDR2 modules are socket incompatible despite having identical module lengths. DDR2 modules have 240 pins compared to DDR’s 184, the electrical configurations are different, and the “key,” or notch, on the edge connector is in a slightly different position to prevent insertion in the wrong socket.
Despite some of the initial negativity surrounding DDR2, we remain convinced that this is an essential technology that resellers should start integrating immediately into many of their system configurations.

“The initial market for DDR2 will be composed of early adopters, enthusiasts, and people wanting the newest features now,” says Kingston’s Tekunoff. “This segment of the market is very technical and would likely be asking for Kingston DDR2 to get the best quality memory to go into the new sytems they are evaluating. It’s supported on the newest motherboards based on Alderwood and Grantsdale chipsets in addition to other new features. DDR2 will be the technology in 2005/2006 and has the headroom needed for performance improvements. It will also help manage power consumption in laptops.”