BTX Desktop PCs – A Thermally Superior System Design That Failed Horribly

BTX Desktop PCs – A Thermally Superior System Design That Failed Horribly

Not much is written these days about the lowly desktop computer. Smartphones, tablets and laptops take center stage. You may disagree, but I believe the desktop is going to be around for a long time. Its relatively low price and super performance numbers can’t be matched in other form factors. Cloud computing that moves processing power to a remote location and continuing advancements in low power performance chips will further erode desktop sales, especially at the low end, but you’re going to have to pry this design from the dead hands of engineers and scientists who demand high-powered local computing. These types of systems are power pigs and even their lower performing counterparts are getting smaller (small form factor desktops), increasing thermal management challenges.

Why didn’t the market embrace a proposed standard in PC architecture design that offered, among other things, substantially improved thermal headroom? Major OEM’s produced it and millions were sold, yet it remains a virtual footnote in PC system design. I’m talking about BTX architecture (motherboard layout and system design).

Since the mid-1990’s the ATX (Advanced Technology Extended) desktop architecture had been king. Power and power densities of CPUs were still relatively low so this Intel design, later adapted to AMD processors, was never meant to address growing thermal challenges. In those days, Pentium II and III processors were still in the 20-40 watt power range with power densities in the low to high teens (Figure 1).

Figure 1: CPU Power Density (Source: Canturk Isci, Workload Adaptive Power Management)

CPU power densities

As you can see in Figure 2 (excuse my crude airflow drawings), a typical ATX layout positioned the CPU toward the rear of the machine with a single 100mm exhaust fan used to move air through the system. When necessary, a small fan was added on top of the CPU heat sink for further cooling but it distributed warm air throughout the system. Cool inlet air was impeded by forward connectors and memory modules that were perpendicular to, and positioned in front of, the CPU. Graphics cards where designed so that heat sink and associated fan faced toward the bottom of the case. From a thermal management standpoint, this design was marginal but adequate for the time.

Figure 2: Typical ATX Architecture

ATX Design

The long lived Pentium IV, with its’ every increasing power (80-115 watts in later iterations) and power densities (roughly 60-90 w/cm2), forced Intel to rethink ATX architecture and in 2003 it pushed a competing architecture whose primary purpose was better thermal management. The idea was simple. Give the CPU ample cool air and place all component in-line with the air flow.

Figure 3: Sample BTX Architecture

BTX Design

Figure 3 shows the first mass produced BTX design from Gateway (2004). Dell followed with a similar design the next year and HP as well as a few others soon got on board. The CPU was moved to the front of the machine where it could benefit from the coolest air. Rear exhaust fan diameter was generally increased to 120mm and in some cases (shown) at additional 120mm intake fan was added. A shroud was sometimes used to direct intake air over the CPU and move it in a straight line toward the exhaust fan. Additionally memory slots where positioned parallel to air flow and the motherboard was moved to the other side of the case. This allowed graphics cards to be flipped upside-down so the GPU fan received additional air and could more easily direct it toward the exhaust fan. Finally, there existed a thermally efficient desktop system design. Yes it was marginally more expensive – on the order of $10-$15 with larger fans and shrouding in addition to a non-standard layout – but the thermal benefit seemed to outweigh the cost premium.

Engineers are often asked to tackle thermal challenges after system layout is locked. I know it’s not best practice, but it happens more than we’d like to admit. BTX gave us a better platform to manage heat, but it failed despite the fact that the latest generation of desktop processors (Core i7) still has power densities considerably higher than their Pentium III counterparts.

So why did BTX fail to replace an aging ATX design? We’d like to hear your comments.

Please contact us if you’d like to learn more about how Celsia can help with your next heat sink project. We’ve worked on everything from consumer devices to industrial test equipment that require heat sinks to cool anywhere from a few watts to a few kilowatts.

Every Electronic Device Is Water Cooled ….. Indirectly

Every Electronic Device Is Water Cooled ….. Indirectly

Computers, ovens, and even light bulbs rely on water cooling to function; nearly 200 billion gallons a day in the US alone, most of it is fresh water. Some of it, roughly 4 billion gallons, evaporates into thin air each day. In today’s guest blog, Dr. Pablo Hidalgo – a thermal management research engineer at Georgia Institute of Technology – provides a wonderful overview of cooling techniques used in thermoelectric power plants and discusses a little used thermal management approach that will help us reclaim one of our most precious resources: fresh water.

Enhancing the Efficiency of Dry Cooling

Did you know that 41% of the water withdrawn in the US is used for cooling thermoelectric power plants while only 14% is dedicated to domestic and public supply? What if I told you that the amount of annual non-recoverable fresh water used to cool these facilities could grow five years’ worth of the worldwide potato crop?

Fresh water management is a challenge we need to tackle. Globally, approximately one billion people lack adequate access to fresh drinking water, and fresh water supplies are diminishing as our population grows and our climate changes. Moreover, the International Food Policy Research Institute has predicted a 120% increase in food prices by 2025 due to the shortage of fresh water resources.

I’m not trying to discuss a completely new form of power generation in this blog, or take a position towards renewable energy or nuclear. I simply want to make readers conscious of what it takes to keep their computer, tablet or phone running while viewing this article. I’ll also touch on how self-oscillating reeds might be used to improve the efficiency of dry cooling, an alternative to liquid cooled power plants.

Thermoelectric power plants generate about 90% of this nation’s energy and cooling them is no small task. There are basically three types of cooling systems used in these facilities: once-through, wet recirculating, and dry cooling. Only by improving the efficiency of the latter can we hope to reduce the estimated 4 billion gallons of fresh water that is lost through evaporation every day.

Figure 1: Once-Through Cooling System  1(Source: GAO Energy-Water Nexus 2009)

Once Through Cooling System

Used in just over 40% US thermoelectric power plants, once-through cooling requires a tremendous amount of natural source water that is pumped into liquid cooled condensers used to convert the steam coming from the turbines back to water. As its name suggests, the cooling water passes through the system only once, after which it is expelled back into the lake, river or ocean. While no appreciable amount of water is lost during the process, this system wreaks havoc on the environment in the form of dead marine life and changes to the ecosystem. Due partly to the Clean Water Act of the 1970s, once-through cooling is usually only seen in facilities built more than forty years ago.

Figure 2: Wet Recirculating Cooling System1

Wet Recirculating Cooling System


Wet-recirculating systems using cooling towers are the predominant type found, and being built, in the United States. Rather than being returned to its source as with once-through systems, warmed cooling water is sent to cooling towers and then reused. The great advantage of this system is that, once primed, it uses up to 95% less source water, greatly reducing the impact to aquatic life. On the flip side, water loss is substantially higher. According to the National Energy Technology Laboratory, a single 520 megawatt power plant using this system will lose 4,000 gallons of water per minute due to evaporation and another 1,000 gallons from blowdown (the discharge of mineral laden water).

And this brings us to the newest system being used – dry cooling.

Figure 3: Dry Cooling System1

Dry Cooling System

Currently, only a small fraction of power generating facilities use dry cooling. Rather than relying on evaporative cooling, these systems pass the steam though air cooled finned tubes which condense it back to water that is re-fed to the boiler. With no water loss due to evaporation or blowdown, these systems are ideal for conserving fresh water supplies, but they come at a higher price tag than either of the other two technologies. A part of that cost is due to the fact that dry cooling is much less efficient that evaporative cooling.

I am going to focus on some specifics of this cooling technology and elaborate on a potential strategy to improve it, which I am currently working on and that I believe can have multiple applications.  The efficiency of air cooled condensers (ACCs) is very poor, on the order of 0.3  and  this efficiency is reduced further when ambient air temperature is high (i.e. summer time). Under these conditions, the temperature difference between the air inlet of the ACC and the steam is reduced resulting in decreased system efficiency and increased back pressure to the turbine. The net result, power generation capability is reduced.

According to the Department of Energy, during the summer the plant efficiency and power generation can decrease up to 25%, which is precisely when the demand of electricity is highest.  The greatest benefit of air cooled condensers is the lack of use of water for cooling but they also have some significant drawbacks.  For instance, the real estate needed to achieve the same cooling performance is over 2 times larger compared to that of a wet cooling tower. This is part of the reason the cost to build one these facilities is several times higher than that of alternative solutions.

Another disadvantage is the dominant air-side thermal resistance.  A representative value of the heat transfer coefficient (HTC) of the fins is on the order of 30 W/m2K.  For a forced convection system, this is a very low value with poor heat transfer capability. The fact that the ACC’s air side channel Reynolds number is in the laminar regime also contributes to such poor heat transfer, but air cooled condensers are designed for minimized flow losses and fan power. Here’s an approach that improves this technology. It’s based on some results I obtained working on a different project but with significant similarities between them.

On a previous DARPA project at the Georgia Institute of Technology, I participated in the development of a novel technology to improve the heat transfer of high-power heat sinks used in aerospace for military applications. We integrated an array of self-oscillating reeds (SORs) in between the channels of a heat sink in order to improve thermal performance.

Figure 4 shows several images of the 2.5mm x 10mm reeds before and after integration in a full scale heat sink – click for video.  The reeds are fabricated using laser micro-machining techniques out of polyester because of its low density and ability to flutter using only a small amount air.

Figure 4: Reed Integration into Heat Sink


The SOR is mounted and constrained as a cantilever beam. As air passes over them, they flutter creating turbulence which minimizes the boundary layer and increases the heat transfer coefficient. Our studies showed a tremendous increase in the coefficient of performance (COP) of the channels. Under the same heat flux and similar wall temperature distribution in our test bed, the SOR-enhanced channel showed an increase in COP of 400% with respect to the baseline channel (no reeds) as shown in Figure 5.

Figure 5: Coefficient of Performance – Baseline (non-reed) and SOR Enhanced Channels

SOR Performance

Here, the flow rate for each case is measured in liters per minute, which represents two different Reynolds number (Re). For constant heat dissipation, the increase in mixing in the SOR-enhanced channel would result in a decrease in wall temperature.  Therefore, it’s possible to dissipate more heat while maintaining the same wall temperature.  In this case, our results have shown an increase in power (heat) dissipation of over 40% using this technology.

In addition to being adapted for use in a wide range of fields, data center and semi-conductor applications among others, self-oscillating reeds could be incorporated into the air cooled condensers used in thermoelectric power plants. It’s an important first step in making dry cooling an appealing alternative to wet-recirculating systems. Four billion gallons for fresh water could be saved every day.

Whether you’re involved in the thermal management of power electronics, servers or mobile devices, it’s important for all of us to help make power generation more economical and environmentally friendly. Increasing the efficiency of dry cooling is an area where thermal engineers can certainly have an impact.

Please comment on other ways we could improve the efficiency, cost and environmental impact of power generating facilities.

I’m also very interested in hearing ideas about how self-oscillating reeds might be used for other applications in your field of expertise.

About the Author

Dr. Pablo Hidalgo is a second level research engineer in the fluid mechanics research lab at the Georgia Institute of Technology. His specialties include experimental fluid mechanics, system level thermal management, convective heat transfer, heat sink design, CFD and conjugate heat transfer simulations, flow control, fluid-structure interactions and aerodynamics.

Dr. Hidalgo received his Ph.D. in Engineering Sciences and Mechanics from The University of Alabama in 2008 and his Bachelor’s and Master’s degrees in Aerospace Engineering from Saint Louis University in 2003 and 2005 respectively. He is actively involved in the program committee of Semi-Therm and is a member of APS-DFD.