What Do Overclockers and Supercomputers Have in Common?

Pages: 1 2

Overclockers Keep Cool

When it comes to PCs, nobody is more dedicated to squeezing the last bit of performance out of a system than enthusiasts. Overclockers in particular are known to go to great lengths to eke out extra performance and run components faster than the specifications. Part of the allure is that mid-range CPUs or GPUs are quite affordable, while the fastest products are exponentially more expensive. Currently, a 3.06GHz i7-950 runs about $250 online; while the 3.33GHz i7-975 is roughly $1000. Boosting the core clock frequency by a modest 10% and the QPI frequency by 33% could save you $750. Conversely, taking a top of the line CPU that runs at 3.4GHz and pushing it up to 5GHz gives you performance that money cannot really buy.

As a previous overclocking article discussed, the speed that a CPU (or any chip) runs at is largely a function of the design and manufacturing. Due to subtle variations in manufacturing, some chips can run faster or slower, while others run cooler or hotter. Companies will then take their production and bin the chips into product SKUs with different TDPs and frequencies. However, the goal of binning is not to achieve the best performance and power for each chip, but is partially determined by marketing. Often chips are specified to run slower, simply to meet demand for lower cost components. Consequently, some chips can run considerably faster than the specifications would indicate, while others are already at the limits of the silicon.

There are many products that cater to overclockers, including power supplies, motherboards and specially selected memory modules. In the last few years, hardware vendors have even launched CPUs and GPUs with unlocked multipliers to make overclocking easier. Enthusiasts are willing to pay to get this extra performance and these products often come at a steep premium. But the area where overclockers tend to get the most extravagant components and spend the most money is cooling.

Conventional wisdom amongst enthusiasts is that cooling is perhaps the single most important factor in performance and stability of overclocked systems. Keeping the CPU and other chips cool is critical for good results and avoiding permanent damage. Early on, many cooling solutions were home-brewed projects, but over the last few years, all sorts of exotic cooling solutions have sprung up. The simplest cooling solutions are oversize heatsinks and fans (HSF) that are higher quality than traditional HSFs. The next step up is replacing the fan with watercooling for even better heat removal. Heat exchangers (e.g. Peltiers or thermo-electric coolers like those found in refrigerators) are more powerful still and can work with either cooling technique to dramatically reduce chip temperatures. The most aggressive systems use liquid nitrogen or other coolants, rather than just water. These products range from perhaps $30 on the low-end to several hundred dollars for the more complex solutions. Figure 1 shows some of these solutions, ranging from the more mundane to exotic.

Figure 1 – Air, water and thermo-electric cooling solutions

While enthusiasts are the most knowledgeable of consumers, few are electrical engineers or physicists with a comprehensive understanding of the impact of changes in voltage, temperature and frequency. Much of the community relies on conventional wisdom and accumulated anecdotes for guidance. While this has worked relatively well, it is no substitute for thousands of systematic tests under controlled conditions with accurate measurements. Chip companies and leading vendors spend considerable time, money and effort to understand these issues, but they rarely share details with the public. In this article, we explore the impact of cooling on performance and power efficiency, and real world examples from CPU manufacturers and discuss the implications.

Pages:   1 2  Next »

Discuss (38 comments)