As I sit writing this blog, the HP Power Assistant tells me my laptop is using about 16 watts.
I can also easily select several profiles to dial up or down performance and power usage. Now if only managing power usage on an HPC cluster was that easy. Other than turning off servers, most HPC systems lack even the most basic abilities of my laptop to monitor and control power usage.
Modern HPC servers, like the HP ProLiant SL390s G7, equipped with multiple GPUs, large memory capacities, and SSDs can approach power usage levels of nearly 100x the 16 watts my laptop is using. Of course in doing so, the SL390s can achieve more than 100x the performance of my laptop, thus it’s no surprise that the SL390s was used to build the greenest production supercomputer in the world, Tokyo Tech’s TSUBAME2.0 system, according to the most recent Green 500 list.
With this week’s tragic events in Japan, my thoughts and prayers go out to everyone in Japan impacted by the disaster. I don’t have any updates on the operational status of TSUBAME2.0, but I expect if it is currently in operation, it is operating at a reduced capacity to save power. Supercomputer centers outside of Japan, such as Lawrence Livermore National Labs, have been using their HPC resources to help assess radiation risks in Japan. TSUBAME2.0 does employ a number of intelligent energy management systems including Dynamic Power Capping which allows the system to control power usage.
Tools like the HP Power Advisor help HP engineers and customers consider power usage trade-offs before their supercomputer is even built. Unfortunately, some HPC customers still focus simply on acquisition costs, versus total cost of ownership which every day is impacted more by power costs. HP’s 94% efficient power supplies may cost a few dollars more than the less efficient power supply found in a low cost server, but the more efficient power supply can easily pay for itself several times over during the operational life of the system. Many other component choices also impact power usage. For instance, the DRAM memory on a server today often accounts for more power usage than the CPUs. Using 8 GB DIMMs may cost slightly more than using 4 GB DIMMs, but since both DIMMs use approximately the same power, the power savings add up quickly.
HP’s holistic approach to energy efficiency goes far beyond simple component choices and extends across systems and even facilities. Our Performance Optimized Datacenter (POD) can achieve significantly better power usage effectiveness than many traditional “brick and mortar” datacenters.
I am sure I speak for all HP employees in saying our immediate concern is for the safety and well being of all Japan’s citizens and others who are in the country. It will be a long road to recovery, but as Japan starts to rebuild, HPC technology from HP and other vendors will no doubt play a critical part in building a safer, more energy efficient, country.