Intel EIST SpeedStep
Nearly seven years ago at the Intel Developer Forum in Palm Springs California, Intel had first demonstrated Geyserville Technology. This dual-speed technology, when paired with a 0.18-micron mobile Pentium processor, had offered a distinct advantage as it was able to run in two processing modes -- maximum performance and battery optimized. Geyserville Technology allowed the processor to run slower and decrease the voltage when the mobile solution was running off of a battery (in order to reduce power consumption and allow robust performance without sacrificing battery life), but when the laptop was then plugged into an AC outlet, the frequency would return to normal. With the introduction of the Intel Mobile Pentium 3 processors, the dual-speed Geyserville was renamed to SpeedStep Technology. Shipping at that time were all sub-1GHz laptops yet SpeedStep/Geyserville proved to be quite a valuable asset to Intel. Since the Geyserville inception, many of Intel's desktop and mobile processing technologies have changed or been eliminated from the Coppermine 180nm process to Katmai 3D, but what has remained, however, is Intel's SpeedStep Technology. During this time SpeedStep has been revitalized with the ability to have multiple power-states as well as the changing the frequency depending upon the computing power needed rather than strictly the power source. This technology can continue to be found in processors from their mobile Pentium M and Core Duo parts to the desktop Pentium 4's (600 series) and Pentium D (excluding 820) CPUs under the name EIST, or rather Enhanced Intel SpeedStep Technology. Another power innovation on the next-generation Core Duo processors is Intel Dynamic Power Coordination and Enhanced Intel Deeper Sleep. These technologies allow dynamic cache sizing and only sending power to the specific areas of the processor that it actually needs to compute a specific operation. The only system requirements for SpeedStep are a supportive Intel CPU, Chipset, BIOS, and operating system. Enhanced Intel SpeedStep Technology, the updated technology found in the Pentium 4/D/M processors, are not Chipset dependent as EIST voltage/frequency selection is software controlled by writing to the processor MSR's (Model Specific Registers). While SpeedStep is compatible with operating systems going back to Microsoft Windows 95, Linux users have been able to benefit from the technology when using the Linux 2.6 kernel and cpufreq module. This vast operating system support without the need for a special software configuration was due to SpeedStep originally being controlled by toggling the GHI# pin. Intel's direct competition to SpeedStep Technology is AMD's Cool 'n' Quiet, which is a fork of their original PowerNow design. Similar to as we had done last week when publishing our AMD Cool 'n' Quiet Linux investigation, we are back today with our findings from our most recent SpeedStep investigation using an Intel Pentium M processor and the Linux 2.6.15 kernel with Fedora Core 4. For reference, below is the notebook computer we had used throughout the testing process of this article as well as the prominent software versions.
|Processor:||Intel Pentium M 750 (1.86GHz)|
|Motherboard:||IBM R52 18494WU (i915PM + ICH-6M)|
|Memory:||2 x 1024MB OCZ DDR2-533|
|Graphics Card:||ATI RADEON X300 64MB (dedicated)|
|Hard Drives:||Toshiba Slim MK1032GAX|
|Operating System:||Fedora Core 4|
|Linux Kernel:||2.6.15-1.1831_FC4 (i686)|
|GCC - GNU Compiler:||4.0.0|
|Graphics Driver:||ATI v8.22.5|
Unlike our AMD Cool 'n' Quiet tests where we were forced to monitor the CPU heatsink temperature, due to the lack of LM_Sensors v2.9.2 detection on both the ASRock 939Dual-SATA2 and Tyan Tomcat K8E-SLI, with our SpeedStep testing we are relying heavily upon some of the integrated IBM/Lenovo ThinkPad features. The CPU temperature was monitored through the ThinkPad ACPI system temperature sensors. In addition, we monitored the power consumption when the notebook was running off its battery and then again when it was connected to a power outlet. We monitored the present Wattage rate when the system was running off a battery using the ACPI battery state, and the Seasonic PowerAngel to monitor the AC adapter Wattage. We dealt with both the battery and AC power tests, due to the system being designed for optimal performance (in both Linux and the IBM BIOS) when connected to an outlet, while additional power management items are enabled when running off the battery. Even though ATI's v8.22.5 drivers support PowerPlay, which is similar to SpeedStep except on the graphics front, we had ran the Mobility X300 64MB at its standard operating frequencies throughout the entire process. Similar to our Cool 'n' Quiet piece, we had also performed some compilation benchmarks to gather its effects on system performance. During the complete testing process, the ambient room temperature was properly maintained as well as all other constants and standard Phoronix testing practices. In addition, no external laptop coolers were used during the testing process. Before moving onto the tests, on the following pages are a few notes regarding cpufreq and Intel's SpeedStep as well as information pertaining to IBM ACPI features.