Slashdot Mirror
Intel Releases Several Projects to Help Save Power
GeekyBodhi writes "LessWatts.org is Intel's new website that hosts several power saving tools. As Linux.com reports, it also shares tips and tricks to help optimize power consumption on hardware from portable devices running on batteries to large data centers. 'LessWatts.org is not about marketing, trying to sell you something or comparing one vendor to another. LessWatts.org is about how you can save real watts, however you use Linux on your computer or computers.' As reported on Slashdot earlier, this isn't the first time Intel has tried to help Linux users cut their power bills. In May, the company launched the PowerTOP program that monitors individual processes to keep track of power consumption. The project comes at a time when more vendors are pre-installing Linux on handhelds and laptops." Linux.com and Slashdot are both owned by SourceForge.
It would be interesting to compare the actual power usage between two sets of IDENTICAL test systems, one set running Linux and the Intel Optimizations described in the site, and the second set running the equivalent commercial MICROSOFT software (Server 2003, Vista, XP, etc..) running equivalent real-world workloads.... (And I suppose some sort of Linux Anti-Virus software would need to be procured and added to the Linux test systems to make it fair...)
Any "PowerTaskManager" or something :(?
Please?
on modern Intel CPUs, the overwhelming majority are used to compensate for the fact that the CPU is ridiculously faster than the memory to which it interfaces. IOW, the cache. These CPUs consume between 80 and 120 watts of power. The reduced power versions use only 50.
By way of comparison, the 1 GHz AMD Geode runs on about 1 watt of power, and ARM processors can get by for even less.
By way of further comparison, a register to register transfer can be completed in 1/2 clock cycle. Contrast this with a read-modify-write memory cycle where a word is fetched from one memory location, modified, and written to a distant location, which will take 4 memory cycles (which typically runs at 1/2 or less of the clock speed of the CPU).
The power consumption problem is due more to the fact that compensating for this difference in speed requires a large SRAM cache on the die. And even then, it's not perfect - if you do things which routinely involve cache misses (such as video encoding, etc...) the CPU is stuck operating at the effective speed of the memory bus.
The key to reducing CPU power consumption is to use lower-latency memories, which require smaller on-die caches for a given performance level. We could double the throughput of DDR SDRAM by simply demultiplexing the address and data busses, similar to the way SRAM functions. There's no requirement in the underlying storage structure of DRAM to require separate row and column addresses; it's just a historical artifact. Originally, before DRAM and SDRAM became popular, computers were built with SRAM because its lower latency allowed even slow CPUs to work efficiently. But DRAM promised lower cost (via fewer bus lines) and lower power consumption (bits stored in charged capacitors, rather than cross-coupled transistors), at the price of latency, and the rest, unfortunately, is history.
The society for a thought-free internet welcomes you.
The sad thing is, efficient power supplies are a really marginal cost increase compared to the whole system. (My 375 Watt 80% efficient one cost me $60, including shipping). But since every OEM in existence insists on spending the absolute minimum on power supplies, you get lousy 60% efficiency, and a lot of blown power supplies.
Liberte, Egalite, Fraternite (TM)
Power supplies are more efficient at lower temperatures (around 25C), so don't send the heat from your processor to the PS as most people do using a horizontal cooler over the heat sink.
Instead use a heat sink that have a vertical cooler like Zalman CNPS9500