Researchers at Harvard and Microsoft have authored a paper that seeks to prove that a small, power-efficient core like the Intel Atom chip can be better suited for search.
The paper compares the performance, reliability, and cost of ownership of two microprocessors: a power-optimized “Harpertown” L5420 four-core Xeon, a 2.5-GHz, 45nm part based on the “Penryn” core; and the “Diamondville” dual-core Atom, a 1.6-GHz microprocessor also fabricated on a 45nm process.
For the test, the researchers used real Bing search queries that had been previously sampled. For each query, the ISN computed overall ranks for pages that matched the query for a 1GB production index, which was stored in memory to eliminate the variable time in which the data would have to otherwise be fetched from a hard disk.
The problem, the researchers concluded, is that the Atom is fundamentally weaker in processing the neural network algorithms that assign page rankings, even at a minimum load of 10 queries per second. The researchers concluded that it was reasonable to assume that Microsoft or another company would want to maintain some level of quality of service, so that users would receive relevant results and stick with the search engine.
Read the full article at extremetech.com.