Microsoft Uses Intel FPGAs to Drive Better Search Results

Microsoft Uses Intel FPGAs to Drive Better Search Results

Microsoft was one of the first companies to experiment with FPGAs in data centers; the company began testing FPGAs with its Project Catapult designs back in 2011. That work has continued through the present day, with successive generations of FPGAs deployed to improve search results and lower response time. Today, Microsoft and Intel have issued joint statements on how the adoption of Intel’s latest Arria and Stratix families allows Bing to improve its overall results and speed delivery.

All of this is being deployed to drive what MS and Intel call Intelligent Search. You might also think of it as a bid by Microsoft to transform Bing from a search engine, which you visit to find websites to learn about something, to a website you learn things from directly. Microsoft notes, for example, that instead of searching for “Tundra Biome” and clicking a link, you can now search for “Tundra Biome facts” and see a quick set of data about the tundra biome.

Microsoft Uses Intel FPGAs to Drive Better Search Results

We set out to test this capability, but had trouble finding categories that would trigger it. Searches for current hot-topic political issues like “Donald Trump,” “Stormy Daniels,” “school shootings” and “March for Our Lives” (all issues and individuals that have been part of the general news cycle in recent weeks) did not return different results when we added the word “facts” to each search. Nor did searches for companies (“AMD” or “Intel”) or searches for many historical events (“Revolutionary War,” “Civil War,” “World War II,” “World Trade Center bombing”). Generic searches for the Republican and Democratic parties also did not return different results.

The only search term that triggered a separate result when we added the word “facts” was “climate change.” Intel’s own press release seems to imply that Bing will be capable of more nuanced parsing than simply requiring the word “facts” when it writes:

Intel FPGAs power the technology that allows Bing to quickly process millions of articles across the web to get you contextual answers. Using machine learning and reading comprehension, Bing will now rapidly provide intelligent answers that help users find what they’re looking for faster, instead of a list of links for the users to manually check. Now you can check it out on your own for lots of types of questions, such as “Is coffee good for you?” or “What are the mental health benefits of yoga?” that generate multiple perspectives. Or even ask “How many calories are in a hot dog?” and Bing will share the calories and the number of minutes running required to burn the calories ingested.

But it’s not actually clear which results are being summarized and presented as accurate and which are not. Bing’s own example uses the phrase “Summarized from 3 sources.” Search for “Climate change facts” and you’ll see the same phrasing. But search for “what are the mental health benefits of yoga” and the answer you get is listed as being “Answer from 3 sources.”

In short, the idea that Bing is trying to transform itself from a search engine to an answer site is interesting. It could lead to trends that reshape how people use the internet. But it’s an approach that seems to be quite limited at this point in time and appears unlikely to impact the majority of search results.

Continue reading

Intel Launches AMD Radeon-Powered CPUs

Intel's new Radeon+Kaby Lake hybrid CPUs are headed for store shelves. Here's how the SKUs break down and what you need to know.

Lasers Used to Create Negative Mass Particles

Researchers at the University of Rochester have worked out a way to create negative mass particles using, what else, lasers. Is there anything lasers can't do?

OnePlus May Have Accidentally Sent Clipboard Data to Chinese Server

The latest beta version of its custom "OxygenOS" Android build was sending user clipboard data to a server in China. Oops.

Astronomers Detect Mysteriously Contorted Fast Radio Bursts

The signals from FRB 121102 appear to be "twisted" in a way that indicates a very extreme stellar environment.