Here's how much computing power Google DeepMind needed to beat Lee Sedol at Go
The London startup, which was acquired by Google for approximately £400 million in 2014, ran its "AlphaGo" algorithm on multiple machines to get the performance it needed to outsmart world number one Lee Sedol.Altogether, DeepMind used 1,202 CPUs and 176 GPUs, according to a paper in Nature that was published by several of DeepMind's employees in January. That is about 25 times as many CPUs (central processing units) and GPUs (graphics processing units) as the single-computer version of AlphaGo.
The game on Wednesday lasted roughly three and a half hours and is the first of five Go games between Google DeepMind's AlphaGo AI and Lee. At the end of Wednesday's game, AlphaGo was down to 5:30 of its time, while Lee had 28:28 left on the clock.
The next game will be held on Thursday at 1PM Korea Standard Time, followed by games on Saturday, Sunday, and Tuesday.The winner of the so-called "Challenge Match" must win at least three of the five games in the tournament.At the end of the tournament, which is being played in a luxury hotel in Seoul, South Korea, the winner will receive $1 million in prize money.
You can watch the entire first game here:
- India surpasses 2 lakh mark in daily COVID-19 cases with 1038 new fatalities
- Macrotech Developers IPO — check share allotment date, listing date and more
- Throwback to some of the best IPL ads made over the years
- Infosys, Wipro, HDFC Bank and other stocks to watch out for on April 15
- India's largest opposition party blames national media for its lack of influence — but its new YouTube channel may only be targeting the Hindi belt