Monday, 29 August 2011

When will an IBM Watson-like computer cost less than $1000? - Quora

When will an IBM Watson-like computer cost less than $1000?

Watson is: "Powered by 10 racks of IBM POWER 750 servers running Linux, using 15 terabytes of RAM, 2,880 processor cores and is capable of operating at 80 teraflops."

3 Answers

Michał Strojnowski, PhD, computer science researcher

1 vote by Jarrod Cugley

This actually can be predicted with fairly high accuracy. Performance development of computing power per $ shows stable trend of doubling every 14 months (see top500.org). Hardware vendors predict that this trend will continue for at least next two decades. In 2011 Watson hardware costs about $1million, so it will cost less than a $1000 around 2023. There is also a question about software cost, because it is very likely that in 12 years this software will be obsolete and no longer supported, replaced with something much more sophisticated (and computation-costly).
This answer .
Please specify the necessary improvements.

Michał Strojnowski

In reality , there's no need for a personal watson computer. You just need answers. So i think a better way to measure watson's price is not price per system , but price per result ,i.e. price per answer.

According to some calculations[1], today it costs watson around $2.84 to answer a question(assuming Watson works at full capacity), using the IBM system.

Using amazon's cloud computing plaform it would go down to about $1.5 per question.

And if you build your own, it would get to around $0.5 per question. And that's before software and hardware optimization that could rapidly decrease the price.

And just for perspective , According to [2], google's revenue per query is $0.2.

[1]http://news.ycombinator.com/item...
[2]http://www.skrenta.com/2007/01/w...

This answer .
Please specify the necessary improvements.

Joe Talin

James William Murdock IV, IBM Researcher: Watson (Jeopardy!)

Michal's comment makes a lot of reasonable points. However, I am very skeptical of any predictions that far into the future. Also, I think a more important point with the cost of Watson, is the possibility that software improvements could reduce the computational requirements of Watson and thus make it cost-effective much sooner. In many cases, it is possible for clever software to compensate for the limits of a hardware platform. Even if it is not possible to completely implement every function of Watson on a less powerful platform, it may be possible to create a version that has many of the important functions and provides similar capabilities with much less demand on the computational hardware.