[Lex Computer & Tech Group/LCTG] AI Energy Requirements

Tue Dec 31 15:51:46 PST 2024

 The Idaho National Laboratory has been the proving ground for several experiments in nuclear reactor design, some of which involved people from Argonne National Laboratory, outside of Chicago. 
My Dad worked at Argonne as a "Senior Chemist", though most of his work was in physics (dimensional changes in materials when exposed to nuclear radiation). A lot of his work involved counting interference fringes as a way of getting very precise measurements of dimensional changes. 
In one memorable test, the Idaho Lab ran a nuclear reactor of a design similar to the Westinghouse reactors with diminishing amounts of coolant, until there was almost no coolant remaining. Unlike Chernobyl (which had graphite-based moderation and a cooling system with a "positive void coefficient") this reactor spewed many gallons of somewhat radioactive water out over the Idaho desert. No threat to anyone from the amounts of residual radiation, but you would not want to be in the direct path of it. 
I will be looking forward to attending the meeting where Dr. Griffith will present. 
-- Bob Primak 

    On Tuesday, December 31, 2024 at 04:17:42 PM EST, Jerome Slate <slatemd at comcast.net> wrote:  

  Dear LTCG Group,
 No worries about the AI energy load ~ 40% increase in USA energy output by 2050.
 Small modular nuclear reactors are coming to the rescue.
 We will have a live speaker on this topic on Feb 12. 
 George W Griffith PhD from the Idaho National Laboratory, part of the Dept of Energy, will present this topic.
 We are all saved.

 Jerry Slate

 On 12/31/2024 11:05 AM, Robert Primak via LCTG wrote:

 The problem is, New Scientist has a paywall. To read the article we would have to subscribe. (Which might not be a bad idea for some of us.)  
  But anyone can listen to this audio-only breakdown of the same fundamental concepts. 
  How Running AI Computations Backwards Can Save Energy
  https://www.youtube.com/watch?v=kKAwpyj98m0   The Singularity Labs | 12:35 m:s

  Fascinating discussion, and makes the concepts much clearer.  
  BTW, reverse-time computation has nothing to do with time as in space-time. It's a technique of doing some calculations back-to-front, as well as front-to-back. It can be much more efficient in terms of CPU and NPU loads, as well as making the computations happen faster.  
  Pruning of irrelevant background details has an even greater effect on computational efficiency.  
  Given the huge energy demands of AI, as well as AI's tendency to use up a lot of GPU and other hardware resources, this is an important area of concern.  
  -- Bob Primak  
      On Tuesday, December 31, 2024 at 10:10:16 AM EST, John Rudy via LCTG <lctg at lists.toku.us> wrote:  

    #yiv2250455876 filtered {}#yiv2250455876 filtered {}#yiv2250455876 p.yiv2250455876MsoNormal, #yiv2250455876 li.yiv2250455876MsoNormal, #yiv2250455876 div.yiv2250455876MsoNormal {margin:0in;font-size:12.0pt;font-family:sans-serif;}#yiv2250455876 a:link, #yiv2250455876 span.yiv2250455876MsoHyperlink {color:#467886;text-decoration:underline;}#yiv2250455876 span.yiv2250455876EmailStyle17 {font-family:sans-serif;color:windowtext;}#yiv2250455876 .yiv2250455876MsoChpDefault {}#yiv2250455876 filtered {}#yiv2250455876 div.yiv2250455876WordSection1 {}   
I have been getting this very interesting magazine for a couple of months.  It is mostly short stories on a VERY wide swath of technical issues.  Seems to be UK based.  It is a weekly of about 40 pages, and  the 12/29 article had this article on the cost of AI computing and the way in which computers use so much money erasing data.  I only partially understand it but it appears to be well worth reading.  Then someone with more knowledge that I can explain it at a meeting.

https://www.newscientist.com/article/mg26435231-300-how-to-fix-computings-ai-energy-problem-run-everything-backwards/

John Rudy

781-861-0402

781-718-8334  cell

13 Hawthorne Lane

Bedford MA

jjrudy1 at comcast.net

    ===============================================
 ::The Lexington Computer and Technology Group Mailing List::
 Reply goes to sender only; Reply All to send to list.
 Send to the list: LCTG at lists.toku.us      Message archives: http://lists.toku.us/pipermail/lctg-toku.us/
 To subscribe: email lctg-subscribe at toku.us  To unsubscribe: email lctg-unsubscribe at toku.us
 Future and Past meeting information: http://LCTG.toku.us
 List information: http://lists.toku.us/listinfo.cgi/lctg-toku.us
 This message was sent to bobprimak at yahoo.com.
 Set your list options: http://lists.toku.us/options.cgi/lctg-toku.us/bobprimak@yahoo.com

  ===============================================
::The Lexington Computer and Technology Group Mailing List::
Reply goes to sender only; Reply All to send to list.
Send to the list: LCTG at lists.toku.us      Message archives: http://lists.toku.us/pipermail/lctg-toku.us/
To subscribe: email lctg-subscribe at toku.us  To unsubscribe: email lctg-unsubscribe at toku.us
Future and Past meeting information: http://LCTG.toku.us
List information: http://lists.toku.us/listinfo.cgi/lctg-toku.us
This message was sent to slatemd at comcast.net.
Set your list options: http://lists.toku.us/options.cgi/lctg-toku.us/slatemd@comcast.net

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.toku.us/pipermail/lctg-toku.us/attachments/20241231/a5504b6a/attachment.htm>