Suggestions
Anand Joshi
AI Analyst & Industry Executive Specializing in AI
Anand Joshi is a prominent figure in the artificial intelligence and semiconductor industries, currently serving as the AI Analyst at TechInsights.12 Here's a comprehensive overview of his background and expertise:
Professional Experience
Anand Joshi has over 25 years of experience in the semiconductor industry and is recognized as an artificial intelligence industry veteran.1 His career includes:
- AI Analyst at TechInsights (September 2022 - Present)3
- Independent Consultant and Managing Director at JP Data (2015 - Present)3
- Board Advisor at Datafoundry (June 2022 - Present)3
Expertise and Contributions
Anand Joshi is known for his:
- Leadership in AI coverage at TechInsights1
- Groundbreaking predictions on the AI semiconductor boom of 20161
- Thorough analysis of AI technologies and markets1
- Collaboration with various professionals to create solutions related to AI technology, market potential, and strategy1
His work has been influential in the industry:
- His graphs, reports, and numbers have been used by many AI chip start-ups, as well as major companies like Intel, Microsoft, and Qualcomm13
- He has advised and consulted for prominent AI companies, many of which have been acquired over the years13
Speaking Engagements and Reputation
Anand is a sought-after speaker and advisor:
- He frequently speaks on topics ranging from AI chipsets to applications and infrastructure13
- His credibility as an AI Analyst has led to numerous speaking engagements1
Educational Background
- MS in Electrical Engineering from Virginia Tech (1991 - 1994)3
- MBA from the University of California, Irvine - The Paul Merage School of Business (1996 - 1999)3
Additional Achievements
- Wrote the industry's first market report on Computer Vision in 20143
- Has built a reputation as a credible analyst in the AI field3
- Worked with the investor community on due diligence of AI product companies3
- Advised Fortune 500 companies on strategy3
Anand Joshi's extensive experience, analytical skills, and industry insights have made him a valuable asset in the rapidly evolving field of artificial intelligence and semiconductor technology.
Highlights
Folks, I'll attend the AI Hardware Summit next week and look forward to seeing some familiar faces. Please DM me if you wish to set up a time for a briefing/meeting.
https://lnkd.in/gEVKXbiC
Meta has released details of the training infrastructure for Llama3.1 in their paper.
- It took them 54 calendar days to train Llama3.1
- They used up to 16384 H100 with 80 GB HBM cards
- Meta servers, each with dual CPU and 8 accelerators were connected by 400 GPBS network
- Storage server supported 240 PB
- The GPU utilization was 38% when using 16K servers
They also mention that power fluctuations and GPU failures due to heat were the leading cause of training interruptions. When all GPUs simultaneously switched states, to write a checkpoint, for example, the power fluctuation was of the order of MegaWatts that stressed the entire power infrastructure. Interestingly GPU-HBM communication was the second leading cause for interruptions, although they don't provide any details on what caused it. There is a lot more in the 92-page paper with contributions from 450+ authors.
https://lnkd.in/gHXQG3me
