Intel Technology | Next-Gen Intel Core Ultra’s New NPU Explained | Talking Tech | Intel Technology @IntelTechnology | Uploaded August 2024 | Updated October 2024, 3 hours ago.
Today on Talking Tech we’re joined by Darren Crews, Sr. Principal Engineer of NPU Architecture, who walks us through a brief history of Intel NPUs up to the fourth and most capable iteration in Lunar Lake. Learn how evolving trends in prominent client AI workloads drive our microarchitectural design and implementation. Discover why we place more emphasis on vector compute this generation, and how we’ve optimized for efficiency by minimizing the NPU’s reliance on system memory.
Learn about Xe 2 architecture, graphics, and AI on Lunar Lake: youtu.be/x2oD4d3hg6I?si=M5k67TdCzQGECjNr
What else is new for Lunar Lake? Here’s the full breakdown: youtu.be/hmlDCAiD1bA?si=GIEQ-IwvPdfD8kAW
Chapters:
00:00 – Intro
00:59 – Designing for future workloads
01:53 – Common workloads today
03:05 – What is a large language model?
04:02 – NPUs through the years
05:39 – The major goals for Lunar Lake
06:28 – Achieving greater efficiency
10:00 – What is a “first token?”
11:49 – Neural compute engine improvements
12:48 – What are TOPS?
15:56 – Optimizing for vector operations
19:15 – DSP block improvements
20:19 – Summary and conclusion
About Intel Technology: Intel has always been at the forefront of developing exciting new technology for business and consumers including emerging technologies, data center servers, business transformation, memory and storage, security, and graphics. The Intel Technology YouTube channel is a place to learn tips and tricks, get the latest news, and watch product demos from both Intel and our many partners across multiple fields.
Connect with Intel Technology:
Visit Intel Technologies WEBSITE: intel.ly/IntelTechnologies
Follow Intel Technology on TWITTER: intel.ly/3MmhqET
Today on Talking Tech we’re joined by Darren Crews, Sr. Principal Engineer of NPU Architecture, who walks us through a brief history of Intel NPUs up to the fourth and most capable iteration in Lunar Lake. Learn how evolving trends in prominent client AI workloads drive our microarchitectural design and implementation. Discover why we place more emphasis on vector compute this generation, and how we’ve optimized for efficiency by minimizing the NPU’s reliance on system memory.
Learn about Xe 2 architecture, graphics, and AI on Lunar Lake: youtu.be/x2oD4d3hg6I?si=M5k67TdCzQGECjNr
What else is new for Lunar Lake? Here’s the full breakdown: youtu.be/hmlDCAiD1bA?si=GIEQ-IwvPdfD8kAW
Chapters:
00:00 – Intro
00:59 – Designing for future workloads
01:53 – Common workloads today
03:05 – What is a large language model?
04:02 – NPUs through the years
05:39 – The major goals for Lunar Lake
06:28 – Achieving greater efficiency
10:00 – What is a “first token?”
11:49 – Neural compute engine improvements
12:48 – What are TOPS?
15:56 – Optimizing for vector operations
19:15 – DSP block improvements
20:19 – Summary and conclusion
About Intel Technology: Intel has always been at the forefront of developing exciting new technology for business and consumers including emerging technologies, data center servers, business transformation, memory and storage, security, and graphics. The Intel Technology YouTube channel is a place to learn tips and tricks, get the latest news, and watch product demos from both Intel and our many partners across multiple fields.
Connect with Intel Technology:
Visit Intel Technologies WEBSITE: intel.ly/IntelTechnologies
Follow Intel Technology on TWITTER: intel.ly/3MmhqET
