Uploaded June 2022 | Updated September 2026, 5 days ago
AMD has revealed efficiency targets and some of the details of the upcoming RDNA 3 architecture, the first GPU with chiplets, slated to arrive later this year. Can AMD's GPU chiplet design revolutionize graphics cards like it did with CPUs? We spoke with Sam Naffziger, VP and Corporate Fellow at AMD, regarding some of the design challenges chip makers are facing and what's being done to continue driving higher performance and increased efficiency.
1:10 - Lighting round, news of the week
11:24 - Begin AMD GPU architecture discussion
11:58 - The Problems Facing the Tech Industry
14:41 - RDNA 3 Overview
15:20 - 5nm and What to Expect
18:42 - Chiplet Packaging
24:12 - How Much Power Will RDNA 3 Use?
29:25 - Rearchitected Compute Unit
31:17 - Optimized Graphics Pipeline
31:53 - No Matrix/Tensor Cores in RDNA 3
34:08 - Next-Gen AMD Infinity Cache and Infinity Fabric
38:15 - Process Nodes, Benefits of Having I/O on Older Process
42:20 - Analog Scaling for I/O Is Terrible
45:53 - Move to Fabless Was Great for AMD, TSMC Also Benefitted
47:42 - Cost of Design, Testing, etc. for Newer Nodes
49:56 - Where Will Nvidia Ada / RTX 40-Series Land
51:52 - AMD Gaining Ground Every Generation of RDNA GPU
53:00 - Moore's Law Slowing Down, Wrapping it Up
AMD has revealed efficiency targets and some of the details of the upcoming RDNA 3 architecture, the first GPU with chiplets, slated to arrive later this year. Can AMD's GPU chiplet design revolutionize graphics cards like it did with CPUs? We spoke with Sam Naffziger, VP and Corporate Fellow at AMD, regarding some of the design challenges chip makers are facing and what's being done to continue driving higher performance and increased efficiency.
1:10 - Lighting round, news of the week
11:24 - Begin AMD GPU architecture discussion
11:58 - The Problems Facing the Tech Industry
14:41 - RDNA 3 Overview
15:20 - 5nm and What to Expect
18:42 - Chiplet Packaging
24:12 - How Much Power Will RDNA 3 Use?
29:25 - Rearchitected Compute Unit
31:17 - Optimized Graphics Pipeline
31:53 - No Matrix/Tensor Cores in RDNA 3
34:08 - Next-Gen AMD Infinity Cache and Infinity Fabric
38:15 - Process Nodes, Benefits of Having I/O on Older Process
42:20 - Analog Scaling for I/O Is Terrible
45:53 - Move to Fabless Was Great for AMD, TSMC Also Benefitted
47:42 - Cost of Design, Testing, etc. for Newer Nodes
49:56 - Where Will Nvidia Ada / RTX 40-Series Land
51:52 - AMD Gaining Ground Every Generation of RDNA GPU
53:00 - Moore's Law Slowing Down, Wrapping it Up
