Sign up to save your podcasts
Or
Share Why Tesla FSD Stalled: The "Memory Wall" Explained
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Why Tesla FSD Stalled: The "Memory Wall" Explained
Why is Full Self-Driving taking so long? It isn't just a code problem—it is a physics problem. We analyze the "Von Neumann Bottleneck" and why current chips spend more time moving data than actually thinking.
The Silicon Roadmap:
HW 2.5: Why Nvidia chips failed the "Batch Size 1" test.
HW 3.0: The "SRAM" breakthrough that solved latency.
HW 4.0: Brute-forcing bandwidth with GDDR6 (Gaming Memory).
AI 5: The liquid-cooled, 800-watt "Data Center on Wheels."
Timestamps:(00:00) The "Server on Wheels" Architecture(00:32) HW 2.5: The Latency Trap (Batch Size 1)(01:25) Over-Precision: Floating Point vs. Integer(01:43) HW 3: The SRAM Breakthrough(02:00) The "Memory Wall" (Transformers vs. Bandwidth)(02:30) HW 4: Brute Forcing with GDDR6(03:03) AI 5: The 800-Watt Liquid Cooled Monster(03:48) The Bifurcation: Are HW3 Cars Obsolete?
View all episodes
By Tech DadWhy is Full Self-Driving taking so long? It isn't just a code problem—it is a physics problem. We analyze the "Von Neumann Bottleneck" and why current chips spend more time moving data than actually thinking.
The Silicon Roadmap:
HW 2.5: Why Nvidia chips failed the "Batch Size 1" test.
HW 3.0: The "SRAM" breakthrough that solved latency.
HW 4.0: Brute-forcing bandwidth with GDDR6 (Gaming Memory).
AI 5: The liquid-cooled, 800-watt "Data Center on Wheels."
Timestamps:(00:00) The "Server on Wheels" Architecture(00:32) HW 2.5: The Latency Trap (Batch Size 1)(01:25) Over-Precision: Floating Point vs. Integer(01:43) HW 3: The SRAM Breakthrough(02:00) The "Memory Wall" (Transformers vs. Bandwidth)(02:30) HW 4: Brute Forcing with GDDR6(03:03) AI 5: The 800-Watt Liquid Cooled Monster(03:48) The Bifurcation: Are HW3 Cars Obsolete?