Broadcast Archives

SR-016 | January 16, 2026 @ 4:00 PM EST

Bandwidth Constraints and the Roofline: Visualizing Performance Bottlenecks

Guest

Dr. Samuel Williams (Computer Scientist, Lawrence Berkeley National Laboratory)

Examined memory bandwidth limitations through the roofline performance model, discussing operational intensity measurement from FLOPS per byte, hierarchical rooflines for multiple cache levels, irregular access pattern challenges, latency versus bandwidth trade-offs, GPU accelerator characteristics with occupancy considerations, reduced precision impacts on compute ceilings and operational intensity, cache blocking and data layout optimizations, compute-bandwidth gap architectural responses, distributed system network bandwidth extensions, model prediction accuracy and systematic deviations, and compiler auto-tuning integration.

SR-015 | January 15, 2026 @ 4:00 PM EST

Redundancy and Reconstruction: Engineering Data Durability at Scale

Guest

Dr. Garth Gibson (Computer Scientist, Carnegie Mellon University)

Examined storage system reliability through RAID and erasure codes, discussing parity-based reconstruction using XOR properties, RAID levels balancing capacity efficiency against fault tolerance, write amplification from read-modify-write cycles, SSD impacts on failure models and performance assumptions, reconstruction time vulnerabilities and distributed strategies, Reed-Solomon and advanced erasure codes with Galois field arithmetic, distributed storage failure domains and hierarchical encoding, silent corruption detection through checksums, reconstruction bandwidth throttling, block versus file system redundancy integration, and evolution toward diverse customized schemes.

SR-014 | January 14, 2026 @ 4:00 PM EST

Open Architecture: RISC-V and the Case for ISA Independence

Guest

Dr. Krste Asanović (Computer Scientist, University of California Berkeley)

Examined RISC-V as open instruction set architecture alternative to proprietary ISAs, discussing modular design with frozen base and optional extensions, custom instruction space for domain-specific optimization, compiler support strategies from intrinsics to full backend extensions, formal verification and architectural testing, ecosystem development from toolchains to operating systems, vector extension's length-agnostic model, memory consistency options, security considerations, adoption patterns in embedded versus established markets, governance through RISC-V International, and vision for ubiquitous deployment.

SR-013 | January 13, 2026 @ 4:00 PM EST

Balancing Fairness, Throughput, and Latency in CPU Scheduling

Guest

Dr. Robert Love (Kernel Developer, Google)

Examined CPU scheduler fundamentals including fairness definitions through proportional time allocation, CFS virtual runtime tracking and sleep fairness, target latency and minimum granularity trade-offs, multicore scheduling with cache affinity and load balancing, real-time scheduling integration with priority inheritance, power management and heterogeneous cores, priority inversion solutions, O(1) versus CFS design philosophy, hierarchical cgroup scheduling, kernel preemption for latency, and future challenges with heterogeneous systems and accelerators.

SR-012 | January 12, 2026 @ 4:00 PM EST

Building Reliable Computation from Noisy Quantum Components

Guest

Dr. John Preskill (Theoretical Physicist, Caltech)

Examined quantum error correction fundamentals including syndrome measurement without state collapse, surface code topology and stabilizer checks, physical-to-logical qubit overhead ratios, fault-tolerant gate implementation constraints, error thresholds and decoder algorithms, qubit technology error characteristics, temperature effects on coherence, measurement back-action management, quantum memory applications, verification challenges, path from NISQ devices to error-corrected systems, and hardware-software co-design opportunities for reducing overhead.

SR-011 | January 11, 2026 @ 4:00 PM EST

Preserving Semantics While Chasing Speed: The Compiler Optimization Challenge

Guest

Dr. Chris Lattner (Co-founder, Modular AI)

Examined compiler optimization trade-offs between correctness and performance, discussing intermediate representations and SSA form, constant propagation and aliasing analysis, loop optimization including vectorization, instruction scheduling and register allocation, undefined behavior exploitation, floating-point reordering constraints, interprocedural optimization, profile-guided optimization, compiler verification, heterogeneous compilation, LLVM design principles, and machine learning's emerging role in optimization heuristics.

SR-010 | January 10, 2026 @ 4:00 PM EST

Layering and Performance: The Protocol Design Dilemma

Guest

Dr. Van Jacobson (Research Fellow, Google)

Examined network protocol design tensions between layered abstractions and performance requirements. Discussed TCP congestion control mechanisms including slow start and congestion avoidance, protocol overhead from headers and processing, hardware offload trade-offs, RDMA's abstraction sacrifices, content-centric networking's architectural shift, modern congestion control variants, backward compatibility constraints, and heterogeneous network challenges balancing transparency against optimization.

SR-009 | January 9, 2026 @ 4:00 PM EST

Between Memory and Storage: The Persistent Memory Dilemma

Guest

Dr. Steven Swanson (Computer Scientist, University of California San Diego)

Examined persistent memory technologies occupying the gap between volatile DRAM and non-volatile storage. Discussed byte-addressable persistence, crash consistency challenges requiring explicit cache flushes and memory fences, write endurance and wear leveling, latency asymmetry between reads and writes, programming models including file systems and transactions, persistent garbage collection, database architecture implications, and security concerns from data persistence.

SR-008 | January 8, 2026 @ 4:00 PM EST

Proof Before Silicon: Formal Verification of Hardware Correctness

Guest

Dr. Randal Bryant (Computer Scientist, Carnegie Mellon University)

Examined formal verification methods for proving hardware correctness mathematically, discussing Binary Decision Diagrams, symbolic model checking, SAT solvers, and theorem proving. Analyzed state explosion challenges, floating-point verification, processor verification complexity, security property verification post-Spectre, and memory consistency model verification. Explored compositional reasoning, abstraction techniques, specification quality issues, and the integration of formal methods in industrial design flows.

SR-007 | January 7, 2026 @ 4:00 PM EST

Silicon Specialization: The Architecture and Economics of Neural Network Accelerators

Guest

Dr. Norm Jouppi (Distinguished Engineer, Google)

Explored specialized hardware for neural networks, examining systolic arrays, reduced precision arithmetic, memory hierarchies, and arithmetic intensity optimization. Discussed programmability challenges, multi-accelerator interconnects, sparsity exploitation, and software compilation stacks. Analyzed power efficiency sources, design trade-offs between specialization and flexibility, verification complexity, and the economics of custom silicon versus general-purpose processors.

SR-006 | January 6, 2026 @ 4:00 PM EST

The Art of Constraint: Fundamental Trade-offs in Programming Language Design

Guest

Dr. Guy Steele (Computer Scientist, Oracle Labs)

Examined fundamental choices in programming language design including performance versus abstraction, static versus dynamic typing, explicit versus implicit parallelism, and error handling approaches. Discussed type system complexity, immutability trade-offs, compilation strategies, backward compatibility, metaprogramming, numeric computation, and principles guiding successful languages. Explored how different constraints lead to different optimal languages for different domains.

SR-005 | January 5, 2026 @ 4:00 PM EST

The Idle Power Problem: Energy Proportionality in Datacenter Computing

Guest

Dr. Luiz André Barroso (VP of Engineering, Google)

Examined energy proportionality in datacenter systems, analyzing why servers consume 50-60% of peak power when idle. Discussed static power sources, workload consolidation strategies, power state transitions, and cluster-level orchestration. Explored memory and network challenges, heterogeneous architectures, cooling efficiency, renewable energy integration, and the path toward more energy-proportional designs that balance efficiency with performance and reliability.

SR-004 | January 4, 2026 @ 4:00 PM EST

Trust No One: Byzantine Fault Tolerance and Adversarial Systems

Guest

Dr. Barbara Liskov (Computer Scientist, MIT)

Examined Byzantine fault tolerance protocols that guarantee correctness despite arbitrary component failures. Discussed the 3f+1 replica requirement, multi-phase consensus protocols, view change mechanisms, and cryptographic overhead. Analyzed threat models justifying Byzantine versus crash fault tolerance, performance trade-offs, and applications in adversarial environments, critical infrastructure, and blockchain systems.

SR-003 | January 3, 2026 @ 4:00 PM EST

Beyond Copper: The Physics and Economics of Photonic Interconnects

Guest

Dr. Keren Bergman (Electrical Engineer, Columbia University)

Explored silicon photonic interconnects for high-bandwidth chip-to-chip communication, examining physical limitations of copper at high data rates, electro-optic conversion efficiency, and wavelength division multiplexing. Discussed fabrication challenges including laser integration, thermal management of ring resonators, and manufacturing tolerances. Analyzed network topology implications, on-chip versus chip-to-chip applications, and the trade-offs between integration complexity and bandwidth scaling.

SR-002 | January 2, 2026 @ 4:00 PM EST

The Price of Safety: Memory Management Trade-offs

Guest

Dr. David Bacon (Computer Scientist, IBM Research)

Examined automatic memory management through garbage collection, analyzing the trade-offs between throughput, latency, and memory usage. Discussed generational collection, concurrent GC algorithms, write barrier overhead, and real-time GC constraints. Explored memory safety benefits, comparison with manual management and Rust's ownership system, large heap challenges, and the balance between runtime complexity and language complexity in achieving safe memory management.

SR-001 | January 1, 2026 @ 4:00 PM EST

The Coherence Contract: Hardware Promises and Software Assumptions

Guest

Dr. Sarita Adve (Computer Scientist, University of Illinois)

Explored cache coherence protocols and memory consistency models in multicore processors, discussing sequential consistency versus relaxed models, the interaction between hardware and language-level memory models, and verification challenges. Examined coherence protocol complexity, false sharing, heterogeneous system challenges, and the fundamental tension between abstraction simplicity and performance optimization in shared memory systems.