Shows / Condor Currents
Condor Currents
Condor Currents delivers concise breakdowns of the latest computer architecture research and RISC-V developments. Each episode covers recent arXiv papers on data prefetching, branch prediction, and microarchitecture innovations, plus news from the open silicon community. Brought to you by Condor Computing, high-performance RISC-V processor IP from Andes Technology.
Episodes
- Condor Currents
SPICE: Speculative Prefetching with Low-Rank Expert Surrogates and Heterogeneous Orchestration for MoE Inference Acceleration
## Episode Summary In this episode, we cover: - **SPICE: Speculative Prefetching with Low-Rank Expert Surrogates and Heterogeneous Orchestration for MoE…
- Condor Currents
Formal Performance and Compile Time Guarantees for Compiler Optimization Heuristics
## Episode Summary In this episode, we cover: - **Formal Performance and Compile Time Guarantees for Compiler Optimization Heuristics** (arXiv) - **Beyond…
- Condor Currents
From Retrieved Context to Runtime Control: Adaptive Compression for Edge-based RAG
## Episode Summary In this episode, we cover: - **From Retrieved Context to Runtime Control: Adaptive Compression for Edge-based RAG** (arXiv) - **DTX: A…
- Condor Currents
A Thread-Register Decoupled GPU Execution Model for Efficient Tensor Computation
## Episode Summary In this episode, we cover: - **A Thread-Register Decoupled GPU Execution Model for Efficient Tensor Computation** (arXiv) - **Performance…
- Condor Currents
FPGA Lifecycle Management for RISC-V Systems
## Episode Summary In this episode, we cover: - **FPGA Lifecycle Management for RISC-V Systems** (arXiv) - **CTTE: An Open Dual-Protocol RISC-V Trace Encoder…
- Condor Currents
ETHEREAL: A 25.6-$μ$s/inf. Low-latency Event-driven Graph-neural-network Processor for High-resolution Vision at the Edge
## Episode Summary In this episode, we cover: - **ETHEREAL: A 25.6-$μ$s/inf. Low-latency Event-driven Graph-neural-network Processor for High-resolution Vision…
- Condor Currents
Testing the EPYC Conjecture on Real Hardware: MoA-Guided Dense Matrix Multiplication on NCSA Delta (AMD EPYC 7763 Milan)
## Episode Summary In this episode, we cover: - **Testing the EPYC Conjecture on Real Hardware: MoA-Guided Dense Matrix Multiplication on NCSA Delta (AMD EPYC…
- Condor Currents
SLAC: Access-Driven CPU-to-GPU Side-channel Attacks via System-Level Cache on Apple Silicon
## Episode Summary In this episode, we cover: - **SLAC: Access-Driven CPU-to-GPU Side-channel Attacks via System-Level Cache on Apple Silicon** (arXiv) -…
- Condor Currents
What Actually Serializes GPU LZ77 Decode: Three Decoders, Three Mechanisms, and an Encode-Time Lever That Removes the Last One
## Episode Summary In this episode, we cover: - **What Actually Serializes GPU LZ77 Decode: Three Decoders, Three Mechanisms, and an Encode-Time Lever That…
- Condor Currents
SLAC: Access-Driven CPU-to-GPU Side-channel Attacks via System-Level Cache on Apple Silicon
## Episode Summary In this episode, we cover: - **SLAC: Access-Driven CPU-to-GPU Side-channel Attacks via System-Level Cache on Apple Silicon** (arXiv) -…
- Condor Currents
Why Do Prefetchers Fail? Let Agents Answer
## Episode Summary In this episode, we cover: - **Why Do Prefetchers Fail? Let Agents Answer** (arXiv) - **Machine Shape and Hierarchical Blocking: A…
- Condor Currents
APEX: Adaptive Expert Prefetching for Memory-Efficient Edge MoE Inference
## Episode Summary In this episode, we cover: - **APEX: Adaptive Expert Prefetching for Memory-Efficient Edge MoE Inference** (arXiv) - **Aneto: Predicting…
- Condor Currents
ArchAgent v2: A Case Study with the Data Prefetching Championship
## Episode Summary In this episode, we cover: - **ArchAgent v2: A Case Study with the Data Prefetching Championship** (arXiv) - **The Fallacy of Independent…
- Condor Currents
A Centralized Performance Monitoring Architecture for Heterogeneous Multicore SoCs
## Episode Summary In this episode, we cover: - **A Centralized Performance Monitoring Architecture for Heterogeneous Multicore SoCs** (arXiv) - **MCHA: A…
- Condor Currents
Heterogeneous LLM Serving with General-Purpose Processing-Near-Memory for Retrieval-Based Sparse Attention
## Episode Summary In this episode, we cover: - **Heterogeneous LLM Serving with General-Purpose Processing-Near-Memory for Retrieval-Based Sparse Attention**…
- Condor Currents
PLoRA: An NDP-Enhanced Pooled-Memory System for Cost-Efficient Multi-LoRA Serving
## Episode Summary In this episode, we cover: - **PLoRA: An NDP-Enhanced Pooled-Memory System for Cost-Efficient Multi-LoRA Serving** (arXiv) - **Breaking…
- Condor Currents
Architectural Implications of Agentic AI Workflows
## Episode Summary In this episode, we cover: - **Architectural Implications of Agentic AI Workflows** (arXiv) - **On the Limits of Machine-Learned Ranking for…
- Condor Currents
LACE: Large Language Model Aided Multi-Agent Framework for Agile RISC-V Instruction Extension
## Episode Summary In this episode, we cover: - **LACE: Large Language Model Aided Multi-Agent Framework for Agile RISC-V Instruction Extension** (arXiv) - **A…
- Condor Currents
Themis: Software-Defined Hardware Prefetching
## Episode Summary In this episode, we cover: - **Themis: Software-Defined Hardware Prefetching** (arXiv) - **Beyond Static Policies: Dynamic Selection Among…
- Condor Currents
Selective KV Cache Protection for Noise-Resilient LLM Inference on Analog Compute-In-Memory Systems
## Episode Summary In this episode, we cover: - **Selective KV Cache Protection for Noise-Resilient LLM Inference on Analog Compute-In-Memory Systems** (arXiv)…
- Condor Currents
Windowed-MTP: Removing the Full-Context Draft-KV Tax at Million-Token Context
## Episode Summary In this episode, we cover: - **Windowed-MTP: Removing the Full-Context Draft-KV Tax at Million-Token Context** (arXiv) - **A Heterogeneous…
- Condor Currents
A Photonic-CXL Memory Appliance for Scalable KV Cache Management in LLM Inference
## Episode Summary In this episode, we cover: - **A Photonic-CXL Memory Appliance for Scalable KV Cache Management in LLM Inference** (arXiv) - **SPARC…
- Condor Currents
PIMID: A Full-System Simulator with Intricacy and Diversity for Processing-in-Memory
## Episode Summary In this episode, we cover: - **PIMID: A Full-System Simulator with Intricacy and Diversity for Processing-in-Memory** (arXiv) -…
- Condor Currents
Investigating reservoir computing for branch predictionin pipelined processors using emerging CMOS memristor devices
## Episode Summary In this episode, we cover: - **Investigating reservoir computing for branch predictionin pipelined processors using emerging CMOS memristor…
- Condor Currents
SpecBox: Speculative Sandbox Scheduling for Efficient LLM Agent Serving
## Episode Summary In this episode, we cover: - **SpecBox: Speculative Sandbox Scheduling for Efficient LLM Agent Serving** (arXiv) - **Reducing…
- Condor Currents
SpacemiT MUSE Pi Pro is an octa-core RISC-V SBC with Armbian support - LinuxGizmos.com
## Episode Summary In this episode, we cover: - **SpacemiT MUSE Pi Pro is an octa-core RISC-V SBC with Armbian support - LinuxGizmos.com** (google_riscv) -…
- Condor Currents
TileSight: A First-Principles Tile-Centric Analytical GPU Performance Model from Cores to Clusters
## Episode Summary In this episode, we cover: - **TileSight: A First-Principles Tile-Centric Analytical GPU Performance Model from Cores to Clusters** (arXiv)…
- Condor Currents
DGNA: Dissecting GPU NUMA Architecture through Microbenchmarking and Data Analysis
## Episode Summary In this episode, we cover: - **DGNA: Dissecting GPU NUMA Architecture through Microbenchmarking and Data Analysis** (arXiv) - **Isolation…
- Condor Currents
RISC-V: The Open-Source Revolution in CPU Architecture - Design And Reuse
## Episode Summary In this episode, we cover: - **RISC-V: The Open-Source Revolution in CPU Architecture - Design And Reuse** (google_arch) - **The RISC-V…
- Condor Currents
AMD reveals CPU architecture roadmap through 2028, following Zen 6 'Venice' launch — Zen 7 'Florence' to de... - Yahoo Tech
## Episode Summary In this episode, we cover: - **AMD reveals CPU architecture roadmap through 2028, following Zen 6 'Venice' launch — Zen 7 'Florence' to…
- Condor Currents
NextSilicon Productizes Arbel RISC-V Core Into 64-Core Enterprise Processor - Quantum Zeitgeist
## Episode Summary In this episode, we cover: - **NextSilicon Productizes Arbel RISC-V Core Into 64-Core Enterprise Processor - Quantum Zeitgeist**…
- Condor Currents
SEAM-V: A Hybrid-Decoupled RISC-V Vector Processor with Backend-Visible EP Context for Sustained Vector Throughput
## Episode Summary In this episode, we cover: - **SEAM-V: A Hybrid-Decoupled RISC-V Vector Processor with Backend-Visible EP Context for Sustained Vector…
- Condor Currents
DSTAR: Accelerating Diffusion Transformers via Spatial and Temporal Redundancy Reduction
## Episode Summary In this episode, we cover: - **DSTAR: Accelerating Diffusion Transformers via Spatial and Temporal Redundancy Reduction** (arXiv) -…
- Condor Currents
A Quantized Native Runtime for On-Device Semantic Audio Generation
## Episode Summary In this episode, we cover: - **A Quantized Native Runtime for On-Device Semantic Audio Generation** (arXiv) - **Pattern-Guided Design Space…
- Condor Currents
Smarter and Cheaper at Once: Byte-Exact KV-Cache Grafting Turns a Frozen Small Model into a Verified-Knowledge Flywheel
## Episode Summary In this episode, we cover: - **Smarter and Cheaper at Once: Byte-Exact KV-Cache Grafting Turns a Frozen Small Model into a…
- Condor Currents
CODA: Algorithm-Hardware Co-design for Edge Video Diffusion via NMP-Enabled Compute-Cache Operator Disaggregation
## Episode Summary In this episode, we cover: - **CODA: Algorithm-Hardware Co-design for Edge Video Diffusion via NMP-Enabled Compute-Cache Operator…
- Condor Currents
Full-Pipeline Inference Optimization for MiMo-V2.5 Series: Pushing Hybrid SWA Efficiency to the Limit
## Episode Summary In this episode, we cover: - **Full-Pipeline Inference Optimization for MiMo-V2.5 Series: Pushing Hybrid SWA Efficiency to the Limit**…
- Condor Currents
CLIP-3D: Closed-Loop Evaluation of Performance and Physical Constraints for 3D ICs
## Episode Summary In this episode, we cover: - **CLIP-3D: Closed-Loop Evaluation of Performance and Physical Constraints for 3D ICs** (arXiv) - **AtomFlow: An…
- Condor Currents
FlashAccel: Leveraging High-Bandwidth Flash for High-Throughput LLM Inference
## Episode Summary In this episode, we cover: - **FlashAccel: Leveraging High-Bandwidth Flash for High-Throughput LLM Inference** (arXiv) - **IRONSmith: A…
- Condor Currents
STEEL: Sparsity-Aware Fused Attention for Energy-Efficient Long-Sequence Inference on AMD's XDNA NPU
## Episode Summary In this episode, we cover: - **STEEL: Sparsity-Aware Fused Attention for Energy-Efficient Long-Sequence Inference on AMD's XDNA NPU**…
- Condor Currents
Embedded Blockchain Infrastructure Management (eBIM): A RISC-V-Empowered Hardware--Software Co-Design Framework Towards Trustworthy Blockchain
## Episode Summary In this episode, we cover: - **Embedded Blockchain Infrastructure Management (eBIM): A RISC-V-Empowered Hardware--Software Co-Design…
- Condor Currents
HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing
## Episode Summary In this episode, we cover: - **HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing** (arXiv) - **ATLAS…
- Condor Currents
Vectorizing Quantum Control: A RISC-V Vector Extension Architecture for Scalable Qubit Systems
## Episode Summary In this episode, we cover: - **Vectorizing Quantum Control: A RISC-V Vector Extension Architecture for Scalable Qubit Systems** (arXiv) -…
- Condor Currents
RISC-V: The Open-Source Revolution in CPU Architecture - Design And Reuse
## Episode Summary In this episode, we cover: - **RISC-V: The Open-Source Revolution in CPU Architecture - Design And Reuse** (google_arch) - **NextSilicon to…
- Condor Currents
TileLens: Efficiently Using Large-Granularity Memory Systems with Transparent Two-Dimensional Memory Layout
## Episode Summary In this episode, we cover: - **TileLens: Efficiently Using Large-Granularity Memory Systems with Transparent Two-Dimensional Memory Layout**…
- Condor Currents
A Reconfigurable and Representation-Adaptive ISA-Based Architecture for Efficient DNN Acceleration
## Episode Summary In this episode, we cover: - **A Reconfigurable and Representation-Adaptive ISA-Based Architecture for Efficient DNN Acceleration** (arXiv)…
- Condor Currents
A Multi-Dimensional, Per-Pass Empirical Study of the LLVM Optimization Pipeline
## Episode Summary In this episode, we cover: - **A Multi-Dimensional, Per-Pass Empirical Study of the LLVM Optimization Pipeline** (arXiv) - **Approximate…
- Condor Currents
The Fourth-Root Complexity of Data Movement
## Episode Summary In this episode, we cover: - **The Fourth-Root Complexity of Data Movement** (arXiv) - **Dynamic Ultrasound Beamforming Using Left-to-Right…
- Condor Currents
RISC-V Mentorship Taught Me the RISC-V ISA Is Far More Than a Reference Manual
## Episode Summary In this episode, we cover: - **RISC-V Mentorship Taught Me the RISC-V ISA Is Far More Than a Reference Manual** (riscv_news) - **RISC-V set…
- Condor Currents
3DLS: A 3D Logic-Stacked Architecture for Disaggregated LLM Serving
## Episode Summary In this episode, we cover: - **3DLS: A 3D Logic-Stacked Architecture for Disaggregated LLM Serving** (arXiv) - **KernelSight-LM: A…
- Condor Currents
High-Performance NTT Accelerators for PQC leveraging Unified Redundant Arithmetic and Fine-Tuned Microarchitecture
## Episode Summary In this episode, we cover: - **High-Performance NTT Accelerators for PQC leveraging Unified Redundant Arithmetic and Fine-Tuned…
- Condor Currents
TraceLab: Characterizing Coding Agent Workloads for LLM Serving
## Episode Summary In this episode, we cover: - **TraceLab: Characterizing Coding Agent Workloads for LLM Serving** (arXiv) - **Croc: Training the Next…
- Condor Currents
HBM Is Not All You Need: Efficient Disaggregated LLM Serving across Memory-heterogeneous Accelerators
## Episode Summary In this episode, we cover: - **HBM Is Not All You Need: Efficient Disaggregated LLM Serving across Memory-heterogeneous Accelerators**…
- Condor Currents
CrossPool: Efficient Multi-LLM Serving for Cold MoE Models through KV-Cache and Weight Disaggregation
## Episode Summary In this episode, we cover: - **CrossPool: Efficient Multi-LLM Serving for Cold MoE Models through KV-Cache and Weight Disaggregation**…
- Condor Currents
SOLAR: AI-Powered Speed-of-Light Performance Analysis
## Episode Summary In this episode, we cover: - **SOLAR: AI-Powered Speed-of-Light Performance Analysis** (arXiv) - **Accelerating Disaggregated RL for Visual…
- Condor Currents
The Serialized Bridge: Understanding and Recovering LLM Serving Performance under Blackwell GPU Confidential Computing
## Episode Summary In this episode, we cover: - **The Serialized Bridge: Understanding and Recovering LLM Serving Performance under Blackwell GPU Confidential…
- Condor Currents
CVA6-RT: an Open-Source Time-Predictable RV64 Processor for Mixed-Criticality Systems
## Episode Summary In this episode, we cover: - **CVA6-RT: an Open-Source Time-Predictable RV64 Processor for Mixed-Criticality Systems** (arXiv) - **Residual…
- Condor Currents
Cache-Resident LLM Inference in GB-Scale Last-Level Caches
## Episode Summary In this episode, we cover: - **Cache-Resident LLM Inference in GB-Scale Last-Level Caches** (arXiv) - **Energy-Efficient CNN Acceleration…
- Condor Currents
CrossPool: Efficient Multi-LLM Serving for Cold MoE Models through KV-Cache and Weight Disaggregation
## Episode Summary In this episode, we cover: - **CrossPool: Efficient Multi-LLM Serving for Cold MoE Models through KV-Cache and Weight Disaggregation**…
- Condor Currents
Non-Uniform L2 Cache Latency Across the Streaming Multiprocessors of an NVIDIA L40
## Episode Summary In this episode, we cover: - **Non-Uniform L2 Cache Latency Across the Streaming Multiprocessors of an NVIDIA L40** (arXiv) - **MOCAP…