Sep 2026· IEEE International Conference on Application-Specific Systems, Architectures, and Processors· pp. 169-170· 0 citations· 4 references
TL;DR
A multi-cluster architecture that replaces inter-cluster communication with localized data replication within ScratchPad Memories (SPMs) is introduced, creating a critical architectural trade-off between the data replication overhead required by overlapping-footprint kernels and the routing delays of larger clusters.
Abstract
To mitigate interconnect scaling bottlenecks $\left(O\left(N^{2}\right)\right)$ and Non-Uniform Memory Access (NUMA) congestion in Programmable Multi-Core Accelerators (PMCAs), this paper introduces a multi-cluster architecture that replaces inter-cluster communication with localized data replication within ScratchPad Memories (SPMs). This design creates a critical architectural trade-off between the data replication overhead required by overlapping-footprint kernels (e.g., matrix multiplication) and the routing delays of larger clusters. We propose an architectural exploration methodology utilizing a Memory-Delay Squared Product $(m d^{2} p)$ cost function to analytically resolve this trade-off. Linear programming optimization reveals two optimal configurations based on the workload data ratio: a 1-PE cluster for element-wise kernels and an 8-PE cluster for overlapping-footprint kernels. Synthesized results for a 64-PE implementation demonstrate that our methodology yields substantial performance gains, achieving up to a $7.36 \times$ speedup over an ARM Cortex A53 and a $6.03 \times$ speedup over a state-of-the-art integrated programmable array (IPA) accelerator.
Modern systems-on-chip (SoCs) rely on heterogeneous accelerators for performance scaling. Memory access is a critical bottleneck, but the complexity of cache coherence and nuances of weak memory consistency models (which may vary from system to system) represent a significant designer burden to every load/store unit th...
Joseph Maheshe, G. Lemieux· IEEE International Conferenc...· 3 citations
CMD, an integrated CGRA framework featuring cluster-based distributed memory design with a co-designed compilation toolchain, is proposed, which includes a novel memory-aware mapper and a design space exploration mechanism that identifies the optimal memory architecture design for specific kernels.
Shang-Kun Li, Cheng Tan, Ze-Yu Li et al.· 0 citations
Modern GPUs are no longer physically symmetric. Die scaling leads to both manufacturing-driven floorsweeping and cache and memory partitioning. The former creates chip-specific compute topologies, while the latter causes non-uniform memory access. These asymmetries are substantial. Topology-oblivious compute unit alloc...
X. Fan, Jia-Hao Wang, Wei-Hao Cui et al.· 1 citation
SA-Scheduler is presented, a structure-derived bottleneck-aware scheduling framework for multitasking MCM-GPUs that determines chip placement without hardware modification or runtime bottleneck profiling, and provides a principled and scalable foundation for multitasking on future MCM-GPUs.
Tie-Jian Zhang, Guangda Zhang, Lu Wang et al.· ACM Transactions on Design A...· 0 citations
TileGEMM is proposed, a high-performance GEMM implementation on AMX that systematically enhances data reuse across the memory hierarchy, and achieves average speedups of 3.27 × and 1.96 × over AVX-512-based implementations TVM and MKL, respectively.
Kang-Kang Chen, Hua-You Su, Meng-Han Jia et al.· International Conference on...· 0 citations
Evaluation using cycle-accurate hardware-software co-simulation demonstrates that CHIPSMORE effectively sustains high throughput across varying model sizes, context lengths, and batch sizes while maintaining favorable power scaling.