Scaling inference for large language models is increasingly constrained by limited GPU memory, primarily due to the expanding intermediate states (KV caches) required for long-context generation and multi-user workloads. Once the KV cache exceeds the capacity of high-bandwidth memory, it must be offloaded to host memory and reloaded on demand, a workflow severely bottlenecked by the CPU–GPU interconnect, typically PCIe. Existing approaches exploiting offload KV caches to CPU memory and selectively reload partial segments for attention computation often underutilize CPU compute resources and suffer from accuracy degradation. We present Beyond, a drop-in runtime that integrates a smart offloading scheme to selectively identify and retain salient KV entries across continuous decoding sessions, together with a hybrid CPU–GPU attention mechanism for scalable inference. Beyond executes dense attention over recent KV entries stored in GPU memory while performing parallel, per-head sparse attention on salient contextual KV entries residing in CPU memory. The outputs are fused efficiently through a log-sum-exp scheme. During the bandwidth-constrained decoding phase, oversized KV caches are processed cooperatively by the aggregated CPU and GPU memory bandwidth, with only minimal PCIe data movement. Experiments across diverse models and workloads demonstrate that Beyond improves scalability, supports longer sequences and larger batch sizes, and outperforms existing sparse attention baselines in both efficiency and accuracy—all on commodity GPU hardware.
Weishu Deng, Yujie Yang, Peiran Du et al.· IEEE International Symposium...· 1 citation
The rapid adoption of large language models (LLMs) has increased the need for efficient multi-tenant inference systems that maximize GPU utilization. However, existing frameworks struggle to scale due to the high memory demands of model weights and key-value (KV) caches. We present DynamoServe, a multi-tenant LLM serving framework that addresses these challenges through three key innovations: (1) leveraging stranded GPU memory to offload model weights and KV caches, (2) mitigating resource fragmentation in multi-workload environments, and (3) improving memory locality through coordinated data placement and demand-driven weight migration across GPUs. Together, these techniques enable high-throughput, low-latency inference. Experiments on state-of-the-art models show that DynamoServe significantly improves memory efficiency without sacrificing latency.
Diman Zad Tootaghaj, Khaled Diab, Bob Lantz et al.· 0 citations
Collective communication operations in AI and HPC workloads generate heavy network traffic. Offloading these operations to network switches reduces latency, but performing arithmetic and replication at line rate is difficult, especially as port counts and link speeds grow. Existing in-network approaches rely on accumulation buffers that not only limit throughput but also require complex state management to handle stragglers and congestion. We present CCSwitch, a modular switching fabric built from 4×4 non-blocking Collective Engines (CEs). Each CE combines spatial and temporal parallelism to perform reductions without accumulation buffers. CEs compose into k-ary n-tree topologies, scaling to 32- and 256-port switches while preserving non-blocking throughput. Source routing and flit-level synchronization keep per-switch state minimal. Our FPGA implementation shows that CCSwitch's quaternary-tree reduction fabric uses up to 23% fewer LUTs and 12–30% fewer flip-flops than a comparable Clos-based design at equal throughput. Enabling the full feature set—source routing, replication, and time-multiplexed VCs—uses 1.4–1.8× more LUTs than the circuit-switched baseline, well below the 3–5× overhead typical of packet-switched NoC routers, while supporting concurrent collectives on shared links.
Sumukh Pinge, Hardik Soni, Bob Lantz et al.· Proceedings of the ACM SIGCO...· 0 citations