TriMoE: Augmenting GPU with AMX-Enabled CPU and DIMM-NDP for High-Throughput MoE Inference via Offloading
To deploy large Mixture-of-Experts (MoE) models cost-effectively, offloading-based single-GPU heterogeneous inference is crucial. While GPU-CPU architectures that offload cold experts are constrained by host memory bandwidth, emerging GPU-NDP architectures utilize DIMM-NDP to offload non-hot experts. However, non-hot experts are not a homogeneous memory-bound group: a significant subset of warm experts exists is severely penalized by high GPU I/O latency yet can saturate NDP compute throughput, exposing a critical compute gap. We present TriMoE, a novel GPU-CPU-NDP architecture that fills this gap by synergistically leveraging AMX-enabled CPU to precisely map hot, warm, and cold experts o...
| Authors | Yudong Pan, Yintao He, Tianhua Han et al. |
|---|---|
| Primary category | cs.AR |
| Published | 1 March 2026 |
| Last updated at source | 1 March 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.