Cost-Efficient Multimodal LLM Inference via Cross-Tier GPU Heterogeneity
This work addresses the high cost and inefficiency of serving multimodal LLMs for AI service providers by enabling deployment on heterogeneous hardware.
The paper tackles the problem of inefficient multimodal LLM inference by identifying that vision encoding and language generation have opposing hardware demands, and shows that partitioning at the modality boundary minimizes cross-device transfer, reducing it from GB-scale to MB-scale. This enables cost-efficient heterogeneous serving, achieving up to 54% higher throughput and 37% better cost efficiency in experiments.
Multimodal large language model (MLLM) inference splits into two phases with opposing hardware demands: vision encoding is compute-bound, while language generation is memory-bandwidth-bound. We show that under standard transformer KV caching, the modality boundary (between vision encoder and language model) minimizes cross-device transfer among all partition points that preserve standard stage-based execution. Partitioning here reduces transfer complexity from $O(L * s_ctx)$ bytes (GB-scale KV caches under stage-level disaggregation) to $O(N_v * d)$ bytes (MB-scale embeddings), an O(L) reduction where L is the transformer depth. The result holds across attention mechanisms (MHA/GQA), dynamic vision resolutions, and model scales, and the advantage grows as models deepen. A direct implication is that existing stage-level disaggregation systems are constrained to high-bandwidth interconnects (e.g., NVLink), whereas modality-level disaggregation enables cross-tier heterogeneous serving over commodity PCIe. A closed-form cost model shows that heterogeneous deployment is cost-optimal under phase-separable workloads (predicts 31.4% savings; observed 40.6%). We build HeteroServe, a phase-aware runtime with modality-level partitioning and cross-tier scheduling, and evaluate it on LLaVA-1.5-7B and Qwen2.5-VL against vLLM v0.3.0. On identical 4xA100 hardware, engine optimizations raise throughput by up to 54%. Under a fixed budget, a heterogeneous cluster (\$38k) improves Tokens/\$ by 37% over a homogeneous baseline (\$64k) without degrading latency.