Hybrid CPU GPU dispatch reduces AI radio processing delays
Workload-Aware Monitoring for AI-RAN
Networking and Internet Architecture
Summary
Processing radio signals and AI tasks on shared computers can cause delays because CPUs and GPUs work differently. The authors found a way to watch how these tasks use the processors and match similar types together before measuring. Their tests showed that using the CPU for small radio data pieces and the GPU for larger ones made the system faster than using the GPU alone. This approach helps make better use of computer parts for combined radio and AI workloads.
What this means in practice
- •For mobile network operators: Optimize network equipment by using CPU and GPU dispatch strategies to reduce radio data processing delays under mixed workloads.
- •For cloud infrastructure teams: Improve resource allocation in shared CPU/GPU cloud servers running AI and radio applications by matching workload sizes to processors.
Authors
Babak Azkaei, Georgios Exarchakos, Kishor Chandra Joshi
Abstract
AI-RAN platforms are expected to share CPU and GPU resources between radio workloads and AI services. Aggregate telemetry reports operating conditions, but timing changes may reflect different workloads, processors, or measurement boundaries. We propose workload-aware monitoring that associates radio-work descriptors with processing paths, timing boundaries, and operating context before comparing like with like. In 16 over-the-air downlink experiments, hybrid dispatch (CPU for transport blocks (TBs) below 4,096 bytes and GPU otherwise) has 29.66% and 22.93% lower mean host-callback durations than direct GPU dispatch, without and with a synthetic GPU co-runner, respectively. Grouping locates the separation in small TBs, while large-TB callback-duration ranges overlap. These observations identify a candidate for complementary CPU/GPU use.