Formula

Residual Host Bandwidth Formulation for CPU Expert Execution

In an edge system where both direct CPU expert evaluation and direct memory access (DMA) transfers across PCIe draw from the same host DRAM subsystem, a saturated PCIe transfer leaves a constrained amount of memory bandwidth for CPU processing. If BHB_H denotes the measured host-side expert-processing bandwidth and BPB_P represents the pinned expert-transfer bandwidth across PCIe, the residual host bandwidth BRB_R available for concurrent in-place CPU expert computation is given by:

BR=max(BHBP,0)B_R = \max(B_H - B_P, 0)

When BH>BPB_H > B_P, the CPU can concurrently execute residual missed experts using this leftover host memory bandwidth without contending with or throttling the saturated PCIe bus.

0

1

Updated 2026-09-07

Tags

Prep Sessions

Edge-Native Mixture-of-Experts Serving with FreeToken @ University of Michigan - Ann Arbor

Ch.3 Adaptive Runtime Policies and Device Execution - Edge-Native Mixture-of-Experts Serving with FreeToken @ University of Michigan - Ann Arbor

Bandwidth-Adaptive Decode and the q* Execution Policy - Edge-Native Mixture-of-Experts Serving with FreeToken @ University of Michigan - Ann Arbor

Related