Hardware Mechanisms to Dynamically Throttle AI Performance
2026-07-20 • Hardware Architecture
Hardware ArchitectureMachine Learning
AI summaryⓘ
The authors explain that current AI safety methods, which limit behavior through software, can be bypassed by very smart models. To provide better control, they propose hardware-level controls called microarchitecture knobs that adjust GPU resources like cache size and memory access rates during runtime. These knobs help slow down AI performance dynamically without much extra hardware and do so quickly and efficiently. Combining multiple knobs can create stronger performance limits, offering a flexible way to enforce AI safety at the hardware level.
AI safetymicroarchitectureGPU memory subsystemcache way maskingrate limitinglatency insertionbank arbitrationhardware throttlingperformance sensitivity
Authors
Haiyue Ma, Lauren Malek, Joseph Forzani, David Wentzlaff
Abstract
As more capable AI models are increasingly integrated into critical computer systems, the lack of control over AI intent motivates safety mechanisms. Existing software safeguards impose only behavioral constraints that can potentially be bypassed by sufficiently intelligent models. While hardware-level safety enforcement has been recognized as an essential last line of defense, few mechanisms have been proposed beyond policy regulations on unauthorized accesses or coarse full-chip shutdown. What is missing is a fine-grained, dynamic intervention mechanism at the architecture level. In this paper, we introduce a set of microarchitecture knobs which dynamically control the available hardware resources to limit AI performance at runtime. We evaluate candidate knobs spanning the GPU memory subsystem, across capacity, bandwidth, latency and frequency dimensions, and narrow down to four strong candidates: L2 size, L2 latency, L2 bandwidth, and shared memory port access rate. To minimize new logic and extra design cost, we build all four mechanisms from well-established microarchitectural primitives: cache way masking, credit-based rate limiting, latency insertion, and bank arbitration. We show that these knobs achieve high performance sensitivity (up to 80% performance cut at 1/8 resource availability), negligible implementation cost (<~10K flip flops), fast stabilization after dynamic throttling (5-80K cycles), and minimal collateral impact on the rest of the chip. Further, multi-knob analysis reveals combinations of knobs that amplify the performance degradation beyond the effect of each knob individually, which enables a broader range of performance targets.