Is this page helpful?

Timing Cache Migration#

ITimingCache and related builder APIs persist from TensorRT but does not affect engines built by TensorRT-RTX. Using them forces a GPU query during ahead-of-time compilation, which prevents CPU-only AOT builds. These APIs were deprecated in TensorRT-RTX 1.2 and will be removed in a future release.

Workflow diagram showing that ITimingCache from ported TensorRT code has no effect in TensorRT-RTX, forces GPU during AOT build, and should be replaced with runtime cache for JIT kernel persistence
Table 6 Timing cache APIs in TensorRT-RTX#

API / behavior

Effect in TensorRT-RTX

Recommendation

ITimingCache at build time

No impact on the serialized engine; requires a GPU during AOT

Do not use; remove from ported TensorRT code

Runtime cache

Stores JIT-compiled kernels for faster startup

Use runtime cache instead for cross-run performance

For porting notes, refer to the Porting Guide for TensorRT Applications. For runtime cache setup, refer to Working with Runtime Cache.