Benchmarking The Performance Impact To AMD Inception Mitigations

Written by Michael Larabel in Software on 15 August 2023 at 10:30 AM EDT. Page 5 of 8. 35 Comments.

Neural Magic DeepSparse benchmark with settings of Model: NLP Document Classification, oBERT base uncased on IMDB, Scenario: Asynchronous Multi-Stream. safe RET no microcode was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: ResNet-50, Baseline, Scenario: Asynchronous Multi-Stream. off was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: BERT-Large, NLP Question Answering, Scenario: Asynchronous Multi-Stream. safe RET no microcode was the fastest.

AI performance with software like Neural Magic DeepSparse wasn't impaired by the new CPU security mitigations.

Redis 7.0.12 + memtier_benchmark benchmark with settings of Protocol: Redis, Clients: 100, Set To Get Ratio: 1:10. off was the fastest.
Blender benchmark with settings of Blend File: BMW27, Compute: CPU-Only. off was the fastest.
OpenVINO benchmark with settings of Model: Person Detection FP16, Device: CPU. off was the fastest.
OpenVINO benchmark with settings of Model: Weld Porosity Detection FP16, Device: CPU. safe RET no microcode was the fastest.
RocksDB benchmark with settings of Test: Read Random Write Random. off was the fastest.
RocksDB benchmark with settings of Test: Update Random. off was the fastest.
Apache Cassandra benchmark with settings of Test: Writes. off was the fastest.
nginx benchmark with settings of Connections: 1000. off was the fastest.

It largely comes down to what workloads you are actively engaged in for whether you may be impacted by any performance slowdowns as a result of the new AMD Inception mitigations. When it comes to EPYC server workloads, the areas where there was additional overhead from the default "safe RET" mitigation modes was involving I/O like the code compilation speed and database workloads. For HPC, AI, CPU-based rendering, and other areas the performance was rather unchanged. The IBPB mitigation approach also showed some performance penalties but again that is an optional alternative mode to using safe RET.


Related Articles
Intel Core Ultra "Meteor Lake" Performance Improves With Linux 6.9
GCC 14 vs. LLVM Clang 18 Compiler Performance On Fedora 40
Intel Xeon Max Sees Some Performance Gains For OpenVINO & ONNX With Linux 6.9
Linux 6.9 Drives AMD 4th Gen EPYC Performance Even Higher For Some Workloads
Intel Core Ultra "Meteor Lake" Yields Faster Performance With Linux 6.9
Google Cloud's C3D Instances Provide Strong Performance Value For PingCAP's TiDB