Intel Xeon Max Enjoying Some Performance Gains With Linux 6.6

Written by Michael Larabel in Software on 12 October 2023 at 01:31 PM EDT. Page 2 of 3. 1 Comment.
High Performance Conjugate Gradient benchmark with settings of X Y Z: 144 144 144, RT: 60. v6.6 6 Oct was the fastest.
NAS Parallel Benchmarks benchmark with settings of Test / Class: BT.C. v6.6 6 Oct was the fastest.
Rodinia benchmark with settings of Test: OpenMP CFD Solver. v6.6 6 Oct was the fastest.
OpenFOAM benchmark with settings of Input: drivaerFastback, Small Mesh Size, Mesh Time. v6.6 6 Oct was the fastest.
PostgreSQL benchmark with settings of Scaling Factor: 100, Clients: 800, Mode: Read Write. v6.6 6 Oct was the fastest.

Across a number of different workloads, Linux 6.6 on this Intel Xeon Max server tended to deliver some small performance improvements over Linux 6.5 stable.

OpenRadioss benchmark with settings of Model: Cell Phone Drop Test. v6.6 6 Oct was the fastest.
nekRS benchmark with settings of Input: TurboPipe Periodic. v6.6 6 Oct was the fastest.
7-Zip Compression benchmark with settings of Test: Decompression Rating. v6.6 6 Oct was the fastest.
SVT-AV1 benchmark with settings of Encoder Mode: Preset 4, Input: Bosphorus 1080p. v6.6 6 Oct was the fastest.

The performance uplifts didn't tend to be too significant but rather consistently delivering incremental gains across a wide-range of tests.

Apache IoTDB benchmark with settings of Device Count: 200, Batch Size Per Write: 100, Sensor Count: 500, Client Number: 100. v6.5 was the fastest.
Apache IoTDB benchmark with settings of Device Count: 500, Batch Size Per Write: 100, Sensor Count: 200, Client Number: 400. v6.5 was the fastest.
Apache IoTDB benchmark with settings of Device Count: 500, Batch Size Per Write: 100, Sensor Count: 500, Client Number: 400. v6.5 was the fastest.
Apache IoTDB benchmark with settings of Device Count: 500, Batch Size Per Write: 100, Sensor Count: 500, Client Number: 400. v6.5 was the fastest.
Rodinia benchmark with settings of Test: OpenMP HotSpot3D. v6.5 was the fastest.
OpenFOAM benchmark with settings of Input: drivaerFastback, Medium Mesh Size, Execution Time. v6.5 was the fastest.
PostgreSQL benchmark with settings of Scaling Factor: 1000, Clients: 1000, Mode: Read Only. v6.5 was the fastest.
PostgreSQL benchmark with settings of Scaling Factor: 1000, Clients: 800, Mode: Read Write. v6.5 was the fastest.
PostgreSQL benchmark with settings of Scaling Factor: 1000, Clients: 800, Mode: Read Write, Average Latency. v6.5 was the fastest.
Apache Hadoop benchmark with settings of Operation: Create, Threads: 50, Files: 1000000. v6.5 was the fastest.

There were also a few cases where Linux 6.6 was slightly slower than Linux 6.5 on this Sapphire Rapids server from Supermicro. It seemed to lean towards the database workloads seeing slightly lower performance on Linux 6.6 for this dual socket Intel Xeon Max server.

Stress-NG benchmark with settings of Test: NUMA. v6.5 was the fastest.

The NUMA performance in particular seems to regress with Linux 6.6 as measured by Stress-NG.


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