Linux 6.19 Kernel Benchmarks With X86_NATIVE_CPU Optimization

Added to the Linux kernel earlier this year was the new X86_NATIVE_CPU Kconfig option to enable compiler optimizations for the local/native CPU in use when building the Linux kernel. In effect about ensuring that the “-march=native” compiler flag is set for the kernel build for optimizing the Linux kernel build for your processor being used. Back with Linux 6.16 I ran some benchmarks of the Linux kernel build with X86_NATIVE_CPU to gauge the impact. Now with the current Linux 6.19 kernel and some different hardware, here are some additional on/off benchmarks for evaluating the impact of the Linux kernel build with X86_NATIVE_CPU.

For this quick round of benchmarking was Linux 6.19 Git built using the GCC 15.2 kernel on Ubuntu 26.04 development.

This testing was on an AMD Ryzen Threadripper PRO 9995WX 96-core workstation.

Linux 6.19 X86_NATIVE_CPU Built Kernel Benchmarks

The same kernel configuration/build was carried out besides one with the X86_NATIVE_CPU option enabled and the other without.

LevelDB benchmark with settings of Benchmark: Random Read. X86_NATIVE_CPU Build was the fastest.

FS-Mark benchmark with settings of Test: 4000 Files, 32 Sub Dirs, 1MB Size. X86_NATIVE_CPU Build was the fastest.

Flexible IO Tester benchmark with settings of Type: Random Read, Engine: IO_uring, Direct: No, Block Size: 4KB, Job Count: 32, Disk Target: Default Test Directory. X86_NATIVE_CPU Build was the fastest.

In some of the I/O benchmarks there were slight performance benefits from the X86_NATIVE_CPU kernel.

Stress-NG benchmark with settings of Test: Socket Activity. X86_NATIVE_CPU Build was the fastest.

And in synthetic kernel micro-benchmarks.

Memcached benchmark with settings of Set To Get Ratio: 1:100. X86_NATIVE_CPU Build was the fastest.

Cryptsetup benchmark with settings of Twofish-XTS 256b Encryption. X86_NATIVE_CPU Build was the fastest.

Cryptsetup benchmark with settings of Twofish-XTS 512b Decryption. X86_NATIVE_CPU Build was the fastest.

DaCapo Benchmark benchmark with settings of Java Test: Eclipse. Stock Kernel was the fastest.

Memcached benchmark with settings of Set To Get Ratio: 1:5. X86_NATIVE_CPU Build was the fastest.

PostgreSQL benchmark with settings of Scaling Factor: 100, Clients: 1000, Mode: Read Only. X86_NATIVE_CPU Build was the fastest.

ClickHouse benchmark with settings of 100M Rows Hits Dataset, First Run / Cold Cache. Stock Kernel was the fastest.

RocksDB benchmark with settings of Test: Read While Writing. X86_NATIVE_CPU Build was the fastest.

But among the real-world workloads there was minimal benefit overall for this GCC 15 + Linux 6.19 X86_NATIVE_CPU kernel on the AMD Ryzen Threadripper PRO 9995WX Zen 5 workstation.

Apache Cassandra benchmark with settings of Test: Writes. Stock Kernel was the fastest.

Pogocache benchmark with settings of Set To Get Ratio: 1:10. X86_NATIVE_CPU Build was the fastest.

nginx benchmark with settings of Connections: 500. X86_NATIVE_CPU Build was the fastest.

Just a quick article out of this round as after 100+ benchmarks, only in a handful of cases were there any improvements worth noting and just within those synthetic I/O and kernel micro-benchmark tests.

Source link

Leave a Reply Cancel reply