Benchmark results edited for A1.

This commit is contained in:
Robin E.R. Davies
2026-04-21 11:18:08 -04:00
parent 3704d45f5c
commit 1d23503e3d
+28 -12
View File
@@ -1,32 +1,50 @@
Benchmark Results:
The benchmark is to run TooB NAM with an A1 model to generate 100s of audio. Lower times are better.
The benchmark is to run TooB Neural Amp Modeler with an A1 model to generate 100s of audio. Lower times are better.
TooB NAM has a compile-time switch that uses either the NeuralAudio NAM implementation or
the Neural Modeler Core NAM implementation.
Results
amd64 (AMD Ryzen 7 7735HS, Ubuntu 24.04 amd64)
Neural Amp Modeler Core: 7.965s
NeuralAudio x64: 5.416s
Compiler options /Ofast, MMX, SSE, SSE2, no AVX.
aarch64 (Raspberry PI 4, ARM Cortex A72, Raspberry PI OS Bookworm)
Neural Amp Modeler Core: 92.677s
Neural Amp Modeler Core: 94.296s
NeuralAudio: 31.589s
Compiler options: "-Ofast" "-mcpu=cortex-a72" "-march=armv8-a" "-mtune=cortex-a72" "-DEIGEN_DEFAULT_L1_CACHE_SIZE=32768"
aarch64 (Raspberry PI 5, ARM Cortex A76, Rasperry PI OS Trixie)
Neural Amp Modeler Core: 16.212s
NeuralAudio: 10.824s
Neural Amp Modeler Core: 16.084s
NeuralAudio: 10.766s
Compiler options: "-Ofast" "-mcpu=cortex-a76" "-march=armv8.2-a" "-mtune=cortex-a76" "-DEIGEN_DEFAULT_L1_CACHE_SIZE=65536"apt
amd64 Native (AMD Ryzen 7 7735HS, Ubuntu 24.04 amd64)
Neural Amp Modeler Core: 4.716s
NeuralAudio x64: 3.952s
Compiler options /Ofast, MMX, SSE, SSE2, no AVX.
amd64 AVX+ (AMD Ryzen 7 7735HS, Ubuntu 24.04 amd64)
Neural Amp Modeler Core: 4.97s
NeuralAudio x64: 4.032s
Compiler options /Ofast, "-march=sandybridge" "-mtune=sandybridge"
amd64 SSE2 (AMD Ryzen 7 7735HS, Ubuntu 24.04 amd64)
Neural Amp Modeler Core: 7.748s
NeuralAudio x64: 5.983s
Compiler options /Ofast (defaults to SSE2, no AVX)
The Model:
Package: Fender Deluxe Reverb %2765 Reissue %7c Clean %7c SM57 %2b Royer R-121 %2b Room
@@ -36,8 +54,6 @@ Downloaded from Tone3000.
The model file is attached.
Compilation flags: