llama cpp grace

ARMv8 Neoverse-V2 testing with a Pegatron JIMBO P4352 (00022432 BIOS) and ASPEED on Ubuntu 24.04 via the Phoronix Test Suite.

HTML result view exported from: https://openbenchmarking.org/result/2411235-NE-LLAMACPPG43&grt&rdt.

llama cpp graceProcessorMotherboardMemoryDiskGraphicsNetworkOSKernelCompilerFile-SystemScreen ResolutionabcdARMv8 Neoverse-V2 @ 3.47GHz (72 Cores)Pegatron JIMBO P4352 (00022432 BIOS)1 x 480GB LPDDR5-6400MT/s NVIDIA 699-2G530-0236-RC11000GB CT1000T700SSD3ASPEED2 x Intel X550Ubuntu 24.046.8.0-49-generic-64k (aarch64)GCC 13.2.0 + Clang 18.1.3 + CUDA 11.8ext41920x1200OpenBenchmarking.orgKernel Details- Transparent Huge Pages: madviseCompiler Details- --build=aarch64-linux-gnu --disable-libquadmath --disable-libquadmath-support --disable-werror --enable-checking=release --enable-clocale=gnu --enable-default-pie --enable-fix-cortex-a53-843419 --enable-gnu-unique-object --enable-languages=c,ada,c++,go,d,fortran,objc,obj-c++,m2 --enable-libphobos-checking=release --enable-libstdcxx-backtrace --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-multiarch --enable-nls --enable-objc-gc=auto --enable-offload-defaulted --enable-offload-targets=nvptx-none=/build/gcc-13-dIwDw0/gcc-13-13.2.0/debian/tmp-nvptx/usr --enable-plugin --enable-shared --enable-threads=posix --host=aarch64-linux-gnu --program-prefix=aarch64-linux-gnu- --target=aarch64-linux-gnu --with-default-libstdcxx-abi=new --with-gcc-major-version-only --with-target-system-zlib=auto --without-cuda-driver -v Processor Details- Scaling Governor: cppc_cpufreq ondemand (Boost: Disabled)Security Details- gather_data_sampling: Not affected + itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + reg_file_data_sampling: Not affected + retbleed: Not affected + spec_rstack_overflow: Not affected + spec_store_bypass: Mitigation of SSB disabled via prctl + spectre_v1: Mitigation of __user pointer sanitization + spectre_v2: Not affected + srbds: Not affected + tsx_async_abort: Not affected

llama cpp gracellama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Text Generation 128llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 512llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 1024llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 2048llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Text Generation 128llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 512llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 1024llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 2048llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Text Generation 128llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Prompt Processing 512llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Prompt Processing 1024llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Prompt Processing 2048abcd20.07121.74118.77105.7121.48122.12119.72107.0050.75123.29132.61131.8120.56121.76118.88105.5621.78122.28119.92106.6550.87123.69132.75134.4220.70121.85119.02105.7520.06122.54119.62106.9051.00123.66128.63130.5218.24120.71118.98106.119.48121.45119.36106.4949.91131.19131.74133.64OpenBenchmarking.org

Llama.cpp

Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128abcd510152025SE +/- 0.21, N = 15SE +/- 0.25, N = 15SE +/- 0.25, N = 1520.0720.5620.7018.241. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 512

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 512abcd306090120150SE +/- 0.11, N = 3SE +/- 0.24, N = 3SE +/- 0.05, N = 3121.74121.76121.85120.711. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 1024

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 1024abcd306090120150SE +/- 0.20, N = 3SE +/- 0.19, N = 3SE +/- 0.10, N = 3118.77118.88119.02118.981. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 2048

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 2048abcd20406080100SE +/- 0.13, N = 3SE +/- 0.03, N = 3SE +/- 0.37, N = 3105.71105.56105.75106.101. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Text Generation 128

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Text Generation 128abcd510152025SE +/- 0.31, N = 15SE +/- 0.32, N = 15SE +/- 0.24, N = 321.4821.7820.0619.481. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 512

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 512abcd306090120150SE +/- 0.12, N = 3SE +/- 0.05, N = 3SE +/- 0.22, N = 3122.12122.28122.54121.451. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 1024

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 1024abcd306090120150SE +/- 0.01, N = 3SE +/- 0.12, N = 3SE +/- 0.28, N = 3119.72119.92119.62119.361. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 2048

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 2048abcd20406080100SE +/- 0.16, N = 3SE +/- 0.15, N = 3SE +/- 0.19, N = 3107.00106.65106.90106.491. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Text Generation 128

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Text Generation 128abcd1224364860SE +/- 0.59, N = 3SE +/- 0.71, N = 3SE +/- 0.45, N = 1550.7550.8751.0049.911. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 512

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 512abcd306090120150SE +/- 1.11, N = 15SE +/- 0.55, N = 3SE +/- 1.07, N = 15123.29123.69123.66131.191. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 1024

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 1024abcd306090120150SE +/- 1.59, N = 4SE +/- 1.65, N = 3SE +/- 0.91, N = 3132.61132.75128.63131.741. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Llama.cpp

Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 2048

OpenBenchmarking.orgTokens Per Second, More Is BetterLlama.cpp b4154Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 2048abcd306090120150SE +/- 1.44, N = 4SE +/- 0.94, N = 3SE +/- 1.22, N = 3131.81134.42130.52133.641. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -mcpu=native -fopenmp -lopenblas

Phoronix Test Suite v10.8.5