llama cpp update AMD Ryzen Threadripper 7980X 64-Cores testing with a System76 Thelio Major (FA Z5 BIOS) and AMD Radeon RX 6700 XT 12GB on Ubuntu 24.04 via the Phoronix Test Suite.
HTML result view exported from: https://openbenchmarking.org/result/2411230-PTS-LLAMACPP41&grr&rdt .
llama cpp update Processor Motherboard Chipset Memory Disk Graphics Audio Monitor Network OS Kernel Desktop Display Server OpenGL Compiler File-System Screen Resolution a b c AMD Ryzen Threadripper 7980X 64-Cores @ 7.79GHz (64 Cores / 128 Threads) System76 Thelio Major (FA Z5 BIOS) AMD Device 14a4 4 x 32GB DDR5-4800MT/s Micron MTC20F1045S1RC48BA2 1000GB CT1000T700SSD5 AMD Radeon RX 6700 XT 12GB AMD Device 14cc DELL P2415Q Aquantia AQC113C NBase-T/IEEE + Realtek RTL8125 2.5GbE + Intel Wi-Fi 6E Ubuntu 24.04 6.8.0-48-generic (x86_64) GNOME Shell 46.0 X Server + Wayland 4.6 Mesa 24.0.9-0ubuntu0.2 (LLVM 17.0.6 DRM 3.57) GCC 13.2.0 ext4 1920x1200 OpenBenchmarking.org Kernel Details - Transparent Huge Pages: madvise Compiler Details - --build=x86_64-linux-gnu --disable-vtable-verify --disable-werror --enable-cet --enable-checking=release --enable-clocale=gnu --enable-default-pie --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-multilib --enable-nls --enable-objc-gc=auto --enable-offload-defaulted --enable-offload-targets=nvptx-none=/build/gcc-13-uJ7kn6/gcc-13-13.2.0/debian/tmp-nvptx/usr,amdgcn-amdhsa=/build/gcc-13-uJ7kn6/gcc-13-13.2.0/debian/tmp-gcn/usr --enable-plugin --enable-shared --enable-threads=posix --host=x86_64-linux-gnu --program-prefix=x86_64-linux-gnu- --target=x86_64-linux-gnu --with-abi=m64 --with-arch-32=i686 --with-default-libstdcxx-abi=new --with-gcc-major-version-only --with-multilib-list=m32,m64,mx32 --with-target-system-zlib=auto --with-tune=generic --without-cuda-driver -v Processor Details - Scaling Governor: amd-pstate-epp powersave (EPP: balance_performance) - CPU Microcode: 0xa108105 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: Mitigation of Safe RET + spec_store_bypass: Mitigation of SSB disabled via prctl + spectre_v1: Mitigation of usercopy/swapgs barriers and __user pointer sanitization + spectre_v2: Mitigation of Enhanced / Automatic IBRS; IBPB: conditional; STIBP: always-on; RSB filling; PBRSB-eIBRS: Not affected; BHI: Not affected + srbds: Not affected + tsx_async_abort: Not affected
llama cpp update llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 2048 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 2048 llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 1024 llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 512 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 512 llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Prompt Processing 2048 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 1024 llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Text Generation 128 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Text Generation 128 llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Prompt Processing 512 llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Prompt Processing 1024 llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-Q8_0 - Text Generation 128 a b c 154.68 149.47 123.08 102.05 103.99 418.96 126.34 15.54 16.46 249.15 328.55 76.34 153.21 149.65 124.40 103.50 105.14 405.82 129.31 15.50 16.52 250.07 329.89 76.53 153.51 154.26 122.97 104.43 103.32 406.84 126.69 15.60 16.45 250.38 341.69 76.56 OpenBenchmarking.org
Llama.cpp Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 2048 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 2048 a b c 30 60 90 120 150 SE +/- 1.85, N = 15 SE +/- 1.13, N = 15 SE +/- 1.32, N = 15 154.68 153.21 153.51 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 2048 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 2048 a b c 30 60 90 120 150 SE +/- 0.58, N = 3 SE +/- 0.97, N = 3 SE +/- 1.30, N = 15 149.47 149.65 154.26 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 1024 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 1024 a b c 30 60 90 120 150 SE +/- 0.81, N = 3 SE +/- 0.96, N = 15 SE +/- 0.46, N = 3 123.08 124.40 122.97 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 512 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 512 a b c 20 40 60 80 100 SE +/- 1.42, N = 15 SE +/- 1.15, N = 3 SE +/- 1.20, N = 15 102.05 103.50 104.43 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 512 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 512 a b c 20 40 60 80 100 SE +/- 1.12, N = 15 SE +/- 1.28, N = 3 SE +/- 1.25, N = 13 103.99 105.14 103.32 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 2048 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 2048 a b c 90 180 270 360 450 SE +/- 3.92, N = 3 SE +/- 3.66, N = 3 SE +/- 4.49, N = 15 418.96 405.82 406.84 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 1024 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 1024 a b c 30 60 90 120 150 SE +/- 0.27, N = 3 SE +/- 0.30, N = 3 SE +/- 1.65, N = 3 126.34 129.31 126.69 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128 a b c 4 8 12 16 20 SE +/- 0.05, N = 3 SE +/- 0.04, N = 3 SE +/- 0.00, N = 3 15.54 15.50 15.60 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Text Generation 128 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Text Generation 128 a b c 4 8 12 16 20 SE +/- 0.02, N = 3 SE +/- 0.00, N = 3 SE +/- 0.01, N = 3 16.46 16.52 16.45 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 512 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 512 a b c 50 100 150 200 250 SE +/- 4.04, N = 12 SE +/- 4.43, N = 12 SE +/- 2.38, N = 3 249.15 250.07 250.38 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 1024 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 1024 a b c 70 140 210 280 350 SE +/- 3.24, N = 3 SE +/- 2.46, N = 3 SE +/- 2.74, N = 3 328.55 329.89 341.69 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Llama.cpp Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Text Generation 128 OpenBenchmarking.org Tokens Per Second, More Is Better Llama.cpp b4154 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Text Generation 128 a b c 20 40 60 80 100 SE +/- 0.37, N = 3 SE +/- 0.45, N = 3 SE +/- 0.34, N = 3 76.34 76.53 76.56 1. (CXX) g++ options: -std=c++11 -fPIC -O3 -pthread -fopenmp -march=native -mtune=native -lopenblas
Phoronix Test Suite v10.8.5