9654 llama ncnn Tests for a future article. 2 x AMD EPYC 9654 96-Core testing with a AMD Titanite_4G (RTI1007B BIOS) and ASPEED on Ubuntu 24.10 via the Phoronix Test Suite.
HTML result view exported from: https://openbenchmarking.org/result/2412296-NE-9654LLAMA11&gru .
9654 llama ncnn Processor Motherboard Chipset Memory Disk Graphics Network OS Kernel Desktop Display Server Compiler File-System Screen Resolution a b 2 x AMD EPYC 9654 96-Core @ 2.40GHz (192 Cores / 384 Threads) AMD Titanite_4G (RTI1007B BIOS) AMD Device 14a4 1520GB 3201GB Micron_7450_MTFDKCB3T2TFS + 257GB Flash Drive ASPEED Broadcom NetXtreme BCM5720 PCIe Ubuntu 24.10 6.11.0-13-generic (x86_64) GNOME Shell 47.0 X Server GCC 14.2.0 ext4 1920x1200 OpenBenchmarking.org Kernel Details - Transparent Huge Pages: madvise Compiler Details - --build=x86_64-linux-gnu --disable-vtable-verify --disable-werror --enable-bootstrap --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,rust --enable-libphobos-checking=release --enable-libstdcxx-backtrace --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-link-serialization=2 --enable-multiarch --enable-multilib --enable-nls --enable-objc-gc=auto --enable-offload-defaulted --enable-offload-targets=nvptx-none=/build/gcc-14-zdkDXv/gcc-14-14.2.0/debian/tmp-nvptx/usr,amdgcn-amdhsa=/build/gcc-14-zdkDXv/gcc-14-14.2.0/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-build-config=bootstrap-lto-lean --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: acpi-cpufreq performance (Boost: Enabled) - CPU Microcode: 0xa101148 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
9654 llama ncnn llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Text Generation 128 llama-cpp: CPU BLAS - Llama-3.1-Tulu-3-8B-Q8_0 - Prompt Processing 512 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 2048 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Text Generation 128 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 512 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 1024 llama-cpp: CPU BLAS - Mistral-7B-Instruct-v0.3-Q8_0 - Prompt Processing 2048 llama-cpp: CPU BLAS - granite-3.0-3b-a800m-instruct-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 - Prompt Processing 2048 ncnn: CPU - mobilenet ncnn: CPU-v2-v2 - mobilenet-v2 ncnn: CPU-v3-v3 - mobilenet-v3 ncnn: CPU - shufflenet-v2 ncnn: CPU - mnasnet ncnn: CPU - efficientnet-b0 ncnn: CPU - blazeface ncnn: CPU - googlenet ncnn: CPU - vgg16 ncnn: CPU - resnet18 ncnn: CPU - alexnet ncnn: CPU - resnet50 ncnn: CPUv2-yolov3v2-yolov3 - mobilenetv2-yolov3 ncnn: CPU - yolov4-tiny ncnn: CPU - squeezenet_ssd ncnn: CPU - regnety_400m ncnn: CPU - vision_transformer ncnn: CPU - FastestDet a b 20.74 80.26 75.61 74.75 21.54 75.15 79.86 77.23 35.6 168.94 166.49 162.34 34.84 24.59 31.68 30.69 22.36 35.79 16.36 40.06 60.49 25.47 10.56 43.4 34.84 44 51.3 142.51 71.8 35.47 20.55 79.31 75.72 78.96 21.41 78.25 78.85 79.17 36.04 169.42 164.59 161.38 37.56 25.39 27.3 31.22 24.67 36.82 16.32 44.6 67.76 27.73 11.32 46.85 37.56 47.74 53.55 141.36 71.49 35.68 OpenBenchmarking.org
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 b4397 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128 a b 5 10 15 20 25 20.74 20.55 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 512 a b 20 40 60 80 100 80.26 79.31 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 1024 a b 20 40 60 80 100 75.61 75.72 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 2048 a b 20 40 60 80 100 74.75 78.96 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Text Generation 128 a b 5 10 15 20 25 21.54 21.41 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 512 a b 20 40 60 80 100 75.15 78.25 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 1024 a b 20 40 60 80 100 79.86 78.85 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 2048 a b 20 40 60 80 100 77.23 79.17 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Text Generation 128 a b 8 16 24 32 40 35.60 36.04 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 512 a b 40 80 120 160 200 168.94 169.42 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 1024 a b 40 80 120 160 200 166.49 164.59 1. (CXX) g++ options: -O3
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 b4397 Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 2048 a b 40 80 120 160 200 162.34 161.38 1. (CXX) g++ options: -O3
NCNN Target: CPU - Model: mobilenet OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: mobilenet a b 9 18 27 36 45 34.84 37.56 MIN: 34.6 / MAX: 36.42 MIN: 37.35 / MAX: 39.06 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU-v2-v2 - Model: mobilenet-v2 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU-v2-v2 - Model: mobilenet-v2 a b 6 12 18 24 30 24.59 25.39 MIN: 24.4 / MAX: 26.24 MIN: 25.06 / MAX: 27.15 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU-v3-v3 - Model: mobilenet-v3 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU-v3-v3 - Model: mobilenet-v3 a b 7 14 21 28 35 31.68 27.30 MIN: 27.33 / MAX: 469.66 MIN: 27.04 / MAX: 38.58 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: shufflenet-v2 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: shufflenet-v2 a b 7 14 21 28 35 30.69 31.22 MIN: 29.52 / MAX: 40.42 MIN: 30.69 / MAX: 67.53 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: mnasnet OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: mnasnet a b 6 12 18 24 30 22.36 24.67 MIN: 22.21 / MAX: 24.51 MIN: 24.37 / MAX: 28.09 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: efficientnet-b0 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: efficientnet-b0 a b 8 16 24 32 40 35.79 36.82 MIN: 35.61 / MAX: 37.49 MIN: 36.52 / MAX: 51.88 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: blazeface OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: blazeface a b 4 8 12 16 20 16.36 16.32 MIN: 16.24 / MAX: 17.79 MIN: 16.09 / MAX: 34.43 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: googlenet OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: googlenet a b 10 20 30 40 50 40.06 44.60 MIN: 39.88 / MAX: 41.52 MIN: 44.28 / MAX: 46.21 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: vgg16 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: vgg16 a b 15 30 45 60 75 60.49 67.76 MIN: 59.38 / MAX: 67.14 MIN: 66.94 / MAX: 70.22 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: resnet18 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: resnet18 a b 7 14 21 28 35 25.47 27.73 MIN: 25.27 / MAX: 26.99 MIN: 27.44 / MAX: 29.28 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: alexnet OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: alexnet a b 3 6 9 12 15 10.56 11.32 MIN: 10.2 / MAX: 11.97 MIN: 10.59 / MAX: 13.13 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: resnet50 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: resnet50 a b 11 22 33 44 55 43.40 46.85 MIN: 42.52 / MAX: 195.1 MIN: 46.41 / MAX: 50.7 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPUv2-yolov3v2-yolov3 - Model: mobilenetv2-yolov3 OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPUv2-yolov3v2-yolov3 - Model: mobilenetv2-yolov3 a b 9 18 27 36 45 34.84 37.56 MIN: 34.6 / MAX: 36.42 MIN: 37.35 / MAX: 39.06 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: yolov4-tiny OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: yolov4-tiny a b 11 22 33 44 55 44.00 47.74 MIN: 42.75 / MAX: 45.75 MIN: 46.8 / MAX: 49.69 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: squeezenet_ssd OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: squeezenet_ssd a b 12 24 36 48 60 51.30 53.55 MIN: 50.89 / MAX: 53.01 MIN: 52.9 / MAX: 69.09 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: regnety_400m OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: regnety_400m a b 30 60 90 120 150 142.51 141.36 MIN: 141.15 / MAX: 211.34 MIN: 139.15 / MAX: 147.1 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: vision_transformer OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: vision_transformer a b 16 32 48 64 80 71.80 71.49 MIN: 68.53 / MAX: 76.02 MIN: 68.86 / MAX: 74.1 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
NCNN Target: CPU - Model: FastestDet OpenBenchmarking.org ms, Fewer Is Better NCNN 20241226 Target: CPU - Model: FastestDet a b 8 16 24 32 40 35.47 35.68 MIN: 35.2 / MAX: 37.17 MIN: 35.42 / MAX: 37.42 1. (CXX) g++ options: -O3 -rdynamic -lgomp -lpthread
Phoronix Test Suite v10.8.5