Google Axion C4A vs. AWS Graviton4 Benchmarks

Google Axion C4A versus AWS Graviton4 cloud performance at 48 vCPU size. Benchmarks by Michael Larabel for a future article.

HTML result view exported from: https://openbenchmarking.org/result/2411071-NE-AXIONC4AG01&sgm=1&rdt&grw.

Google Axion C4A vs. AWS Graviton4 BenchmarksProcessorMotherboardMemoryDiskNetworkChipsetOSKernelCompilerFile-SystemSystem LayerC4A AxionR8g.12xlarge Graviton4C4A Axion highmemARMv8 Neoverse-V2 (48 Cores)KVM Google Compute Engine12 x 16GB RAM215GB nvme_card-pdGoogle Compute Engine VirtualUbuntu 24.046.8.0-1015-gcp (aarch64)GCC 13.2.0ext4googleAmazon EC2 r8g.12xlarge (1.0 BIOS)Amazon Device 0200384GB215GB Amazon Elastic Block StoreAmazon Elastic6.8.0-1016-aws (aarch64)amazonKVM Google Compute Engine24 x 16GB RAM215GB nvme_card-pdGoogle Compute Engine Virtual6.8.0-1015-gcp (aarch64)googleOpenBenchmarking.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 Java Details- C4A Axion: OpenJDK Runtime Environment (build 11.0.24+8-post-Ubuntu-1ubuntu324.04.1)Python Details- Python 3.12.3Security 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

Google Axion C4A vs. AWS Graviton4 Benchmarksxmrig: GhostRider - 1Mcloverleaf: clover_bm64_shortgromacs: MPI CPU - water_GMX50_barelammps: 20k Atomsrodinia: OpenMP CFD Solveraskap: tConvolve MPI - Degriddingaskap: tConvolve MPI - Griddingaskap: tConvolve OpenMP - Griddingaskap: tConvolve OpenMP - Degriddingaskap: Hogbom Clean OpenMPpennant: sedovbigpennant: leblancbigamg: lulesh: minife: Smallopenssl: ChaCha20-Poly1305openssl: AES-128-GCMopenssl: AES-256-GCMincompact3d: input.i3d 193 Cells Per Directionopenssl: ChaCha20coremark: CoreMark Size 666 - Iterations Per Secondcompress-7zip: Compression Ratingcompress-7zip: Decompression Ratingjohn-the-ripper: bcryptjohn-the-ripper: Blowfishjohn-the-ripper: HMAC-SHA512build-llvm: Ninjabuild-godot: Time To Compilebuild-nodejs: Time To Compilesrsran: PDSCH Processor Benchmark, Throughput Totalsrsran: PUSCH Processor Benchmark, Throughput Totalclickhouse: 100M Rows Hits Dataset, First Run / Cold Cacheclickhouse: 100M Rows Hits Dataset, Second Runclickhouse: 100M Rows Hits Dataset, Third Runrocksdb: Rand Readrocksdb: Update Randrocksdb: Read While Writingrocksdb: Read Rand Write RandC4A AxionR8g.12xlarge Graviton4C4A Axion highmem4298.226.153.98440.8313.10617695.617890.625980.143530.72254.717.3101305.041512218835300020794.05247199.6704291528702661410288032318576660039.676580111008372281801441191.863234306082266745457494576389864667204.046153.601260.46113010.91152.5438.85465.02480.41288511691999493628242945500504498.721.143.70037.8063.32817302.216748.71479219993.61796.537.5089635.522060250138633319648.30857879.65753548226722777101207319985367832010.0560172785752084031371372.253740303046249500428134280086991000218.480166.732426.88011558.01103.3430.54446.90450.802633798701076237662108146060784393.926.403.98940.9733.07517624.117892.225854.640995.02380.957.3219225.091936217368233320782.75146384.5704444925732658803620972318042495209.872513451008589210531464649.879962307992266620457584570190421667204.758153.336262.17613225.51152.5438.49449.54486.43281897910101370663054344565384OpenBenchmarking.org

Xmrig

Variant: GhostRider - Hash Count: 1M

OpenBenchmarking.orgH/s, More Is BetterXmrig 6.21Variant: GhostRider - Hash Count: 1MC4A AxionR8g.12xlarge Graviton4C4A Axion highmem10002000300040005000SE +/- 12.97, N = 3SE +/- 14.92, N = 3SE +/- 13.21, N = 34298.24498.74393.91. (CXX) g++ options: -fexceptions -fno-rtti -O3 -Ofast -static-libgcc -static-libstdc++ -rdynamic -lssl -lcrypto -luv -lpthread -lrt -ldl -lhwloc

CloverLeaf

Input: clover_bm64_short

OpenBenchmarking.orgSeconds, Fewer Is BetterCloverLeaf 1.3Input: clover_bm64_shortC4A AxionR8g.12xlarge Graviton4C4A Axion highmem612182430SE +/- 0.03, N = 3SE +/- 0.10, N = 3SE +/- 0.01, N = 326.1521.1426.401. (F9X) gfortran options: -O3 -march=native -funroll-loops -fopenmp

GROMACS

Implementation: MPI CPU - Input: water_GMX50_bare

OpenBenchmarking.orgNs Per Day, More Is BetterGROMACS 2024Implementation: MPI CPU - Input: water_GMX50_bareC4A AxionR8g.12xlarge Graviton4C4A Axion highmem0.89751.7952.69253.594.4875SE +/- 0.001, N = 3SE +/- 0.001, N = 3SE +/- 0.002, N = 33.9843.7003.9891. (CXX) g++ options: -O3 -lm

LAMMPS Molecular Dynamics Simulator

Model: 20k Atoms

OpenBenchmarking.orgns/day, More Is BetterLAMMPS Molecular Dynamics Simulator 23Jun2022Model: 20k AtomsC4A AxionR8g.12xlarge Graviton4C4A Axion highmem918273645SE +/- 0.03, N = 3SE +/- 0.04, N = 3SE +/- 0.07, N = 340.8337.8140.971. (CXX) g++ options: -O3 -lm -ldl

Rodinia

Test: OpenMP CFD Solver

OpenBenchmarking.orgSeconds, Fewer Is BetterRodinia 3.1Test: OpenMP CFD SolverC4A AxionR8g.12xlarge Graviton4C4A Axion highmem0.74881.49762.24642.99523.744SE +/- 0.024, N = 3SE +/- 0.073, N = 14SE +/- 0.011, N = 33.1063.3283.0751. (CXX) g++ options: -O2 -lOpenCL

ASKAP

Test: tConvolve MPI - Degridding

OpenBenchmarking.orgMpix/sec, More Is BetterASKAP 1.0Test: tConvolve MPI - DegriddingC4A AxionR8g.12xlarge Graviton4C4A Axion highmem4K8K12K16K20KSE +/- 229.66, N = 3SE +/- 109.76, N = 3SE +/- 65.53, N = 317695.617302.217624.11. (CXX) g++ options: -O3 -fstrict-aliasing -fopenmp

ASKAP

Test: tConvolve MPI - Gridding

OpenBenchmarking.orgMpix/sec, More Is BetterASKAP 1.0Test: tConvolve MPI - GriddingC4A AxionR8g.12xlarge Graviton4C4A Axion highmem4K8K12K16K20KSE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 117.41, N = 317890.616748.717892.21. (CXX) g++ options: -O3 -fstrict-aliasing -fopenmp

ASKAP

Test: tConvolve OpenMP - Gridding

OpenBenchmarking.orgMillion Grid Points Per Second, More Is BetterASKAP 1.0Test: tConvolve OpenMP - GriddingC4A AxionR8g.12xlarge Graviton4C4A Axion highmem6K12K18K24K30KSE +/- 286.07, N = 15SE +/- 0.00, N = 3SE +/- 408.18, N = 1525980.114792.025854.61. (CXX) g++ options: -O3 -fstrict-aliasing -fopenmp

ASKAP

Test: tConvolve OpenMP - Degridding

OpenBenchmarking.orgMillion Grid Points Per Second, More Is BetterASKAP 1.0Test: tConvolve OpenMP - DegriddingC4A AxionR8g.12xlarge Graviton4C4A Axion highmem9K18K27K36K45KSE +/- 575.94, N = 15SE +/- 487.63, N = 3SE +/- 845.25, N = 1543530.719993.640995.01. (CXX) g++ options: -O3 -fstrict-aliasing -fopenmp

ASKAP

Test: Hogbom Clean OpenMP

OpenBenchmarking.orgIterations Per Second, More Is BetterASKAP 1.0Test: Hogbom Clean OpenMPC4A AxionR8g.12xlarge Graviton4C4A Axion highmem5001000150020002500SE +/- 37.89, N = 15SE +/- 10.82, N = 3SE +/- 0.00, N = 32254.711796.532380.951. (CXX) g++ options: -O3 -fstrict-aliasing -fopenmp

Pennant

Test: sedovbig

OpenBenchmarking.orgHydro Cycle Time - Seconds, Fewer Is BetterPennant 1.0.1Test: sedovbigC4A AxionR8g.12xlarge Graviton4C4A Axion highmem246810SE +/- 0.029628, N = 3SE +/- 0.010434, N = 3SE +/- 0.014612, N = 37.3101307.5089637.3219221. (CXX) g++ options: -fopenmp -lmpi_cxx -lmpi

Pennant

Test: leblancbig

OpenBenchmarking.orgHydro Cycle Time - Seconds, Fewer Is BetterPennant 1.0.1Test: leblancbigC4A AxionR8g.12xlarge Graviton4C4A Axion highmem1.24252.4853.72754.976.2125SE +/- 0.017045, N = 3SE +/- 0.001283, N = 3SE +/- 0.003493, N = 35.0415125.5220605.0919361. (CXX) g++ options: -fopenmp -lmpi_cxx -lmpi

Algebraic Multi-Grid Benchmark

OpenBenchmarking.orgFigure Of Merit, More Is BetterAlgebraic Multi-Grid Benchmark 1.2C4A AxionR8g.12xlarge Graviton4C4A Axion highmem500M1000M1500M2000M2500MSE +/- 2463059.28, N = 3SE +/- 4844514.33, N = 3SE +/- 4022914.05, N = 32188353000250138633321736823331. (CC) gcc options: -lparcsr_ls -lparcsr_mv -lseq_mv -lIJ_mv -lkrylov -lHYPRE_utilities -lm -fopenmp -lmpi

LULESH

OpenBenchmarking.orgz/s, More Is BetterLULESH 2.0.3C4A AxionR8g.12xlarge Graviton4C4A Axion highmem4K8K12K16K20KSE +/- 282.43, N = 3SE +/- 176.92, N = 3SE +/- 216.02, N = 320794.0519648.3120782.751. (CXX) g++ options: -O3 -fopenmp -lm -lmpi_cxx -lmpi

miniFE

Problem Size: Small

OpenBenchmarking.orgCG Mflops, More Is BetterminiFE 2.2Problem Size: SmallC4A AxionR8g.12xlarge Graviton4C4A Axion highmem12K24K36K48K60KSE +/- 12.79, N = 3SE +/- 13.99, N = 3SE +/- 51.46, N = 347199.657879.646384.51. (CXX) g++ options: -O3 -fopenmp -lmpi_cxx -lmpi

OpenSSL

Algorithm: ChaCha20-Poly1305

OpenBenchmarking.orgbyte/s, More Is BetterOpenSSLAlgorithm: ChaCha20-Poly1305C4A AxionR8g.12xlarge Graviton4C4A Axion highmem15000M30000M45000M60000M75000MSE +/- 36045318.84, N = 3SE +/- 938120.62, N = 3SE +/- 3273978.88, N = 37042915287057535482267704444925731. OpenSSL 3.0.13 30 Jan 2024 (Library: OpenSSL 3.0.13 30 Jan 2024)

OpenSSL

Algorithm: AES-128-GCM

OpenBenchmarking.orgbyte/s, More Is BetterOpenSSLAlgorithm: AES-128-GCMC4A AxionR8g.12xlarge Graviton4C4A Axion highmem60000M120000M180000M240000M300000MSE +/- 7311961.78, N = 3SE +/- 4106302.62, N = 3SE +/- 138446489.36, N = 32661410288032277710120732658803620971. OpenSSL 3.0.13 30 Jan 2024 (Library: OpenSSL 3.0.13 30 Jan 2024)

OpenSSL

Algorithm: AES-256-GCM

OpenBenchmarking.orgbyte/s, More Is BetterOpenSSLAlgorithm: AES-256-GCMC4A AxionR8g.12xlarge Graviton4C4A Axion highmem50000M100000M150000M200000M250000MSE +/- 119817516.04, N = 3SE +/- 8479460.94, N = 3SE +/- 130040183.67, N = 32318576660031998536783202318042495201. OpenSSL 3.0.13 30 Jan 2024 (Library: OpenSSL 3.0.13 30 Jan 2024)

Xcompact3d Incompact3d

Input: input.i3d 193 Cells Per Direction

OpenBenchmarking.orgSeconds, Fewer Is BetterXcompact3d Incompact3d 2021-03-11Input: input.i3d 193 Cells Per DirectionC4A AxionR8g.12xlarge Graviton4C4A Axion highmem3691215SE +/- 0.00471123, N = 3SE +/- 0.01105122, N = 3SE +/- 0.01510711, N = 39.6765801110.056017209.872513451. (F9X) gfortran options: -cpp -O2 -funroll-loops -floop-optimize -fcray-pointer -fbacktrace -lmpi_usempif08 -lmpi_mpifh -lmpi -lopen-rte -lopen-pal -lhwloc -levent_core -levent_pthreads -lm -lz

OpenSSL

Algorithm: ChaCha20

OpenBenchmarking.orgbyte/s, More Is BetterOpenSSLAlgorithm: ChaCha20C4A AxionR8g.12xlarge Graviton4C4A Axion highmem20000M40000M60000M80000M100000MSE +/- 48601049.47, N = 3SE +/- 2744256.77, N = 3SE +/- 7430771.27, N = 3100837228180785752084031008589210531. OpenSSL 3.0.13 30 Jan 2024 (Library: OpenSSL 3.0.13 30 Jan 2024)

Coremark

CoreMark Size 666 - Iterations Per Second

OpenBenchmarking.orgIterations/Sec, More Is BetterCoremark 1.0CoreMark Size 666 - Iterations Per SecondC4A AxionR8g.12xlarge Graviton4C4A Axion highmem300K600K900K1200K1500KSE +/- 20293.48, N = 3SE +/- 7155.40, N = 3SE +/- 11247.92, N = 101441191.861371372.251464649.881. (CC) gcc options: -O2 -lrt" -lrt

7-Zip Compression

Test: Compression Rating

OpenBenchmarking.orgMIPS, More Is Better7-Zip Compression 24.05Test: Compression RatingC4A AxionR8g.12xlarge Graviton4C4A Axion highmem70K140K210K280K350KSE +/- 748.65, N = 3SE +/- 430.32, N = 3SE +/- 354.33, N = 33060823030463079921. (CXX) g++ options: -lpthread -ldl -O2 -fPIC

7-Zip Compression

Test: Decompression Rating

OpenBenchmarking.orgMIPS, More Is Better7-Zip Compression 24.05Test: Decompression RatingC4A AxionR8g.12xlarge Graviton4C4A Axion highmem60K120K180K240K300KSE +/- 67.68, N = 3SE +/- 97.12, N = 3SE +/- 45.09, N = 32667452495002666201. (CXX) g++ options: -lpthread -ldl -O2 -fPIC

John The Ripper

Test: bcrypt

OpenBenchmarking.orgReal C/S, More Is BetterJohn The Ripper 2023.03.14Test: bcryptC4A AxionR8g.12xlarge Graviton4C4A Axion highmem10K20K30K40K50KSE +/- 25.03, N = 3SE +/- 3.33, N = 3SE +/- 19.55, N = 34574942813457581. (CC) gcc options: -lssl -lcrypto -fopenmp -lm -lrt -lz -ldl -lcrypt

John The Ripper

Test: Blowfish

OpenBenchmarking.orgReal C/S, More Is BetterJohn The Ripper 2023.03.14Test: BlowfishC4A AxionR8g.12xlarge Graviton4C4A Axion highmem10K20K30K40K50KSE +/- 14.62, N = 3SE +/- 11.70, N = 3SE +/- 25.54, N = 34576342800457011. (CC) gcc options: -lssl -lcrypto -fopenmp -lm -lrt -lz -ldl -lcrypt

John The Ripper

Test: HMAC-SHA512

OpenBenchmarking.orgReal C/S, More Is BetterJohn The Ripper 2023.03.14Test: HMAC-SHA512C4A AxionR8g.12xlarge Graviton4C4A Axion highmem20M40M60M80M100MSE +/- 28852.11, N = 3SE +/- 24062.42, N = 3SE +/- 34275.03, N = 38986466786991000904216671. (CC) gcc options: -lssl -lcrypto -fopenmp -lm -lrt -lz -ldl -lcrypt

Timed LLVM Compilation

Build System: Ninja

OpenBenchmarking.orgSeconds, Fewer Is BetterTimed LLVM Compilation 16.0Build System: NinjaC4A AxionR8g.12xlarge Graviton4C4A Axion highmem50100150200250SE +/- 0.23, N = 3SE +/- 0.04, N = 3SE +/- 0.31, N = 3204.05218.48204.76

Timed Godot Game Engine Compilation

Time To Compile

OpenBenchmarking.orgSeconds, Fewer Is BetterTimed Godot Game Engine Compilation 4.0Time To CompileC4A AxionR8g.12xlarge Graviton4C4A Axion highmem4080120160200SE +/- 0.21, N = 3SE +/- 0.52, N = 3SE +/- 0.33, N = 3153.60166.73153.34

Timed Node.js Compilation

Time To Compile

OpenBenchmarking.orgSeconds, Fewer Is BetterTimed Node.js Compilation 21.7.2Time To CompileC4A AxionR8g.12xlarge Graviton4C4A Axion highmem90180270360450SE +/- 0.26, N = 3SE +/- 0.49, N = 3SE +/- 0.23, N = 3260.46426.88262.18

srsRAN Project

Test: PDSCH Processor Benchmark, Throughput Total

OpenBenchmarking.orgMbps, More Is BettersrsRAN Project 23.10.1-20240325Test: PDSCH Processor Benchmark, Throughput TotalC4A AxionR8g.12xlarge Graviton4C4A Axion highmem3K6K9K12K15KSE +/- 68.81, N = 3SE +/- 67.62, N = 3SE +/- 29.21, N = 313010.911558.013225.51. (CXX) g++ options: -O3 -fno-trapping-math -fno-math-errno -ldl

srsRAN Project

Test: PUSCH Processor Benchmark, Throughput Total

OpenBenchmarking.orgMbps, More Is BettersrsRAN Project 23.10.1-20240325Test: PUSCH Processor Benchmark, Throughput TotalC4A AxionR8g.12xlarge Graviton4C4A Axion highmem2004006008001000SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.03, N = 31152.51103.31152.5MIN: 694.9MIN: 637MIN: 695 / MAX: 1152.61. (CXX) g++ options: -O3 -fno-trapping-math -fno-math-errno -ldl

ClickHouse

100M Rows Hits Dataset, First Run / Cold Cache

OpenBenchmarking.orgQueries Per Minute, Geo Mean, More Is BetterClickHouse 22.12.3.5100M Rows Hits Dataset, First Run / Cold CacheC4A AxionR8g.12xlarge Graviton4C4A Axion highmem100200300400500SE +/- 5.15, N = 9SE +/- 7.17, N = 9SE +/- 2.27, N = 3438.85430.54438.49MIN: 35.57 / MAX: 6666.67MIN: 33.69 / MAX: 6666.67MIN: 35.48 / MAX: 6666.67

ClickHouse

100M Rows Hits Dataset, Second Run

OpenBenchmarking.orgQueries Per Minute, Geo Mean, More Is BetterClickHouse 22.12.3.5100M Rows Hits Dataset, Second RunC4A AxionR8g.12xlarge Graviton4C4A Axion highmem100200300400500SE +/- 4.32, N = 9SE +/- 8.12, N = 9SE +/- 6.63, N = 3465.02446.90449.54MIN: 35.65 / MAX: 7500MIN: 33.3 / MAX: 6666.67MIN: 35.63 / MAX: 5000

ClickHouse

100M Rows Hits Dataset, Third Run

OpenBenchmarking.orgQueries Per Minute, Geo Mean, More Is BetterClickHouse 22.12.3.5100M Rows Hits Dataset, Third RunC4A AxionR8g.12xlarge Graviton4C4A Axion highmem110220330440550SE +/- 3.70, N = 9SE +/- 5.59, N = 9SE +/- 3.07, N = 3480.41450.80486.43MIN: 35.46 / MAX: 6666.67MIN: 33.33 / MAX: 6666.67MIN: 35.46 / MAX: 6666.67

RocksDB

Test: Random Read

OpenBenchmarking.orgOp/s, More Is BetterRocksDB 9.0Test: Random ReadC4A AxionR8g.12xlarge Graviton4C4A Axion highmem60M120M180M240M300MSE +/- 88148.93, N = 3SE +/- 9107.63, N = 3SE +/- 35258.83, N = 32885116912633798702818979101. (CXX) g++ options: -O3 -march=native -pthread -fno-builtin-memcmp -fno-rtti

RocksDB

Test: Update Random

OpenBenchmarking.orgOp/s, More Is BetterRocksDB 9.0Test: Update RandomC4A AxionR8g.12xlarge Graviton4C4A Axion highmem200K400K600K800K1000KSE +/- 9437.64, N = 3SE +/- 2869.27, N = 3SE +/- 1257.72, N = 3999493107623710137061. (CXX) g++ options: -O3 -march=native -pthread -fno-builtin-memcmp -fno-rtti

RocksDB

Test: Read While Writing

OpenBenchmarking.orgOp/s, More Is BetterRocksDB 9.0Test: Read While WritingC4A AxionR8g.12xlarge Graviton4C4A Axion highmem1.4M2.8M4.2M5.6M7MSE +/- 42437.10, N = 3SE +/- 84244.73, N = 3SE +/- 6269.58, N = 36282429662108163054341. (CXX) g++ options: -O3 -march=native -pthread -fno-builtin-memcmp -fno-rtti

RocksDB

Test: Read Random Write Random

OpenBenchmarking.orgOp/s, More Is BetterRocksDB 9.0Test: Read Random Write RandomC4A AxionR8g.12xlarge Graviton4C4A Axion highmem1000K2000K3000K4000K5000KSE +/- 4377.96, N = 3SE +/- 1527.21, N = 3SE +/- 6470.49, N = 34550050460607845653841. (CXX) g++ options: -O3 -march=native -pthread -fno-builtin-memcmp -fno-rtti

Geometric Mean Of All Test Results

Result Composite - Google Axion C4A vs. AWS Graviton4 Benchmarks

OpenBenchmarking.orgGeometric Mean, More Is BetterGeometric Mean Of All Test ResultsResult Composite - Google Axion C4A vs. AWS Graviton4 BenchmarksC4A AxionR8g.12xlarge Graviton4C4A Axion highmem7K14K21K28K35K33153.5630297.3533124.37

Phoronix Test Suite v10.8.5