Skip to main content

GeekBench5

Geekbench5 is a third party tool that runs its own, pre-defined set of workloads to measure CPU performance. GeekBench5 is often used to compare cloud performance.

How to package GeekBench5

GeekBench5 is a commercial workload. VirtualClient cannot distribute the license and binary. You need to follow the following steps to package this workload and make it available locally or in a storage that you own.

  1. GeekBench can be downloaded here https://www.geekbench.com/download/.
  2. You need to purchase a license for it to run properly https://www.primatelabs.com/store/.
  3. Geekbench 5.preferences is the license file you get after purchasing. Package the binaries and license in this structure. You only need to pack the runtimes you are going to run. (You don't need "win-arm64, win-x64" if you don't plan on running on Windows).
geekbench-5.1.0
├───geekbench5.vcpkg
│
├───linux-x64/
│       Geekbench 5.preferences
│       geekbench.plar
│       geekbench5
│       geekbench_x86_64
│
├───win-arm64/
│       Geekbench 5.preferences
│       geekbench.plar
│       geekbench5.exe
│       geekbench_aarch64.exe
│
└───win-x64/
        amd_ags_x64.dll
        cpuidsdk64.dll
        Geekbench 5.preferences
        geekbench.plar
        geekbench5.exe
        geekbench_x86_64.exe
        pl_opencl_x86_64.dll
  1. Zip the geekbench-5.1.0 directory into geekbench-5.1.0.zip, make sure the runtimes folders are the top level after extraction, and no extra /geekbench-5.1.0/ top directory is created.
7z a geekbench-5.1.0.zip ./geekbench-5.1.0/*

or

cd geekbench-5.1.0; zip -r ../geekbench-5.1.0.zip *
  1. Modify the GeekBench profile as needed. If you are using your own blob storage, you can use the profile as is. If you are copying the zip file locally under vc/packages, you can simply remove the DependencyPackageInstallation step.

What is Being Measured?

Geekbench5 reports four different types of scores, a value that's calculated by comparing the device's performance against a baseline. The baseline score is 1000, with higher numbers being better and double the score indicating double the performance.

  1. Geekbench score. This is the overall score, calculated using the single-core and multi-core scores.

  2. Single-core and multi-core score. This is calculated using the corresponding single-core and multi-core scores from the three subsections detailed below.

  3. Subsection score. The workloads are grouped by how the workload exercise the system, and are calculated using the individual workload scores. The subsections are Cryptography, Integer, and Floating-Point workloads.

  4. Workload score. Each individual workload reports a score. Additionally, each individual workload reports the numerical result of the workload - 5 GB/sec, 35 images/sec, etc.

Geekbench runs a corresponding single-core and multi-core version for each of the following common CPU-intensive algorithms.

NameDescription
AES-XTSRuns the AES encryption algorithm, widely used to secure communication channels
Text CompressionCompresses and decompress text using the LZMA compression algorithm
Image CompressionCompresses and decompress a photograph using libjpeg-turbo for JPEG and libpng for PNG
NavigationComputes driving directions with approximated times using Dijkstra's algorithm
HTML5Models DOM creation from both server-side and client-side rendered HTML5 documents
SQLiteExecutes SQL queries against an in-memory database that mimics financial data
PDF RenderingParses and renders a PDF using the PDFium library
Text RenderingParses a Markdown-formatted document and renders it as rich text to a bitmap
ClangCompiles a C source file using AArch64 as the target architecture
CameraReplicates a photo sharing app, including image manipulation, using only the CPU
N-Body PhysicsComputes a 3D gravitation simulation using the Barnes-Hut method
Rigid Body PhysicsComputes a 2D physics simulation for rigid bodies using Lua and the BoxD physics lib
Gaussian BlurBlurs an image using a Gaussian spatial filter
Face DetectionDetects faces with the algorithm presented in “Rapid Object Detection using a Boosted Cascade of Simple Features” (2001) by Viola and Jones.
Horizon DetectionSearches for the horizon line in an image and rotates the image to make horizon line level if found
Image InpaintingReconstructs part of an image with data from other parts of the image
HDRTakes four SDR images and produces an HDR image
Ray TracingGenerates an image by tracing the path of light through an image plane and simulating the effects of its encounters with virtual objects
Structure from MotionTakes two 2D images of the same scene and constructs an estimate of the 3D coordinates of the points visible in both images
Speech RecognitionPerforms recognition of arbitrary English speech using PocketSphinx
Machine LearningExecutes a Convolutional Neural Network to perform an image classification task

Workload Metrics

The following metrics are examples of those captured by the Virtual Client when running the GeekBench5 workload.

ScenarioMetric NameExample Value (min)Example Value (max)Example Value (avg)Unit
AES-XTS (Multi-Core)Raw Value1.262.352.2361644951140167GB/sec
AES-XTS (Single-Core)Raw Value1.212.091.9808686210640743GB/sec
AES-XTS Score (Multi-Core)Test Score742.01379.01311.7852877307276score
AES-XTS Score (Single-Core)Test Score708.01226.01162.0089576547233score
Camera (Multi-Core)Raw Value5.029.218.824011943539614images/sec
Camera (Single-Core)Raw Value4.588.768.354275244299679images/sec
Camera Score (Multi-Core)Test Score433.0794.0761.2090119435396score
Camera Score (Single-Core)Test Score395.0756.0720.6886536373507score
Clang (Multi-Core)Raw Value2.517.837.425597176981524Klines/sec
Clang (Single-Core)Raw Value3.476.516.240559174810017Klines/sec
Clang Score (Multi-Core)Test Score322.01005.0953.0298588490771score
Clang Score (Single-Core)Test Score445.0836.0800.9562975027144score
Cryptography Score (Multi-Core)Test Score742.01379.01311.7852877307276score
Cryptography Score (Single-Core)Test Score708.01226.01162.0089576547233score
Face Detection (Multi-Core)Raw Value5.07.377.172499999999988images/sec
Face Detection (Single-Core)Raw Value3.836.976.708667209554847images/sec
Face Detection Score (Multi-Core)Test Score649.0957.0931.612920738328score
Face Detection Score (Single-Core)Test Score498.0905.0871.3536916395223score
FloatingPoint Score (Multi-Core)Test Score875.01099.01068.0024429967428score
FloatingPoint Score (Single-Core)Test Score642.0920.0892.807546145494score
Gaussian Blur (Multi-Core)Raw Value22.946.144.88300760043444Mpixels/sec
Gaussian Blur (Single-Core)Raw Value26.539.337.38417480998918Mpixels/sec
Gaussian Blur Score (Multi-Core)Test Score416.0839.0816.5222584147666score
Gaussian Blur Score (Single-Core)Test Score482.0716.0680.0738327904452score
HDR (Multi-Core)Raw Value14.124.824.09500542888169Mpixels/sec
HDR (Single-Core)Raw Value11.622.722.018431053202997Mpixels/sec
HDR Score (Multi-Core)Test Score1035.01817.01768.249457111835score
HDR Score (Single-Core)Test Score853.01662.01615.9367535287732score
HTML5 (Multi-Core)Raw Value437.1922.7886.5702225841469KElements/sec
HTML5 (Single-Core)Raw Value383.8951.7892.1498099891434KElements/sec
HTML5 Score (Multi-Core)Test Score372.0786.0755.0415309446254score
HTML5 Score (Single-Core)Test Score327.0811.0759.8032030401737score
Horizon Detection (Multi-Core)Raw Value17.426.025.055456026058658Mpixels/sec
Horizon Detection (Single-Core)Raw Value12.621.020.195276872964159Mpixels/sec
Horizon Detection Score (Multi-Core)Test Score705.01054.01016.4828990228014score
Horizon Detection Score (Single-Core)Test Score510.0853.0819.3455483170467score
Image Compression (Multi-Core)Raw Value30.151.549.88078175895753Mpixels/sec
Image Compression (Single-Core)Raw Value24.944.142.71745385450586Mpixels/sec
Image Compression Score (Multi-Core)Test Score636.01088.01054.4169381107493score
Image Compression Score (Single-Core)Test Score527.0933.0903.0100434310532score
Image Inpainting (Multi-Core)Raw Value49.885.381.46028773072759Mpixels/sec
Image Inpainting (Single-Core)Raw Value39.273.269.65016286644959Mpixels/sec
Image Inpainting Score (Multi-Core)Test Score1014.01738.01660.501628664495score
Image Inpainting Score (Single-Core)Test Score799.01492.01419.764115092291score
Integer Score (Multi-Core)Test Score625.0941.0914.0887622149837score
Integer Score (Single-Core)Test Score485.0832.0804.042345276873score
Machine Learning (Multi-Core)Raw Value15.927.125.364087947882785images/sec
Machine Learning (Single-Core)Raw Value14.724.222.42418566775247images/sec
Machine Learning Score (Multi-Core)Test Score410.0702.0656.4011943539631score
Machine Learning Score (Single-Core)Test Score380.0627.0580.3208469055375score
Multi-Core ScoreTest Score767.01005.0980.1767100977198score
N-Body Physics (Multi-Core)Raw Value752.3992.2927.4999999999999Kpairs/sec
N-Body Physics (Multi-Core)Raw Value1.011.251.1558626328699893Mpairs/sec
N-Body Physics (Single-Core)Raw Value596.1941.1888.0819761129206Kpairs/sec
N-Body Physics Score (Multi-Core)Test Score601.0995.0923.1102062975027score
N-Body Physics Score (Single-Core)Test Score476.0752.0709.78664495114score
Navigation (Multi-Core)Raw Value1.482.892.7101140065146289MTE/sec
Navigation (Single-Core)Raw Value1.532.452.2940499457112084MTE/sec
Navigation Score (Multi-Core)Test Score524.01024.0961.0165580890337score
Navigation Score (Single-Core)Test Score544.0867.0813.499457111835score
PDF Rendering (Multi-Core)Raw Value23.554.851.804044516829538Mpixels/sec
PDF Rendering (Single-Core)Raw Value25.743.942.409690553746077Mpixels/sec
PDF Rendering Score (Multi-Core)Test Score433.01010.0954.5415309446254score
PDF Rendering Score (Single-Core)Test Score473.0809.0781.46335504886score
Ray Tracing (Multi-Core)Raw Value1.01.11.059050544540641Mpixels/sec
Ray Tracing (Multi-Core)Raw Value770.8999.9962.744660194175Kpixels/sec
Ray Tracing (Single-Core)Raw Value510.2914.4866.0622964169361Kpixels/sec
Ray Tracing Score (Multi-Core)Test Score960.01365.01315.3683496199784score
Ray Tracing Score (Single-Core)Test Score635.01139.01078.4891422366994score
Rigid Body Physics (Multi-Core)Raw Value5001.67430.67166.121661237785FPS
Rigid Body Physics (Single-Core)Raw Value3365.65952.75784.287106406104FPS
Rigid Body Physics Score (Multi-Core)Test Score807.01199.01156.6815960912052score
Rigid Body Physics Score (Single-Core)Test Score543.0961.0933.6305646036916score
SQLite (Multi-Core)Raw Value102.0283.0269.9680238870789Krows/sec
SQLite (Single-Core)Raw Value131.8258.3246.84967426710126Krows/sec
SQLite Score (Multi-Core)Test Score326.0903.0861.6783387622149score
SQLite Score (Single-Core)Test Score421.0825.0787.8889793702497score
Single-Core ScoreTest Score605.0875.0848.5890336590662score
Speech Recognition (Multi-Core)Raw Value28.742.239.75352877307272Words/sec
Speech Recognition (Single-Core)Raw Value14.329.728.06102062975032Words/sec
Speech Recognition Score (Multi-Core)Test Score898.01319.01243.3998371335507score
Speech Recognition Score (Single-Core)Test Score448.0929.0877.7090119435396score
Structure from Motion (Multi-Core)Raw Value4.547.37.1137079261672009Kpixels/sec
Structure from Motion (Single-Core)Raw Value3.36.546.369581976112893Kpixels/sec
Structure from Motion Score (Multi-Core)Test Score507.0815.0794.0420738327905score
Structure from Motion Score (Single-Core)Test Score369.0730.0710.978555917481score
Text Compression (Multi-Core)Raw Value3.876.276.006568946796989MB/sec
Text Compression (Single-Core)Raw Value2.494.794.55621064060805MB/sec
Text Compression Score (Multi-Core)Test Score765.01241.01187.6183496199784score
Text Compression Score (Single-Core)Test Score493.0946.0900.8458197611292score
Text Rendering (Multi-Core)Raw Value142.5276.8262.3624592833867KB/sec
Text Rendering (Single-Core)Raw Value128.8268.0251.43379478827428KB/sec
Text Rendering Score (Multi-Core)Test Score447.0869.0823.4541259500543score
Text Rendering Score (Single-Core)Test Score404.0841.0789.1682953311618score

GeekBench6

Geekbench6 is a third party tool that runs its own, pre-defined set of workloads to measure CPU performance. GeekBench6 is often used to compare cloud performance, so it's valuable to see how our changes impact GeekBench6 results.

How to package GeekBench6

GeekBench6 is a commercial workload. VirtualClient cannot distribute the license and binary. You need to follow the following steps to package this workload and make it available locally or in a storage that you own.

  1. GeekBench can be downloaded here https://www.geekbench.com/download/.
  2. You need to purchase a license for it to run properly https://www.primatelabs.com/store/.
  3. Geekbench 5.preferences is the license file you get after purchasing. Package the binaries and license in this structure. You only need to pack the runtimes you are going to run. (You don't need "win-arm64, win-x64" if you don't plan on running on Windows).
geekbench-5.1.0
├───geekbench6.vcpkg
│
├───linux-arm64/
│       Geekbench 6.preferences
│       geekbench.plar
│       geekbench6
│       geekbench_aarch64
│
├───linux-x64/
│       Geekbench 6.preferences
│       geekbench.plar
│       geekbench6
│       geekbench_x86_64
│
├───win-arm64/
│       Geekbench 6.preferences
│       geekbench.plar
│       geekbench6.exe
│       geekbench_aarch64.exe
│
└───win-x64/
        amd_ags_x64.dll
        cpuidsdk64.dll
        Geekbench 6.preferences
        geekbench.plar
        geekbench6.exe
        geekbench_x86_64.exe
        pl_opencl_x86_64.dll
  1. Zip the geekbench-6.2.2 directory into geekbench-6.2.2.zip, make sure the runtimes folders are the top level after extraction, and no extra /geekbench-6.2.2/ top directory is created.
7z a geekbench-6.2.2.zip ./geekbench-6.2.2/*

or

cd geekbench-6.2.2; zip -r ../geekbench-6.2.2.zip *
  1. Modify the GeekBench profile as needed. If you are using your own blob storage, you can use the profile as is. If you are copying the zip file locally under vc/packages, you can simply remove the DependencyPackageInstallation step.

What is Being Measured?

Geekbench6 reports four different types of scores, a value that's calculated by comparing the device's performance against a baseline. The baseline score is 1000, with higher numbers being better and double the score indicating double the performance.

  1. Geekbench score. This is the overall score, calculated using the single-core and multi-core scores.

  2. Single-core and multi-core score. This is calculated using the corresponding single-core and multi-core scores from the two subsections detailed below.

  3. Subsection score. The workloads are grouped by how the workload exercise the system, and are calculated using the individual workload scores. The subsections are Integer, and Floating-Point workloads.

  4. Workload score. Each individual workload reports a score. Additionally, each individual workload reports the numerical result of the workload - 5 GB/sec, 35 images/sec, etc.

Geekbench runs a corresponding single-core and multi-core version for each of the following common CPU-intensive algorithms.

NameDescription
File CopmressionCompresses and decompresses files using the LZ4 and ZSTD compression codecs
Text CompressionCompresses and decompresses text using the LZMA compression algorithm
Image CompressionCompresses and decompresses a photograph using libjpeg-turbo for JPEG and libpng for PNG
NavigationComputes driving directions with approximated times using Dijkstra's algorithm
HTML5Models DOM creation from both server-side and client-side rendered HTML5 documents
PDF RenderingParses and renders a PDF using the PDFium library
Photo LibraryRuns image classification to categorize and tag photos based on the objects they contain
ClangCompiles a C source file using AArch64 as the target architecture
Text ProcessingLoads files, parses them with regex, stores metadata in SQLite databse, and exports content to different format
Asset CompressionCompresses 3D textural and geometric assets using a variety of popular compression codecs (ASTC, BC7, DXT5)
Object DetectionUses CNN to detect and classify objects in photo and then highlights them in photo
Background BlurSeparates background from foreground in a video stream and blurs the background
Image EditingMeasures how well you CPU handles making simple and complex image edits
Object RemoverRemoves an object from a photo and automatically fills in the gap left behind
Horizon DetectionSearches for the horizon line in an image and rotates the image to make horizon line level if found
Photo FilterApplies filters to photos to enhance their appearance
HDRTakes four SDR images and produces an HDR image
Image SynthesisMeasures how well your CPU handles creating artificial images
Ray TracerGenerates an image by tracing the path of light through an image plane and simulating the effects of its encounters with virtual objects
Structure from MotionTakes two 2D images of the same scene and constructs an estimate of the 3D coordinates of the points visible in both images

Workload Metrics

The following metrics are examples of those captured by the Virtual Client when running the GeekBench6 workload.

MetricNameMetricValueMetricUnit
SingleCore-File Compression269.8MB/sec
SingleCore-Navigation11.5routes/sec
SingleCore-HTML5 Browser41.7pages/sec
SingleCore-PDF Renderer47.7Mpixels/sec
SingleCore-Photo Library27images/sec
SingleCore-Clang9.71Klines/sec
SingleCore-Text Processing152.4pages/sec
SingleCore-Asset Compression62.3MB/sec
SingleCore-Object Detection59.1images/sec
SingleCore-Background Blur9.76images/sec
SingleCore-Horizon Detection84.5Mpixels/sec
SingleCore-Object Remover123.2Mpixels/sec
SingleCore-HDR55.8Mpixels/sec
SingleCore-Photo Filter25.3images/sec
SingleCore-Ray Tracer1.5Mpixels/sec
SingleCore-Structure from Motion68.7Kpixels/sec
SingleCoreScore-File Compression1878Score
SingleCoreScore-Navigation1901Score
SingleCoreScore-HTML5 Browser2038Score
SingleCoreScore-PDF Renderer2069Score
SingleCoreScore-Photo Library1988Score
SingleCoreScore-Clang1971Score
SingleCoreScore-Text Processing1903Score
SingleCoreScore-Asset Compression2012Score
SingleCoreScore-Object Detection1976Score
SingleCoreScore-Background Blur2358Score
SingleCoreScore-Horizon Detection2714Score
SingleCoreScore-Object Remover1602Score
SingleCoreScore-HDR1900Score
SingleCoreScore-Photo Filter2546Score
SingleCoreScore-Ray Tracer1550Score
SingleCoreScore-Structure from Motion2170Score
MultiCore-File Compression774MB/sec
MultiCore-Navigation44.2routes/sec
MultiCore-HTML5 Browser134.8pages/sec
MultiCore-PDF Renderer114.8Mpixels/sec
MultiCore-Photo Library46.2images/sec
MultiCore-Clang17.8Klines/sec
MultiCore-Text Processing145.2pages/sec
MultiCore-Asset Compression127.7MB/sec
MultiCore-Object Detection75.5images/sec
MultiCore-Background Blur13.8images/sec
MultiCore-Horizon Detection158Mpixels/sec
MultiCore-Object Remover248.5Mpixels/sec
MultiCore-HDR157.5Mpixels/sec
MultiCore-Photo Filter63.2images/sec
MultiCore-Ray Tracer4.91Mpixels/sec
MultiCore-Structure from Motion198Kpixels/sec
MultiCoreScore-File Compression5390Score
MultiCoreScore-Navigation7332Score
MultiCoreScore-HTML5 Browser6586Score
MultiCoreScore-PDF Renderer4978Score
MultiCoreScore-Photo Library3401Score
MultiCoreScore-Clang3615Score
MultiCoreScore-Text Processing1813Score
MultiCoreScore-Asset Compression4122Score
MultiCoreScore-Object Detection2524Score
MultiCoreScore-Background Blur3339Score
MultiCoreScore-Horizon Detection5076Score
MultiCoreScore-Object Remover3232Score
MultiCoreScore-HDR5368Score
MultiCoreScore-Photo Filter6371Score
MultiCoreScore-Ray Tracer5073Score
MultiCoreScore-Structure from Motion6254Score
SingleCoreSummary-Single-Core Score2007Score
SingleCoreSummary-Integer Score1970Score
SingleCoreSummary-Floating Point Score2077Score
MultiCoreSummary-Multi-Core Score4308Score
MultiCoreSummary-Integer Score4061Score
MultiCoreSummary-Floating Point Score4808Score