Measurement Cluster

Our Measurement Cluster#

Our Measurement Cluster runs on NOP Linux changes → and is the driver behind our Hosted Service.

We have the following machines available for running measurements in our cluster. The Machine ID is the number shown in the ID column of our Cluster Status page and is the identifier used in the API and when a job is dispatched to a specific machine. The live availability, queue state and detailed machine configuration of every machine can always be looked up there.

Machines marked [PREMIUM] are only available in our paid plans.

• Machine ID 7 — CO2 Profiling - Esprimo P956
  • Use Case: For profiling of a software to get a value for an off-the-shelf Ubuntu system with default configuration
  • Vendor: Fujitsu ESPRIMO P956
  • OS: Ubuntu 24.04 (NOP Linux)
  • Type: Single-Tenant Server
  • CPU: Intel(R) Core(TM) i5-6500 CPU @ 3.20GHz
  • Metrics Provider for Machine Power: MCP39F511N
  • Special: Blue Angel compatible for Server applications
  • ⚙️ Detailed machine specs and configuration like CPU Cores, Memory, TurboBoost etc. are always up to date on the (ℹ︎) icon in the Cluster Machine Listing

• Machine ID 14 — CO2 Benchmarking 2.0 - TX1330 M2 [PREMIUM]
  • Use Case: For benchmarking of a software where configuration is tuned for reproducability
  • Vendor: Fujitsu TX1330 M2
  • OS: Ubuntu 24.04 (NOP Linux)
  • Type: Single-Tenant Server
  • CPU: Intel(R) Xeon(R) CPU E3-1240L v5 @ 2.10GHz
  • Metrics Provider for Machine Power: MCP39F511N
  • Special: Blue Angel compatible for Client-Server and Server applications
  • This machine replaces the former CO2 Benchmarking machine (Machine ID 5), see Old Machines
  • ⚙️ Detailed machine specs and configuration like CPU Cores, Memory, TurboBoost etc. are always up to date on the (ℹ︎) icon in the Cluster Machine Listing

• Machine ID 6 — Micro Benchmarking - TX1330 M2 [PREMIUM]
  • Use Case: For micro benchmarking of a software where configuration is tuned for reproducability. Reporters are set to 2 ms sampling_rate and limited to RAPL CPU / Memory as well as network I/O per cgroup
  • Vendor: Fujitsu TX1330 M2
  • OS: Ubuntu 24.04 (NOP Linux)
  • Type: Single-Tenant Server
  • CPU: Intel(R) Xeon(R) CPU E3-1240L v5 @ 2.10GHz
  • Metrics Provider for Machine Power: None
  • ⚙️ Detailed machine specs and configuration like CPU Cores, Memory, TurboBoost etc. are always up to date on the (ℹ︎) icon in the Cluster Machine Listing

• Machine ID 12 — GUI Benchmarking - TX1330 M3 [PREMIUM]
  • Use Case: For benchmarking of a GUI / Desktop software that uses X11 or Wayland window management systems
  • Vendor: Fujitsu TX1330 M3
  • OS: Ubuntu 24.04 (default) with Wayland Window manager
  • Type: Single-Tenant Server
  • CPU: Intel(R) Xeon(R) CPU E3-1220 v6 @ 3.00GHz
  • Metrics Provider for Machine Power: MCP39F511N
  • Special: Blue Angel compatible for Client-Server, Server and Desktop applications
  • Special: External VGA Monitor connected to fully measure GPU Output overhead
  • ⚙️ Detailed machine specs and configuration like CPU Cores, Memory, TurboBoost etc. are always up to date on the (ℹ︎) icon in the Cluster Machine Listing

• Machine ID 15 — GUI High Performance Benchmarking - TX1330 M4 [PREMIUM]
  • Use Case: For benchmarking GUI / Desktop software that needs more cores, a higher clock rate or more memory than the TX1330 M3 and uses a virtual X window
  • Vendor: Fujitsu TX1330 M4
  • OS: Ubuntu 24.04 without active window manager but active X-Libraries for virtual / containerized X-Windows
  • Type: Single-Tenant Server
  • CPU: Intel(R) Xeon(R) E-2176G CPU @ 3.70GHz
  • Metrics Provider for Machine Power: MCP39F511N
  • ⚙️ Detailed machine specs and configuration like CPU Cores, Memory, TurboBoost etc. are always up to date on the (ℹ︎) icon in the Cluster Machine Listing

• Machine ID 11 — ML/AI Benchmarking - GTX-1080 - [PREMIUM]
  • Use Case: For GPU measurements and AI training measurements
  • OS: Ubuntu 24.04 (NOP Linux)
  • Type: Desktop-PC (MSI MS-7B98)
  • CPU: Intel(R) Core(TM) i5-9600K CPU @ 3.70GHz
  • Graphics: GeForce GTX 1080 (NVIDIA GP104)
    • VRAM: 8 GB
    • Installed NVIDIA kernel driver packages: 535.309.01 and 580.173.02
  • Metrics Provider for Machine Power: MCP39F511N
  • Metrics Provider for GPU Energy: NVIDIA NVML
  • Special: Blue Angel compatible for Server applications
  • ⚙️ Detailed machine specs and configuration like CPU Cores, Memory, TurboBoost etc. are always up to date on the (ℹ︎) icon in the Cluster Machine Listing

• Machine ID 13 — High Performance Benchmarking - TX1330 M4 [PREMIUM]
  • Use Case: For workloads with higher CPU count, higher clock frequency or high memory requirements
  • Vendor: Fujitsu TX1330 M4
  • OS: Ubuntu 24.04 (NOP Linux)
  • Type: Single-Tenant Server
  • CPU: Intel(R) Xeon(R) E-2176G CPU @ 3.70GHz
  • Metrics Provider for Machine Power: MCP39F511N
  • ⚙️ Detailed machine specs and configuration like CPU Cores, Memory, TurboBoost etc. are always up to date on the (ℹ︎) icon in the Cluster Machine Listing

Old Machines#

These machines are no longer available for measurements. They are listed here because older runs in our database still reference them.

• Machine ID 5 — CO2 Benchmarking (DVFS OFF, TB OFF, HT OFF) - TX1330 M2 - (Phased out Late 2025)
  • Use Case: For benchmarking of a software where configuration is tuned for reproducability
  • Vendor: Fujitsu TX1330 M2
  • OS: Ubuntu 24.04 (NOP Linux)
  • Type: Single-Tenant Server
  • CPU: Intel(R) Xeon(R) CPU E3-1240L v5 @ 2.10GHz
  • Cores: 4
  • Threads: 4
  • Hyper-Threading: Off
  • Turbo Boost: Off
  • DVFS: Off (Fixed to 2.1 GHz)
  • C-States: All
  • Memory: 8 GB
  • Sample measurement with machine specs
  • Metrics Provider for Machine Power: MCP39F511N
  • Replaced by CO2 Benchmarking 2.0 (Machine ID 14)

• Machine ID 8 — Palit GPU - (Phased out Early 2025)
  • Use Case: For GPU measurements and AI training measurements
  • OS: Ubuntu 22.04 (NOP Linux)
  • Type: Desktop-PC
  • CPU: Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz
  • Cores: 4
  • Threads: 8
  • Hyper-Threading: On
  • Turbo Boost: On
  • Memory: 12 GB
  • Graphics: Palit GF110 (GeForce GTX 570)
    • Installed CUDA version: 12
    • NVIDIA kernel driver version 390
  • Metrics Provider for Machine Power: MCP39F511N & IPMI
  • Replaced by ML/AI Profiling - GTX-1080 (Machine ID 11)

• Machine ID 2 — Quanta Leopard-DDR3 - (Phased out Early 2025)
  • Use Case: Heavy parallelized workloads / HPC
  • Type: 48-Thread Multi-Tenant Server
  • OS: Ubuntu 22.04 (default)
  • CPU: Intel(R) Xeon(R) CPU E5-2678 v3 @ 2.50GHz (2x)
  • Cores: 24
  • Threads: 48
  • Hyper-Threading: On
  • Turbo Boost: Off
  • Memory: 32 GB
  • Sample measurement with machine specs
  • Metrics Provider for Machine Power: IPMI
  • Special: SoftAWERE compatible

• Machine ID 3 — Fujitsu Esprimo P956 (Blue Angel compatible)
  • Predecessor of the current CO2 Profiling machine (Machine ID 7)

• Machine ID 10 — MacBook Pro 13" - 2015 (Premium)
  • Use Case: macOS measurements

How to choose#

Profiling Machines#

These machines are configured to be best representative of an off-the-shelf energy configuration.

This means:

• CPU Power Hungry but user friendly features like TurboBoost are on
• Low-Load multi-tasking optimizing features like HyperThreading are turned on
• Frequency Limits and C-State-Limits which typically make workloads more reproducible for all tenants (and thus are typically activated in the cloud) are not set.

You should choose these kind of machines to get an idea of how much actual energy an application might be using on an off-the-shelf installation. This energy value is thus very representative for desktop, home user or shared hosting situations.

It is also very good to show how good your software can leverage sleep states or how good it mitigates wakeups and thus highlight it’s power saving features.

Benchmarking Machines#

These machines are configured to increase reproducability by turning off some variance introducing features but still allow a reasonable difference between low load and high load scenarios on the machine

• Strongly non linear features like TurboBoost and DVFS are turned off
• Frequency is limited to only the base clock rate of the CPU
• C-States are limited to C1 reduce latency from wakeups, but still let CPU not stay in a spin-lock

You should choose these machine types if you want have strongly reproducible runs that are not impacted by home user configurations like power saving features. This also corresponds to configurations of most cloud vendors and thus are more represeantative of a cloud workload.

Also consider to choose a machine that has the metric providers with sampling rates apt for the effect you want to capture. See Sampling Rate Best Practicesfor more details.

Drawbacks are that it cannot show the full potential how good an application leverages power saving features.

HPC Benchmarking#

These machines are configured for extreme compute scenarios effectively turning off all latency introducing features completely. Optionally some compute power improvements features like Pre-Fetching can be enabled

HPC Benchmarking machines are only available in the enterprise plan as the must be configured to the users request. If you plan to make HPC Benchmarking on our cluster please contact us and send us your desired CPU, DRAM and GPU configuration and power capping settings.

Setting up your own measurement cluster#

Please refer to the page Installation of a Cluster