Integrate RAPL energy usage measurements

[ta1H3n]

This could be a great addition to this framework if energy usage measurements could be included in the benchmarks. Energy efficiency is becoming more relevant, and it is known that energy usage does not always correlate with execution time.

Recent research suggests that RAPL provides good enough measurements:

Our observations reveal that RAPL readings are highly correlated with plug power, promisingly accurate enough, and have negligible performance overhead. Experimental results suggest RAPL can be a very useful tool to measure and monitor the energy consumption of servers without deploying any complex power meters.

https://dl.acm.org/doi/abs/10.1145/3177754

[wild0ne]

@AndreyAkinshin Hi!

Frankly, the feature looks quite interesting. Would you be so kind as to share a couple of ideas regarding PoC? What might the minimum results look like? I assume that RAPL counters might be useless if Frequency and Time (and certain scalar values, of course) are enough to calculate power consumption.

[AndreyAkinshin]

Hey @wild0ne, thanks for the interest! Here are some thoughts on a possible PoC.

Why RAPL counters matter

Frequency and time alone are not sufficient to reliably estimate power consumption. Two code paths can run at the same frequency for the same duration but have very different energy profiles — e.g., AVX-heavy workloads vs simple integer arithmetic. Modern CPUs also apply dynamic voltage/frequency scaling (DVFS), selectively power-gate execution units, and transition through various C-states — all of which affect real consumption in ways that a simple formula cannot capture. RAPL provides energy readings via hardware counters that are far more accurate than any frequency/time-based estimation — especially on modern hardware (Haswell+), where RAPL uses integrated voltage regulators for direct measurement. It also covers DRAM energy, which cannot be derived from CPU frequency or timing data.

What the output could look like

The whole point is to surface cases where faster ≠ more efficient:

| Method | Mean     | Energy/Op |
|--------|----------|-----------|
| Foo    | 1.234 ms | 0.045 J   |
| Bar    | 1.198 ms | 0.071 J   |

Bar is slightly faster but consumes ~58% more energy — this kind of insight is invisible with time-only measurements.

Package structure

This should live in a separate package (e.g., BenchmarkDotNet.Diagnostics.Energy) to keep the core lightweight and avoid pulling in platform-specific dependencies for users who don't need this.

Minimum PoC scope

• A new EnergyDiagnoser (similar to MemoryDiagnoser) that reads RAPL energy counters before and after benchmark execution and reports the delta.
• Additional columns in the summary table: Energy/Op (J) for the CPU package domain, optionally DRAM.
• The diagnoser should be attachable via an attribute ([EnergyDiagnoser]) or config.

One important design consideration: RAPL counters update roughly every ~1 ms, so for very short benchmarks (nanosecond-scale), measuring individual iterations won't be meaningful. The diagnoser would need to measure energy across a batch of iterations and compute per-operation values — similar to how BDN already handles timing.

OS-specific implementations

Energy measurement APIs differ across platforms, so the implementation needs a platform abstraction layer with OS-specific backends:

• Linux: read from /sys/class/powercap/intel-rapl/. Note that since the Platypus side-channel attack (CVE-2020-8694), many distros restrict access to these files — root privileges or adjusted file permissions may be required.
• Windows: requires MSR access via a kernel driver (MSR reads are ring-0 only).
• macOS: lower priority for the PoC; energy APIs on macOS are platform-specific and largely undocumented.

Both Intel (Sandy Bridge+) and AMD (Zen and later) support RAPL, so the implementation should not be Intel-specific.

The diagnoser should detect at runtime whether energy measurement is supported on the current OS/hardware combination and gracefully degrade (skip energy columns with a warning) when it's not available. This follows the same pattern we use for other platform-dependent diagnosers.

For the PoC, starting with Linux would be the simplest path — the powercap sysfs interface is the most straightforward to work with.

[wild0ne]

Hey @wild0ne, thanks for the interest! Here are some thoughts on a possible PoC.

Why RAPL counters matter

Frequency and time alone are not sufficient to reliably estimate power consumption. Two code paths can run at the same frequency for the same duration but have very different energy profiles — e.g., AVX-heavy workloads vs simple integer arithmetic. Modern CPUs also apply dynamic voltage/frequency scaling (DVFS), selectively power-gate execution units, and transition through various C-states — all of which affect real consumption in ways that a simple formula cannot capture. RAPL provides energy readings via hardware counters that are far more accurate than any frequency/time-based estimation — especially on modern hardware (Haswell+), where RAPL uses integrated voltage regulators for direct measurement. It also covers DRAM energy, which cannot be derived from CPU frequency or timing data.

What the output could look like

The whole point is to surface cases where faster ≠ more efficient:

| Method | Mean     | Energy/Op |
|--------|----------|-----------|
| Foo    | 1.234 ms | 0.045 J   |
| Bar    | 1.198 ms | 0.071 J   |

Bar is slightly faster but consumes ~58% more energy — this kind of insight is invisible with time-only measurements.

Package structure

This should live in a separate package (e.g., BenchmarkDotNet.Diagnostics.Energy) to keep the core lightweight and avoid pulling in platform-specific dependencies for users who don't need this.

Minimum PoC scope

• A new EnergyDiagnoser (similar to MemoryDiagnoser) that reads RAPL energy counters before and after benchmark execution and reports the delta.
• Additional columns in the summary table: Energy/Op (J) for the CPU package domain, optionally DRAM.
• The diagnoser should be attachable via an attribute ([EnergyDiagnoser]) or config.

One important design consideration: RAPL counters update roughly every ~1 ms, so for very short benchmarks (nanosecond-scale), measuring individual iterations won't be meaningful. The diagnoser would need to measure energy across a batch of iterations and compute per-operation values — similar to how BDN already handles timing.

OS-specific implementations

Energy measurement APIs differ across platforms, so the implementation needs a platform abstraction layer with OS-specific backends:

• Linux: read from /sys/class/powercap/intel-rapl/. Note that since the Platypus side-channel attack (CVE-2020-8694), many distros restrict access to these files — root privileges or adjusted file permissions may be required.
• Windows: requires MSR access via a kernel driver (MSR reads are ring-0 only).
• macOS: lower priority for the PoC; energy APIs on macOS are platform-specific and largely undocumented.

Both Intel (Sandy Bridge+) and AMD (Zen and later) support RAPL, so the implementation should not be Intel-specific.

The diagnoser should detect at runtime whether energy measurement is supported on the current OS/hardware combination and gracefully degrade (skip energy columns with a warning) when it's not available. This follows the same pattern we use for other platform-dependent diagnosers.

For the PoC, starting with Linux would be the simplest path — the powercap sysfs interface is the most straightforward to work with.

Thanks!

Let me dig into it if that's ok

[wild0ne]

Hi @AndreyAkinshin !

I've made some code, and here is a quick update along with early results

The energy diagnoser works exactly as memory one does - via extra iteration

It reads full power package - /sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:0/energy_uj

Here is my test code:

Details

//...

    [EnergyDiagnoser]
    public class EnergyDemo
    {

//...
        [Benchmark]
        [MinIterationCount(1)]
        [MaxIterationCount(5)]
        public void Sqrt_10()
        {
            Random r1 = new Random();
            double d;
            for (int i = 0; i < 10; i++)
                d = Math.Sqrt(r1.NextDouble());
        }

        [Benchmark]
        [MinIterationCount(1)]
        [MaxIterationCount(5)]
        public void Sqrt_10_second()
        {
            Random r1 = new Random();
            double d;
            for (int i = 0; i < 10; i++)
                d = Math.Sqrt(r1.NextDouble());
        }

// ...

        [Benchmark]
        [MinIterationCount(1)]
        [MaxIterationCount(5)]
        public void Sqrt_2_10p8()
        {
            Random r1 = new Random();
            double d;
            for (long i = 0; i < 2 * 100_000_000; i++)
                d = Math.Sqrt(r1.NextDouble());
        }

        [Benchmark]
        [MinIterationCount(1)]
        [MaxIterationCount(5)]
        public void Sqrt_2_10p8_second()
        {
            Random r1 = new Random();
            double d;
            for (long i = 0; i < 2 * 100_000_000; i++)
                d = Math.Sqrt(r1.NextDouble());
        }

// ...

I run it on Linux Mint with high performance energy profile enabled

Here are the results:

// * Summary *

BenchmarkDotNet v0.16.0-develop (2026-02-23), Linux Linux Mint 22.3 (Zena)
Intel Core i7-4710HQ CPU 2.50GHz (Max: 3.29GHz) (Haswell), 1 CPU, 8 logical and 4 physical cores
.NET SDK 10.0.103
  [Host]     : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3
  Job-FSAAOM : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3

MaxIterationCount=5  MinIterationCount=1  

| Method             | Mean             | Error             | StdDev           | Energy/Op   |
|------------------- |-----------------:|------------------:|-----------------:|------------:|
| Sqrt_10            |         649.9 ns |          23.16 ns |          6.01 ns |    0.006122 |
| Sqrt_10_second     |         652.2 ns |           7.95 ns |          1.23 ns |    0.006077 |
| Sqrt_10_third      |         651.0 ns |           7.50 ns |          1.95 ns |    0.005975 |
| Sqrt_10_4th        |         644.3 ns |           6.16 ns |          0.95 ns |    0.005888 |
| Sqrt_10_5th        |         648.2 ns |           4.08 ns |          0.22 ns |    0.005985 |
| Sqrt_2_10p8        | 693,869,948.7 ns |   6,732,541.99 ns |    369,033.35 ns | 8538.797000 |
| Sqrt_2_10p8_second | 694,280,136.2 ns | 195,880,899.09 ns | 50,869,647.56 ns | 5569.322000 |
| Sqrt_2_10p8_third  | 693,049,041.3 ns |  13,067,746.67 ns |    716,287.29 ns | 8656.533000 |
| Sqrt_2_10p8_4th    | 693,932,424.2 ns |   7,253,614.13 ns |  1,122,504.81 ns | 8589.150000 |
| Sqrt_2_10p8_5th    | 578,094,217.0 ns |   3,539,756.82 ns |    547,781.28 ns | 5552.110000 |

// * Hints *
Outliers
  EnergyDemo.Sqrt_2_10p8_second: MaxIterationCount=5, MinIterationCount=1 -> 1 outlier  was  detected (606.44 ms)
  EnergyDemo.Sqrt_2_10p8_5th: MaxIterationCount=5, MinIterationCount=1    -> 1 outlier  was  removed (693.33 ms)

// * Legends *
  Mean      : Arithmetic mean of all measurements
  Error     : Half of 99.9% confidence interval
  StdDev    : Standard deviation of all measurements
  Energy/Op : Energy consumption per operation
  1 ns      : 1 Nanosecond (0.000000001 sec)

// * Diagnostic Output - EnergyDiagnoser *


// ***** BenchmarkRunner: End *****
Run time: 00:02:48 (168.17 sec), executed benchmarks: 10

Global total time: 00:03:39 (219.85 sec), executed benchmarks: 10
// * Artifacts cleanup *
Artifacts cleanup is finished

What I'm questioning here is the reliability of energy/op measures

As you can see, it deviates about 3000 Joules per op, which doesn't look good to me

I'm going to see what the deviations of more specific energy packages like

Hopefully, I will provide more detailed results later this week

[timcassell]

Don't use new Random() in your benchmarks. Use a constant seed, like new Random(12345).

[wild0ne]

Don't use new Random() in your benchmarks. Use a constant seed, like new Random(12345).

thanks, will try it

[wild0ne]

Hey @AndreyAkinshin @timcassell

Please have a look at the draft PR #3032

Results:

namespace demo
{
    [EnergyDiagnoser(EnergyCountersSetup.All)]
    public class EnergyDemo
    {
        [Params(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)]
        public int RunNumber { get; set; }

        [Benchmark]
        public void Sqrt_10()
        {
            double d;
            for (int i = 0; i < 10; i++)
                d = Math.Sqrt(i + 0.1);
        }

        [Benchmark]
        public void Sqrt_1_10p9()
        {
            double d;
            for (long i = 0; i < 1 * 1000_000_000; i++)
                d = Math.Sqrt(i + 0.1);
        }
    }
}

// * Summary *

BenchmarkDotNet v0.16.0-develop (2026-02-27), Linux Linux Mint 22.3 (Zena)
Intel Core i7-4710HQ CPU 2.50GHz (Max: 3.19GHz) (Haswell), 1 CPU, 8 logical and 4 physical cores
.NET SDK 10.0.103
  [Host]     : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3
  DefaultJob : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3


| Method      | RunNumber | Mean               | Error              | StdDev             | Median             | EC package-0     | EC core          | EC uncore      | EC dram        |
|------------ |---------- |-------------------:|-------------------:|-------------------:|-------------------:|-----------------:|-----------------:|---------------:|---------------:|
| Sqrt_10     | 1         |           5.824 ns |          0.0090 ns |          0.0080 ns |           5.824 ns |       000.098 uj |       000.050 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 1         | 290,629,914.000 ns |  2,315,503.1699 ns |  1,933,550.0170 ns | 289,919,152.000 ns | 4,607,033.118 uj | 2,239,891.176 uj | 240,639.471 uj | 497,542.941 uj |
| Sqrt_10     | 2         |           5.834 ns |          0.0074 ns |          0.0069 ns |           5.836 ns |       000.097 uj |       000.049 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 2         | 294,074,720.267 ns |  3,112,004.8359 ns |  2,910,971.2557 ns | 294,091,916.000 ns | 4,329,186.100 uj | 2,067,383.633 uj | 166,381.433 uj | 468,366.333 uj |
| Sqrt_10     | 3         |           5.822 ns |          0.0151 ns |          0.0142 ns |           5.823 ns |       000.096 uj |       000.048 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 3         | 291,639,795.042 ns |  2,226,694.3968 ns |  1,738,456.5197 ns | 291,312,569.500 ns | 4,355,567.467 uj | 2,082,412.467 uj | 152,487.800 uj | 472,978.533 uj |
| Sqrt_10     | 4         |           5.829 ns |          0.0054 ns |          0.0048 ns |           5.830 ns |       000.096 uj |       000.048 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 4         | 326,603,440.320 ns | 14,178,641.9401 ns | 41,806,048.6983 ns | 301,922,308.500 ns | 5,071,500.680 uj | 2,450,680.980 uj | 262,561.580 uj | 540,364.830 uj |
| Sqrt_10     | 5         |           5.826 ns |          0.0129 ns |          0.0121 ns |           5.828 ns |       000.097 uj |       000.049 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 5         | 292,196,238.544 ns |  1,901,212.2995 ns |  4,133,079.3821 ns | 291,647,892.000 ns | 4,749,226.120 uj | 2,269,188.533 uj | 254,688.480 uj | 517,466.933 uj |
| Sqrt_10     | 6         |           5.816 ns |          0.0123 ns |          0.0115 ns |           5.816 ns |       000.096 uj |       000.048 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 6         | 291,191,755.467 ns |  1,049,194.9864 ns |    981,417.6417 ns | 291,036,138.000 ns | 4,222,946.567 uj | 1,987,794.000 uj | 148,903.033 uj | 466,974.733 uj |
| Sqrt_10     | 7         |           5.824 ns |          0.0075 ns |          0.0067 ns |           5.826 ns |       000.097 uj |       000.049 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 7         | 291,461,637.067 ns |  1,837,885.7532 ns |  1,719,159.4746 ns | 291,461,540.000 ns | 4,280,911.467 uj | 2,030,276.400 uj | 164,959.267 uj | 467,338.900 uj |
| Sqrt_10     | 8         |           5.812 ns |          0.0134 ns |          0.0125 ns |           5.815 ns |       000.097 uj |       000.049 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 8         | 290,649,540.214 ns |    361,824.5175 ns |    320,748.0105 ns | 290,681,584.250 ns | 4,233,206.533 uj | 2,022,323.500 uj | 147,423.933 uj | 463,684.900 uj |
| Sqrt_10     | 9         |           5.817 ns |          0.0111 ns |          0.0099 ns |           5.818 ns |       000.097 uj |       000.049 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 9         | 292,170,748.508 ns |  2,963,256.9501 ns |  6,566,373.2089 ns | 290,586,349.000 ns | 4,700,289.893 uj | 2,245,436.960 uj | 248,135.760 uj | 510,728.667 uj |
| Sqrt_10     | 10        |           5.828 ns |          0.0157 ns |          0.0147 ns |           5.829 ns |       000.097 uj |       000.049 uj |     000.003 uj |     000.009 uj |
| Sqrt_1_10p9 | 10        | 295,813,397.679 ns |  3,057,943.2574 ns |  2,710,787.0484 ns | 295,632,190.500 ns | 4,353,919.500 uj | 2,066,346.033 uj | 165,185.133 uj | 473,171.800 uj |

// * Hints *
Outliers
  EnergyDemo.Sqrt_10: Default     -> 1 outlier  was  removed, 2 outliers were detected (5.80 ns, 5.84 ns)
  EnergyDemo.Sqrt_1_10p9: Default -> 4 outliers were removed (306.76 ms..313.15 ms)
  EnergyDemo.Sqrt_1_10p9: Default -> 3 outliers were removed (305.42 ms..319.16 ms)
  EnergyDemo.Sqrt_10: Default     -> 1 outlier  was  removed, 2 outliers were detected (5.82 ns, 5.84 ns)
  EnergyDemo.Sqrt_1_10p9: Default -> 18 outliers were removed (339.92 ms..409.61 ms)
  EnergyDemo.Sqrt_10: Default     -> 1 outlier  was  removed (5.87 ns)
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  removed, 2 outliers were detected (289.86 ms, 291.74 ms)
  EnergyDemo.Sqrt_10: Default     -> 1 outlier  was  removed, 3 outliers were detected (5.79 ns, 5.80 ns, 5.83 ns)
  EnergyDemo.Sqrt_1_10p9: Default -> 16 outliers were removed (328.42 ms..399.16 ms)
  EnergyDemo.Sqrt_10: Default     -> 2 outliers were detected (5.80 ns, 5.80 ns)
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  removed (308.46 ms)

// * Legends *
  RunNumber    : Value of the 'RunNumber' parameter
  Mean         : Arithmetic mean of all measurements
  Error        : Half of 99.9% confidence interval
  StdDev       : Standard deviation of all measurements
  Median       : Value separating the higher half of all measurements (50th percentile)
  EC package-0 : Average energy consumption of unit package-0, uj/op
  EC core      : Average energy consumption of unit core, uj/op
  EC uncore    : Average energy consumption of unit uncore, uj/op
  EC dram      : Average energy consumption of unit dram, uj/op
  1 ns         : 1 Nanosecond (0.000000001 sec)

// * Diagnostic Output - EnergyDiagnoser *


// ***** BenchmarkRunner: End *****
Run time: 00:08:12 (492.06 sec), executed benchmarks: 20

[wild0ne]

I've run it on two other AMD notebooks

AMD Ryzen 5 5500U

Details

// * Summary *

BenchmarkDotNet v0.16.0-develop (2026-03-06), Linux Linux Mint 22.3 (Zena)
AMD Ryzen 5 5500U with Radeon Graphics 1.11GHz, 1 CPU, 12 logical and 6 physical cores
.NET SDK 10.0.103
  [Host]     : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3
  DefaultJob : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3


| Method      | RunNumber | Mean               | Error           | StdDev          | EC package-0     | EC core        |
|------------ |---------- |-------------------:|----------------:|----------------:|-----------------:|---------------:|
| Sqrt_10     | 1         |           4.142 ns |       0.0464 ns |       0.0434 ns |       000.039 uj |     000.029 uj |
| Sqrt_1_10p9 | 1         | 248,765,878.867 ns | 224,719.7402 ns | 210,202.9845 ns | 1,673,246.889 uj |   4,619.111 uj |
| Sqrt_10     | 2         |           3.614 ns |       0.0262 ns |       0.0245 ns |       000.033 uj |     000.000 uj |
| Sqrt_1_10p9 | 2         | 249,014,235.786 ns | 371,423.2799 ns | 329,257.0633 ns | 1,606,135.400 uj | 470,252.067 uj |
| Sqrt_10     | 3         |           4.337 ns |       0.0095 ns |       0.0088 ns |       000.037 uj |     000.000 uj |
| Sqrt_1_10p9 | 3         | 248,853,929.923 ns | 108,923.2373 ns |  90,955.8363 ns | 1,665,969.822 uj |  13,501.644 uj |
| Sqrt_10     | 4         |           3.610 ns |       0.0017 ns |       0.0016 ns |       000.032 uj |     000.000 uj |
| Sqrt_1_10p9 | 4         | 248,823,242.800 ns | 220,766.9751 ns | 206,505.5656 ns | 1,669,444.911 uj | 896,906.600 uj |
| Sqrt_10     | 5         |           3.818 ns |       0.0229 ns |       0.0214 ns |       000.033 uj |     000.000 uj |
| Sqrt_1_10p9 | 5         | 248,586,908.310 ns |  66,917.8634 ns |  59,320.9429 ns | 1,661,998.333 uj |   8,803.533 uj |
| Sqrt_10     | 6         |           3.669 ns |       0.0047 ns |       0.0042 ns |       000.032 uj |     000.000 uj |
| Sqrt_1_10p9 | 6         | 248,804,050.089 ns | 109,523.9653 ns | 102,448.7851 ns | 1,658,491.378 uj |  10,153.333 uj |
| Sqrt_10     | 7         |           3.777 ns |       0.0691 ns |       0.0647 ns |       000.031 uj |     000.000 uj |
| Sqrt_1_10p9 | 7         | 248,629,874.846 ns |  24,456.0714 ns |  20,421.9272 ns | 1,645,252.844 uj |   6,711.156 uj |
| Sqrt_10     | 8         |           3.936 ns |       0.0554 ns |       0.0518 ns |       000.033 uj |     000.000 uj |
| Sqrt_1_10p9 | 8         | 248,504,381.524 ns |  70,676.6063 ns |  62,652.9706 ns | 1,614,178.400 uj |   9,067.333 uj |
| Sqrt_10     | 9         |           4.129 ns |       0.0563 ns |       0.0526 ns |       000.034 uj |     000.000 uj |
| Sqrt_1_10p9 | 9         | 248,551,905.905 ns |  23,591.0506 ns |  20,912.8519 ns | 1,653,796.644 uj |   2,658.289 uj |
| Sqrt_10     | 10        |           3.818 ns |       0.0479 ns |       0.0448 ns |       000.031 uj |     000.024 uj |
| Sqrt_1_10p9 | 10        | 248,566,699.311 ns |  51,308.8887 ns |  47,994.3664 ns | 1,628,683.333 uj | 112,076.111 uj |

// * Hints *
Outliers
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  removed, 2 outliers were detected (248.33 ms, 251.36 ms)
  EnergyDemo.Sqrt_1_10p9: Default -> 2 outliers were removed, 3 outliers were detected (248.68 ms, 249.14 ms, 249.16 ms)
  EnergyDemo.Sqrt_10: Default     -> 1 outlier  was  detected (3.76 ns)
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  removed (248.84 ms)
  EnergyDemo.Sqrt_10: Default     -> 1 outlier  was  removed (3.69 ns)
  EnergyDemo.Sqrt_1_10p9: Default -> 2 outliers were removed (248.73 ms, 248.83 ms)
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  removed (248.72 ms)
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  removed (248.64 ms)

// * Legends *
  RunNumber    : Value of the 'RunNumber' parameter
  Mean         : Arithmetic mean of all measurements
  Error        : Half of 99.9% confidence interval
  StdDev       : Standard deviation of all measurements
  EC package-0 : Average energy consumption of unit package-0, uj/op
  EC core      : Average energy consumption of unit core, uj/op
  1 ns         : 1 Nanosecond (0.000000001 sec)

// * Diagnostic Output - EnergyDiagnoser *


// ***** BenchmarkRunner: End *****
Run time: 00:07:31 (451.45 sec), executed benchmarks: 20

AMD Ryzen 7 7730U

Details

// * Summary *

BenchmarkDotNet v0.16.0-develop (2026-03-06), Linux Linux Mint 22.3 (Zena)
AMD Ryzen 7 7730U with Radeon Graphics 1.09GHz, 1 CPU, 16 logical and 8 physical cores
.NET SDK 10.0.103
  [Host]     : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3
  DefaultJob : .NET 10.0.3 (10.0.3, 10.0.326.7603), X64 RyuJIT x86-64-v3


| Method      | RunNumber | Mean               | Error           | StdDev          | EC package-0     | EC core      |
|------------ |---------- |-------------------:|----------------:|----------------:|-----------------:|-------------:|
| Sqrt_10     | 1         |           3.007 ns |       0.0024 ns |       0.0022 ns |       000.022 uj |   000.000 uj |
| Sqrt_1_10p9 | 1         | 223,690,613.956 ns | 125,543.3504 ns | 117,433.3280 ns |   944,252.178 uj | 1,507.156 uj |
| Sqrt_10     | 2         |           3.005 ns |       0.0029 ns |       0.0027 ns |       000.020 uj |   000.000 uj |
| Sqrt_1_10p9 | 2         | 223,669,275.905 ns | 159,931.2881 ns | 141,774.9212 ns |   957,852.133 uj | 6,794.556 uj |
| Sqrt_10     | 3         |           3.005 ns |       0.0021 ns |       0.0019 ns |       000.021 uj |   000.000 uj |
| Sqrt_1_10p9 | 3         | 223,611,638.711 ns | 130,980.4280 ns | 122,519.1737 ns |   982,299.533 uj | 6,051.667 uj |
| Sqrt_10     | 4         |           3.004 ns |       0.0026 ns |       0.0024 ns |       000.021 uj |   000.000 uj |
| Sqrt_1_10p9 | 4         | 223,664,302.933 ns | 208,998.2568 ns | 195,497.0992 ns | 1,036,044.978 uj | 4,161.022 uj |
| Sqrt_10     | 5         |           3.004 ns |       0.0023 ns |       0.0022 ns |       000.021 uj |   000.000 uj |
| Sqrt_1_10p9 | 5         | 223,637,848.444 ns | 193,319.8587 ns | 180,831.5159 ns |   978,753.578 uj | 2,073.733 uj |
| Sqrt_10     | 6         |           3.004 ns |       0.0021 ns |       0.0019 ns |       000.021 uj |   000.000 uj |
| Sqrt_1_10p9 | 6         | 223,774,844.133 ns | 166,763.6036 ns | 155,990.7783 ns |   984,577.044 uj | 7,006.822 uj |
| Sqrt_10     | 7         |           3.005 ns |       0.0027 ns |       0.0026 ns |       000.021 uj |   000.000 uj |
| Sqrt_1_10p9 | 7         | 223,589,124.800 ns | 188,017.9964 ns | 175,872.1505 ns |   929,321.378 uj | 2,448.067 uj |
| Sqrt_10     | 8         |           3.004 ns |       0.0027 ns |       0.0025 ns |       000.020 uj |   000.000 uj |
| Sqrt_1_10p9 | 8         | 223,613,591.289 ns | 183,089.0963 ns | 171,261.6544 ns |   936,206.800 uj | 1,967.956 uj |
| Sqrt_10     | 9         |           3.005 ns |       0.0024 ns |       0.0023 ns |       000.021 uj |   000.000 uj |
| Sqrt_1_10p9 | 9         | 223,609,985.689 ns | 223,858.8281 ns | 209,397.6868 ns |   952,351.800 uj | 4,520.089 uj |
| Sqrt_10     | 10        |           3.005 ns |       0.0023 ns |       0.0022 ns |       000.021 uj |   000.000 uj |
| Sqrt_1_10p9 | 10        | 223,657,827.200 ns | 143,356.8328 ns | 134,096.0704 ns |   937,605.800 uj | 3,142.822 uj |

// * Hints *
Outliers
  EnergyDemo.Sqrt_1_10p9: Default -> 2 outliers were detected (223.43 ms, 223.44 ms)
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  removed, 2 outliers were detected (223.32 ms, 224.12 ms)
  EnergyDemo.Sqrt_1_10p9: Default -> 2 outliers were detected (223.32 ms, 223.36 ms)
  EnergyDemo.Sqrt_1_10p9: Default -> 1 outlier  was  detected (223.41 ms)
  EnergyDemo.Sqrt_10: Default     -> 2 outliers were detected (3.00 ns, 3.00 ns)

// * Legends *
  RunNumber    : Value of the 'RunNumber' parameter
  Mean         : Arithmetic mean of all measurements
  Error        : Half of 99.9% confidence interval
  StdDev       : Standard deviation of all measurements
  EC package-0 : Average energy consumption of unit package-0, uj/op
  EC core      : Average energy consumption of unit core, uj/op
  1 ns         : 1 Nanosecond (0.000000001 sec)

// * Diagnostic Output - EnergyDiagnoser *


// ***** BenchmarkRunner: End *****
Run time: 00:07:59 (479.13 sec), executed benchmarks: 20