Meadow_Performance_Benchmarks

Contains applications that measure the performance of Meadow, as well as the results of those tests.

This is primarily to track the progress of our performance improvements.

To validate, make sure to run application in release mode.

Please Note - Small variances will occur between runs and between boards.

Benchmark results

Below this markdown you will see the kinds of tests we perform on our boards to benchmark its performance for each release.

• Pi Calculation
• List Benchmark
• Digital Output Operations
• Soft PWM Generation

Pi Calculation

Calculates pi (3.14159..) to [x] digits and records the amount of time (in seconds) it takes. Here are the results per release:

Operation	b3.5	b3.6	b3.7	b6.0.1	b6.3	RC1	RC1 w/ JIT	RC-2	RC-3	3.0-preview
Calculate 50 digits	11s	11s	2.3s	1.8s	1.7s	2.3s	2.3s	2.2s	2.5s	2.1s
Calculate 100 digits	54s	54s	10.8s	8.9s	8.3s	11.2s	11.1s	10.1s	11.9s	9.9s
Calculate 150 digits	127s	126s	24.4s	20.6s	19.3s	26.3s	25.8s	23.4s	27.9s	23.0s

List Benchmark

List Operations

Create a 1,000 item List<int>, and perform basic list operations on it.

Operation	b3.5	b3.6	b3.7	b4.3	b5.1	b6.0.1	b6.3	RC1	RC1+JIT	RC-2	RC-3	3.0-preview
Instantiation	30ms	30ms	30ms	19ms	10ms	11ms	11ms	5ms	13ms	14ms	15ms	14ms
Population	120ms	120ms	20ms	44ms	20ms	21ms	10ms	21ms	28ms	29ms	31ms	30ms
Summation	130ms	120ms	30ms	21ms	19ms	19ms	11ms	19ms	26ms	26ms	29ms	22ms

List Clearing

Clear the 1,000 item list by removing an item one at a time:

Operation	b3.5	b3.6	b3.7	b4.3	b5.1	b6.0.1	b6.3	RC1	RC1+JIT	RC-2	RC-3	3.0-preview
Clearing	59000ms	59000ms	7700ms	6100ms	7100ms	9000ms	8800ms	7500ms	115ms	130ms	134ms	114ms

Digital Output Port Operations

Digital Port initialization

Initializes three DigitalOutputPort instances, one for each of the onboard LED components and writes to them.

Operation	b3.5	b3.6	b3.7	b4.3	b5.1	b6.0.1	b6.3	RC1	RC1+JIT	RC-2	RC-3	3.0-preview
Port initialization	2700ms	2800ms	2000ms	500ms	450ms	460ms	470ms	480ms	740ms	704ms	770ms	1413ms

Digital Port Writes

Alternate states to 3 Digital Output ports in 100 cycles.

Operation	b3.5	b3.6	b3.7	b4.3	b5.1	b6.0.1	b6.3	RC1	RC1 w/ JIT	RC-2	RC-3	3.0-preview	3.0-preview (raw BSRR)
300 writes	48000ms 13000ms	150ms	1400ms	1330ms	150ms	140ms	140ms	50ms	50ms	54ms	5ms	1ms

NOTE: Due to the drastic change of performance for this benchmark going from 4800 milliseconds (or 4.8 seconds) to 50 milliseconds (or 0.05 seconds), the graph below is expressed on a logarithmic scale.

Digital Port Average Time Per Write

Average time calculated between creating and writing on the digital output ports.

Operation	b3.5	b3.6	b3.7	b4.3	b5.1	b6.0.1	b6.3	RC1	RC1 w/ JIT	RC-2	RC-3	3.0-preview	3.0-preview (raw BSRR)
Avg time per write	159ms	42ms	0.5ms	0.5ms	0.44ms	0.51ms	0.46ms	0.47ms	0.16ms	0.17ms	0.18ms	0.017ms	0.003ms

NOTE: Due to the drastic change of performance for this benchmark going from 159 milliseconds to 0.17 milliseconds, the graph below is expressed on a logarithmic scale.

Soft PWM Generation

Generates a PWM signal in software. Currently is a visual test. Run the test and note which was the last frequency change that was noticeable.

Operation	b3.5	b3.6	b3.7	b4.3	b5.1	b6.0.1	b6.3	RC1	RC1+JIT	3.0-preview
Maximum Frequency @ 50% duty	~50hz	~50hz	~100hz	~100hz	~100hz	~100hz	~100hz	~100Hz	~1500Hz	TBD

Methodology Notes

3.0-preview: .NET 10 Mono runtime (Meadow.OS 2.999.2.2) on F7CoreComputeV2 in a ProjectLab V5, Release build of both the native runtime and the managed app. The benchmark project was bumped from Meadow.Sdk/1.1.0 / netstandard2.1 / App<F7FeatherV1> to Microsoft.NET.Sdk / net10.0 / App<F7CoreComputeV2>, and the three OnboardLedRed/Green/Blue pin references in the Digital Output and Soft PWM tests were retargeted to free F7CC pins (PA0, PA3, PA9); the test logic is otherwise unchanged. Soft PWM max frequency requires visual / scope inspection — _TBD_ until measured. The dedicated GPIO microbenchmark in the Meadow repo (F7DigitalOutputPort fast path) reports ~0.26µs per State = write in a tight 200k-iteration loop; this benchmark's 0.017ms (≈17µs) per write reflects the same path under the README's short 300-iteration / 3-distinct-port test pattern, which has less JIT warm-up and fewer inlining opportunities.

3.0-preview (raw BSRR): Same 100×3 alternation as the standard "300 writes" measurement, but written through F7DigitalOutputPort.GetRawWriteHandle(out uint* bsrr, out uint setMask, out uint clearMask) — a new escape hatch in Meadow.Core that hands back a pointer to the STM32 BSRR register and the set/clear masks for the pin. The hot loop becomes three back-to-back register stores per iteration with no virtual dispatch and no _state field maintenance. Stopwatch's ~1ms resolution caps measurement at the low end; the 200k-iteration GPIO microbenchmark in the Meadow repo resolves the raw path at closer to 45ns/write (≈22 MHz square wave when unrolled). The relative ~5× advantage over the managed State = path on the same hardware with the same loop shape is the headline; the absolute number is bounded by timer granularity.