Async machinery benchmarks

BENCHMARKS · ASYNC MACHINERY

Task vs ValueTask overhead and channel throughput. Suite and raw reports: github.com/gdhami-net/dotnet-benchmarks.

Intel Core Ultra 9 285HX · Windows 11 · BenchmarkDotNet 0.15.8 · SDK 10.0.201 · 2026-09-01 · median of 4 quiet runs · typical spread ±3.3% · bars/tables: mean time, lower is better

AwaitOverhead

Await 10,000 already-completed operations: Task vs ValueTask — the overhead argument, measured.

TAKEAWAY ValueTask pays off when a method usually completes synchronously (caches, buffered readers). If the result is usually awaited across threads or stored, plain Task stays simpler and safe.
In this run: ValueTask_completed is fastest — 10.9× faster than the baseline.

RESULTS

method	net9.0	net10.0	ratio	allocated	Δ net10.0
Task_completed baseline	73.5 µs ±1.7 µs	75.0 µs ±2.0 µs	1.00×	702.6 KB	+2%
ValueTask_completed	5.9 µs ±20 ns	6.9 µs ±12 ns	0.09×	72 B	+17%

RATIO VS BASELINE · net10.0

Task_completed1.00×

ValueTask_completed0.09×

THE LEDGER · SAME WORKLOAD ACROSS RELEASES (LOG)

Task_completedValueTask_completed

THE CODE BEING MEASURED

Task_completed — what this measures

Awaiting Task.FromResult in a loop — one allocation per call when not cached.

[Benchmark (Baseline)]
public async Task<int> Task_completed()
{
    var sum = 0;
    for (var i = 0; i < 10_000; i++) sum += await GetTask(i);
    return sum;
}

ValueTask_completed — what this measures

Awaiting a completed ValueTask — no allocation on the synchronous path.

[Benchmark]
public async Task<int> ValueTask_completed()
{
    var sum = 0;
    for (var i = 0; i < 10_000; i++) sum += await GetValueTask(i);
    return sum;
}

ChannelThroughput

Push 10,000 items through a Channel and read them back: unbounded vs bounded.

TAKEAWAY Bounded channels cost a little more per item - that's the price of backpressure, and on a real queue you almost always want it. Unbounded means your memory is the backpressure.
In this run: Unbounded_channel is fastest (it is the baseline).

RESULTS

method	net9.0	net10.0	ratio	allocated	Δ net10.0
Unbounded_channel baseline	472.4 µs ±1.2 µs	458.8 µs ±1.1 µs	1.00×	130.8 KB	-3%
Bounded_channel	524.7 µs ±983 ns	494.2 µs ±878 ns	1.08×	129.4 KB	-6%

RATIO VS BASELINE · net10.0

Unbounded_channel1.00×

Bounded_channel1.08×

THE LEDGER · SAME WORKLOAD ACROSS RELEASES (LOG)

Unbounded_channelBounded_channel

THE CODE BEING MEASURED

Unbounded_channel — what this measures

Channel.CreateUnbounded — no backpressure, no capacity checks.

[Benchmark (Baseline)]
public async Task<int> Unbounded_channel()
{
    var ch = Channel.CreateUnbounded<int>();
    for (var i = 0; i < 10_000; i++) await ch.Writer.WriteAsync(i);
    ch.Writer.Complete();
    var sum = 0;
    await foreach (var v in ch.Reader.ReadAllAsync()) sum += v;
    return sum;
}

Bounded_channel — what this measures

Channel.CreateBounded with room for everything — the capacity bookkeeping alone.

[Benchmark]
public async Task<int> Bounded_channel()
{
    var ch = Channel.CreateBounded<int>(10_000);
    for (var i = 0; i < 10_000; i++) await ch.Writer.WriteAsync(i);
    ch.Writer.Complete();
    var sum = 0;
    await foreach (var v in ch.Reader.ReadAllAsync()) sum += v;
    return sum;
}

speed vs allocation

Fast is one axis. What it costs the GC is the other.

Task_completedValueTask_completedUnbounded_channelBounded_channel