mpsc::channel Mac and Ubuntu performance difference

[boranby]

I am running this benchmark on Mac Studio with M1 Max and Ubuntu Server with AMD 9374F. When I run it on Ubuntu it's very slow compared to Mac. I tried to increase the BUFFER_SIZE but no performance change is detected. Can you help me to figure out why there is a significant performance difference?

Mac: time: [179.77 ns 183.14 ns 186.90 ns]
Ubuntu: time: [756.74 ns 757.13 ns 757.52 ns]

use criterion::{black_box, criterion_group, criterion_main, Criterion};
use std::sync::Arc;
use tokio::runtime::{Builder, Runtime};
use tokio::sync::mpsc;

const BUFFER_SIZE: usize = 1024 * 8;

fn bench(c: &mut Criterion) {
    let runtime = get_runtime();

    let (data_tx, mut data_rx) = mpsc::channel::<Arc<[u8]>>(BUFFER_SIZE);
    runtime.spawn(async move {
        while let Some(_data) = data_rx.recv().await {
            //println!("data: {:?}", data);
        }
    });

    c.bench_function("send", move |b| {
        b.to_async(&runtime).iter(|| async {
            data_tx
                .send(black_box(Arc::new([0u8; 1024])))
                .await
                .unwrap();
        });
    });
}

fn get_runtime() -> Runtime {
    let runtime = Builder::new_multi_thread()
        .worker_threads(6)
        .enable_all()
        .build()
        .unwrap();
    runtime
}

criterion_group!(benches, bench);
criterion_main!(benches);

Bench profile:

[profile.bench]
opt-level = 3
lto = true
debug = false
debug-assertions = false
codegen-units = 1
overflow-checks = false

If I change it to this, Mac is still almost 2 times faster.

...
let buffer = Arc::new([0u8; 1024]);
    c.bench_function("send", move |b| {
        b.to_async(&runtime).iter(|| async {
            data_tx.send(black_box(buffer.clone())).await.unwrap();
        });
    });
...