VERIFIED MULTICHANNEL AES67

256 channels received.
256 transmitted.

Measured simultaneously on the same quad-core Arm Cortex-A72 reference platform: 256 receive channels and 256 transmit channels in one unified engine.

Benchmark your target platform Review the data
BAMA BOX VERIFIED SCALINGSAME ARM PLATFORM
1 ms TARGET64

channels

8 streams × 8 channels
2 ms TARGET128

channels

16 streams × 8 channels
3 ms TARGET256

channels

32 streams × 8 channels

Configured receiver targets · verified logical ALSA output channels

MEASURED FULL DUPLEX

256 RX + 256 TX.
One unified userspace engine.

32 streams × 8 channels per direction ran simultaneously over physical Ethernet at a 3 ms configured receiver target.

SIMULTANEOUS CAPACITY256 + 256

RX + TX channels

32 streams × 8 channels per direction · 1 ms packets

LINUX AUDIO BOUNDARYALSA ↔ AES67

transmit and receive

ALSA to AES67 for transmit · AES67 to ALSA for receive

UNIFIED ENGINE RESOURCES2.244 cores

aggregate CPU

166.95 MiB memory on the quad-core Cortex-A72 reference platform.

Measured media boundary: simultaneous AES67 receive to ALSA and transmit from the Linux audio side. Physical multichannel audio interfaces are qualified on the OEM target.

MEASURED AES67 TRANSMIT

256 channels transmitted.
On the same Arm platform.

The quad-core Cortex-A72 reference platform transmitted 32 streams × 8 channels over direct physical 1 Gb Ethernet.

TRANSMIT CAPACITY256

AES67 channels

32 streams × 8 channels · RTP/L16 · 48 kHz

PRODUCT MEDIA PATHALSA → AES67

Linux audio to network

A standard integration boundary for the OEM audio source.

SENDER RESOURCES0.6048 cores

sender process CPU

28.5 MiB RSS on the quad-core Cortex-A72 reference platform.

Measured boundary: the 256-channel transmit path ran from the Linux audio side to AES67 over physical Ethernet. Hardware-specific audio input and timing are qualified on the OEM platform.

EXECUTIVE BENCHMARK SUMMARY

Capacity rises with the configured receiver target.

Reference platform: Raspberry Pi 4, quad-core Arm Cortex-A72, Debian 12, direct physical 1 Gb Ethernet and software PTP/software timestamps. Final performance is established on the OEM target SoC.

ChannelsMeasured topologyPacket timeReceiver targetReceiver CPUSystem CPURSSOutput
648 streams × 8 channels0.5 ms1 ms0.5200 coresSee report note90.7 MiBAES67 → ALSA
12816 streams × 8 channels1 ms2 ms0.7926 cores1.3868 cores98.9 MiBAES67 → ALSA
25632 streams × 8 channels1 ms3 ms2.2317 cores2.3415 cores115.5 MiBAES67 → ALSA

Receiver capacity was measured from AES67 input through ALSA output. Physical outputs, SAI/TDM interface counts and final product capacity are established on the OEM target.

MEASURED HEREAES67 receiver densityPackets through final logical ALSA PCM validation
ESTABLISHED ON THE OEM TARGETPhysical product I/OSAI/TDM lanes, codecs, DSPs, routing and board interfaces
MATCHED 64-CHANNEL WORKLOAD

Same 8 × 8-channel workload.
Different first passing targets.

The unchanged stock RAVENNA/Linux path and Bama Box were exercised through the common daemon configuration at 64 channels.

BAMA BOX1 ms

strict presentation gate passed

Packet, clock, presentation and final ALSA PCM gates passed for the 64-channel workload.

TESTED STOCK RAVENNA/LINUX3 ms

first final ALSA capture pass

Final ALSA capture continuity failed at 1 and 2 ms and passed at the configured 3 ms target.

AT THAT SAME 3 MS TARGET256 / 64

channels exposed by tested configurations

Bama Box sustained 256 logical channels; the tested common RAVENNA configuration exposed 64 usable input routes.

Scope: the four-times figure compares usable channel counts exposed by the tested configurations. The tested RAVENNA control/configuration path exposed 64 routes; its CPU ceiling and the wider capacity of the RAVENNA standard were outside this measurement. The validation boundaries differ.

REAL HARDWARE OUTPUT

1 ms configured receiver target sustained through PCM5122/I²S.

One stereo RTP/L16 stream sustained through ALSA → I²S → PCM5122 with a maximum runtime ALSA queue of 34 frames / 0.71 ms.

SENDERmacOSRTP/L16 + PTPd2 GM
→
NETWORKDirect 1 GbEno intervening switch
→
RECEIVERBCM27114× Cortex-A72
→
OUTPUTPCM5122 / I²S34 frames / 0.71 ms max queue

This validates the configured receiver target through the physical PCM device path. Analog and network-to-DAC end-to-end latency sit outside this measurement boundary.

X86 BACKSTAGE POTENTIAL

Receiver efficiency beyond embedded endpoints.

A separate cloud KVM/Linux-container benchmark exercised a 32-channel workload through the UDP fallback available in that environment. It shows potential for mixers, console engines and backstage servers; direct packet access depends on the OEM security configuration.

6×lower CPU load
11×lower memory

32 audio channels · compared with FFmpeg on the same x86/KVM environment

METHODOLOGY + SCOPE

What the results establish.

The measurements remain separated so receive density, transmit continuity, physical output and x86 resource evidence are each read at their measured boundary.

01

Receive boundary

AES67 network input to logical ALSA output. Physical PCM and SAI/TDM implementation are qualified on the OEM target.

02

Clock environment

Software PTP and software timestamps on the Arm reference platform; no PHC or hardware timestamps.

03

Full-duplex scope

The unified engine sustained 256 receive and 256 transmit channels simultaneously across the ALSA ↔ AES67 product boundary.

04

Comparison boundary

RAVENNA validates final ALSA capture continuity; Bama Box adds exact RTP-to-presentation instrumentation.

05

Transmit boundary

Linux audio from the ALSA side to AES67 network output. Target hardware supplies the physical audio source and timing integration.

MEASURE YOUR PRODUCT TARGET

Move from reference data to your hardware.

We can benchmark receive, transmit and simultaneous operation using your SoC, timing source, NIC and audio interface.

Plan a joint benchmark