CONCEPT Cited by 1 source
Low-end device inclusion¶
Definition¶
Low-end device inclusion is the product-level constraint that engineering choices — codec, ML model, rendering pipeline — must serve users on older, slower, cheaper devices, not just flagship handsets. For a billions-of-users RTC product like WhatsApp, the low-end population is a first-class audience, not a rounding error. This constraint frequently rules out otherwise-attractive ML-based solutions on complexity grounds (Source: sources/2024-06-13-meta-mlow-metas-low-bitrate-audio-codec).
The specific numbers (Meta 2024)¶
Meta's MLow announcement is unusually explicit about the size of the low-end population they design for:
- "More than 20 percent of our calls are made on ARMv7 devices."
- "10's of millions of daily calls on WhatsApp are on 10-year-old-plus devices."
These numbers are load-bearing on the decision to build a classic-DSP codec rather than ship Meta's own Encodec ML codec. The argument is not about absolute capability — Encodec is better — but about addressable user population: a codec that only runs on expensive phones helps only expensive-phone users.
The Meta quote¶
"While these AI/ML-based codecs are able to achieve great quality at low bitrates, it often comes at the expense of heavy computational cost. Consequently, only the very high-end (expensive) mobile handsets are able to run these codecs reliably, while users running on lower-end devices continue to experience audio quality issues in low-bitrate conditions. So the net impact of these newer computationally expensive codecs is actually limited to a small portion of users."
The operative phrase — "net impact… limited to a small portion of users" — is the reusable constraint. If an intervention helps 15% of users and breaks on 85%, its net impact is bounded.
Recurs across this wiki¶
The low-end-device-inclusion pattern echoes across other compute-constrained production decisions:
- concepts/quantization / Instacart's explicit choice to not ship FP8 quantization on their SRL student despite a 10% latency win — quality over throughput, at the cost of addressable hardware diversity.
- concepts/quality-bitrate-complexity-tradeoff — the complexity axis is the one low-end devices enforce.
- patterns/classic-dsp-over-ml-for-compute-constrained — the direct pattern MLow exemplifies.
What the constraint doesn't imply¶
MLow is not "dumber" than Encodec — it's classical rather than learned. Meta put 2.5 years of development into MLow to get the quality-per-CPU-cycle that hits the low-end target. The pattern is not "use simpler things"; it's "pay the engineering cost of the compute-constrained solution when your install base demands it."
Seen in¶
- sources/2024-06-13-meta-mlow-metas-low-bitrate-audio-codec — canonical articulation, with specific install-base numbers.