Transcoding Profiles¶

VaultStream transcodes uploaded video into an HLS adaptive bitrate ladder. Custom profiles allow organizations to optimize for their specific delivery requirements.

Default Profile¶

The default_h264 profile produces a standard 4-rung ladder:

Resolution	Bitrate (kbps)	Codec	Audio
1920×1080	5000	H.264 High@L4.2	AAC-LC 192k
1280×720	2800	H.264 High@L4.0	AAC-LC 128k
854×480	1400	H.264 Main@L3.1	AAC-LC 96k
640×360	800	H.264 Main@L3.0	AAC-LC 64k

Available Profiles¶

Profile	Codec	Hardware Accel	Best For
`default_h264`	H.264	Auto	General purpose
`high_quality_h264`	H.264	Auto	Premium content, 4K source
`bandwidth_saving_h264`	H.264	Auto	Low-bandwidth viewers
`archive_h265`	H.265	QSV/NVENC	Long-term archival
`audio_only_aac`	AAC only	N/A	Podcasts, audio content

Custom Profiles¶

Enterprise customers can define custom transcoding profiles via the API:

POST /v1/admin/transcode-profiles

{
  "name": "training_h264",
  "codec": "h264",
  "variants": [
    {"resolution": "1280x720", "bitrate_kbps": 3000},
    {"resolution": "854x480", "bitrate_kbps": 1500},
    {"resolution": "640x360", "bitrate_kbps": 800}
  ],
  "segment_duration_seconds": 6,
  "keyframe_interval_frames": 120
}

Hardware Acceleration¶

CYFR's transcoding farm supports:

Intel Quick Sync Video (QSV) — Default for H.264/HEVC
VAAPI — Linux hardware acceleration fallback
NVIDIA NVENC — Available for enterprise deployments
Software (libx264/libx265) — Universal fallback

Note: Custom scripts can invoke specific encoder flags through the ffmpeg_options field when creating a custom profile. Contact your solutions architect for the full options reference.

High-Density Encoding (ASIC & FPGA)¶

For deployments exceeding 1,000 concurrent encodes — enterprise training platforms with tens of thousands of employees, multi-tenant MSP deployments, or content libraries with millions of hours — GPU-based encoding reaches density limits. CYFR supports dedicated hardware acceleration options at the enterprise tier:

Architecture	Example Hardware	Encode Density	Best For
ASIC	NETINT Codensity, Google Argos-class silicon	8–32 streams per card	Fixed-function, highest density per watt
FPGA	AMD Alveo U30, Aupera	8–16 streams per card	Reprogrammable pipelines, custom codec support
GPU	Intel QSV, NVIDIA NVENC	2–8 streams per GPU	General purpose, H.264/HEVC/AV1
CPU (Software)	libx264, libx265, SVT-AV1	1–4 streams per core	Universal fallback, format flexibility

ASIC Encoding¶

Application-specific integrated circuits (ASICs) are fixed-function silicon purpose-built for video encoding. Unlike GPUs, which include general-purpose compute units, ASIC encoders dedicate all transistor area to the encode pipeline — yielding 3–5× higher transcode density per rack unit than GPU-based solutions at equivalent power draw.

Google's Argos VCU — the ASIC powering YouTube's transcoding fleet since 2021 — demonstrated that a single custom chip can replace the work of approximately 10 conventional server nodes for H.264 and VP9 workloads. NETINT Technologies offers commercially available ASIC encoders (Codensity G5, T408) with similar density characteristics for H.264, HEVC, and AV1.

CYFR offers ASIC-based encoding nodes for enterprise customers with sustained high-throughput requirements. These are dedicated bare-metal servers provisioned in the customer's selected colocation facility.

FPGA Encoding¶

Field-programmable gate arrays (FPGA) occupy a middle ground: reprogrammable hardware that achieves near-ASIC throughput for encode workloads while retaining the flexibility to support new codecs through firmware updates. AMD Alveo accelerator cards and Aupera's video processing FPGAs are used in production by CDN operators (Twitch, Tencent Cloud) for adaptive bitrate ladder generation.

FPGA encoding is appropriate when: - Codec requirements may evolve (e.g., AV1 adoption timeline) - Latency requirements preclude software encoding - Power-per-stream efficiency matters more than raw throughput

CYFR supports FPGA encoding through qualified hardware partners. Contact your solutions architect for a capacity assessment.

Comparison¶

	ASIC	FPGA	GPU	CPU
Density per 1U	32–64 streams	16–32 streams	8–16 streams	4–8 streams
Codec flexibility	Fixed per chip	Reprogrammable	Driver-limited	Unlimited
Power efficiency	Highest	High	Medium	Low
Typical TCO (3yr)	Enterprise licensing	Volume-dependent	Per-GPU cost	Included
CYFR availability	Enterprise	Enterprise	All tiers	All tiers

Note: Custom scripts can invoke specific encoder flags through the ffmpeg_options field when creating a custom profile. For ASIC and FPGA deployments, encoding parameters are configured at the hardware provisioning layer rather than per-job — contact your solutions architect for the full configuration reference.