Netint technologies


Video Server

Transcoding | Codensity ASIC G4
Logan Video Server powered by Video Processing Unit (VPU) with ASIC Codensity G4 from NETINT

Transcoding workhorse

Built on the Supermicro 1114S-WN10RT server platform, NETINT’s Logan Video Server boasts ultra-high density encoding capacity enabled by ten T408 video transcoders.
NETINT application specific integrated circuits (ASICs) are the secret to replacing software-driven video encoding for video platforms and delivery services wanting to decrease carbon emissions up to 40x, operating costs as much as 20x and reduce the number of servers needed by 90% for the equivalent video encoding workloads.
The results are profound and transformational.

Designed as a quickstart solution for high density live video encoding and transcoding, the Logan Video Server comprises ten T408 video transcoders in a 1RU chassis that performs the equivalent work of roughly 10 dedicated servers running a typical open-source FFmpeg and x264 or x265 configuration. The server delivers the lowest TCO of any solution in the market, and is a drop-in replacement for existing CPU and GPU encoding stacks.

ASIC Benefits

Early adoption of ASICs slingshots video streaming platforms ahead of their competition due to their low operational cost basis.

Low Cost
The industry’s most cost efficient server platform with ultra-low CAPEX and OPEX.

Ultra High Density
Up to 80x 1080p30 live streams in 1RU. 10x the density of software encoding.

High Power Efficiency
One T408 U.2 module consumes only 7W power at full load. The server contains 10.

Enterprise NVMe Integration
Available in U.2 and PCI form factors, drop-in for any x86 or Arm-based server.

Low Latency
ASICs deliver sub-frame latency for interaction and live ABR streaming.

Encode up to 10x 4Kp60 live streams supporting mass streaming applications..

Fully integrated 1RU encoding solution can be easily deployed for additional capacity.

HEVC, H.264
Ultra flexible server platform, HEVC and H.264 live encoding and decoding.

Environmental Impact of ASICs.png

This is why Google built a custom chip for YouTube

For everyone else who isn’t Google,
we did the heavy lifting for you.

Google’s ASIC called ‘Argos,’ helps YouTube encode and process videos much more efficiently. Argos chips provide “up to 20-33x improvements in compute efficiency compared to its previous traditional server set up,” according to a Google executive. Another report suggests that Argos replaced over 10 million Intel CPUs in YouTube.


Required to deliver 10K concurrent live HD streams



Servers required to deliver 10K concurrent live HD streams


NETINT ASIC video processing units (VPU)


NVIDIA T4 graphic processing units


Intel SVT with CPU-based encoding

The real cost
of live streaming

CAPEX and OPEX comparison using
3 video process/encoding scenarios.
Test parameters: Servers running 100 concurrent five-rung encoding ladders. x264 very fast preset used for CPU-only encoding.

SOURCE: How to Slash CAPEX, OPEX, and Carbon Emissions Using the NETINT T408 Video Transcoder

Logan Video Server, Quadra Video Server - The real cost of live streaming

ASICs will be the engine powering all future video streaming experiences.

Early adoption of ASICs slingshots video streaming platforms ahead of their competition due to their low operational cost basis.

Designed for the Cloud

High-density live UHD transcoding

The T408 video transcoder takes full advantage of the video processing capability inside the Codensity G4 ASIC to support H.264 and HEVC live encode and transcode functionality of 4K UHD video in SDR and HDR with HDR10 and other popular high dynamic range standards. By offloading complex encode and decode processing to the Codensity G4 ASIC, the T408 video transcoder minimizes host CPU utilization. The result is a significant improvement in real-time transcoding density compared to any software or GPU-based transcoding solution.

As many as ten software-based video encoding servers may be replaced for every NETINT Logan Video Server that is installed in the data center.

High power efficiency

Each NETINT T408 U.2 module consumes only 7W of power at full load. This makes the Logan Video Server the most energy efficient video transcoder available.

Enterprise NVMe Integration

Available in the U.2 form factor, the T408 offers a simple upgrade path from CPU-based software to ASIC-powered video transcoding on any x86 or Arm-based server


Video Server

Transcoding | Codensity ASIC G4
Logan Server Diagram

Simple Integration

Open-source suite of processing tools.

Many video processing and transcoding applications developers use FFmpeg and GStreamer, two open-source software libraries offering a vast suite of video processing functions. The T408 includes highly efficient FFmpeg and GStreamer compatible SDKs, allowing operators to apply a patch to complete the integration.

The libavcodec patch on the host server functions between the T408 NVMe interface and the FFmpeg and GStreamer software layers, simplifying integration and enabling fast and efficient performance and capacity upgrades.


CPU Options

AMD EPYC™ 7232P Server Processor (8-core)

AMD EPYC 7543P Server Processor (32-core)

AMD EPYC 7713P Server Processor (64-core)

Operating System

Ubuntu 20.04.05 LTS (as of May 2023)


8x 16GB DDR4-3200


400GB M.2 SSD

NVMe Support


PCIe Expansion

Up to 3x PCIe slots

Network Options

Dual 10GBase-T LAN

Power Supply

700W: 100 – 140Vac

750W: 200 – 240Vac

750W: 200 – 240Vdc (CCC only)


10x NETINT T408

Encoding Capacity

Up to 10x 4Kp60 or 80x 1080p30 (HEVC and H.264)

Codec Support

H.264 – Encode/Decode

HEVC – Encode/Decode

Transcoder Software

FFmpeg, GStreamer


Video Transcoder

Codensity ASIC G4
NETINT Codensity, ASIC-based T408 Video Transcoder


Form Factor

U.2 (SFF-8639)


PCIe 3.0 x4



Power Consumption (Typ)



24/7 Operation

Operation Temperature

0 degrees C to 70 degrees C

RoHS Compliance

Meets requirements of European Union (EU) ROHS

Compliance Directives

Product Health

Self-Monitoring, Analysis, and Reporting Technology


commands Temperature Monitoring and Logging

Hardware Interface

Available U.2 slot

The video engine inside.

The Quadra T1U is a U.2 form-factor video processing unit with one Codensity G5 ASIC. Operating in x86 and Arm-based servers, the T1U enables video operators to move from software to hardware-based encoding to power real-time video applications and the metaverse at a TCO that is up to 85-99% lower than CPUbased solutions.

The Quadra T1U uses the Codensity G5 ASIC to support AV1, HEVC, and H.264 real-time video encoding at up to 8K resolution with 10-bit HDR. The Quadra T1U’s exceptional throughput enables ultra-low latency encoding of up to 16 broadcast quality 1080p60 streams using AV1, HEVC, or H.264.

The addition of two Deep Neural Network engines capable of 18 trillion operations per second (TOPS), enables functions such as object detection, classification and segmentation, and ROI to provide additional data to the encoding engine for image quality improvement and content-adaptive rate control.

H.264 AVC Encode/Decode H.265 HEVC Encode/Decode


CBP / BP / XP / MP / HiP / Hi10P
Main / Main 10


1 to 6.2

1 to 6.2 Main Tier

Min / Max Resolution

Min: 32 x 32 / Max: 8192 x 5120

Scan Type



64kbit/s to 700Mbit/s

VP9 decode

Profile 0, Profile 2, Level 6

Software Integration

FFmpeg and GStreamer SDKs and

direct integration with LibXcoder API


4K @ 60 fps | 1080p @ 240 fps

Region of Interest (ROI)

ROI enables the quality of some regions to be

improved at the expense of other regions

Closed Captioning

T408 supports EIA CEA-708 closed captions for

H.264 and H.265 encode and decode

High Dynamic Range (HDR)

T408 supports HDR10 & HDR10+ for H.264 & H.265

encode and decode

Low Latency 

T408 supports sub-frame latency

IDR Insert

Forced IDR frame inserts at any location

Flexible GOP Structure

8 presets plus customizable GOP structure


Transcoding with Scaling

This table details the H.264 and HEVC output at the specified resolutions and frame rates; and the associated cost per stream. All inputs are scaled to the designated targets. Though the host CPU performs the scaling in these tests, Note CPU utilization remains exceptionally low, reducing power costs and carbon emissions.

Logan Server - Transcoding with Scaling


This table shows the number of full encoding ladders produced by the server and the cost per ladder for that output. Note the low CPU usage, despite all lower resolution rungs being scaled by the host CPU.

Logan Server - Ladders

Power Consumption

One of the key strengths of ASIC-based transcoders is ultra-low power consumption, which reduces OPEX and carbon emissions. You see this in the power figures, particularly the Watts/Output, which are orders of magnitude lower than comparable figures for CPU-based transcoding.