I would like to start a thread detailing management suites that support hardware acceleration, I've started a few I know of below, please add to the list any you know including details!
Milestone - Intel
Genetec - Intel & NVIDIA
DigiFort - Intel
Hanwha SSM - NVIDIA
CNL Software - Intel, NVIDIA & AMD (ONVIF interfacing)
Milestone supports NVIDIA as well:
Video decoding in XProtect Smart Client using NVIDIA graphics card
The computer running XProtect Smart Client can use NVIDIA® and Intel® display adapters for hardware accelerated video decoding. You can add multiple NVIDIA display adapters for better XProtect Smart Client performance and monitor the CPU and NVIDIA GPU load from the system monitor.
In addition in the 2018 R2 release:
Hardware acceleration in the Recording Server using NVIDIA: Maximize the potential of your system and save money on costly hardware
XProtect, the world’s best performing VMS, is now also the only one with support for multiple NVIDIA GPUs in the Recording Server. XProtect 2018 R2 makes it possible to harness the power of multiple NVIDIA graphics cards added to the Recording Server and to connect more cameras per server than ever before.
Thanks for the info. One question, how is acceleration benefiting the recording server?
"how is acceleration benefiting the recording server?"
My take:
As each camera is added to recording servers, the CPU rises.
Off-loading some of this processing to GPU allows the recording servers to be able to record more cameras than they could using just CPU.
Four benefits on the recording server I can think of. These all center around the ability to have lower-cost, lower-powered systems with maximum compute resources, handling many hundreds of cameras without sacrificing features.
1) Milestone is building a framework for 3rd-party server-side analytics and server-side analytics require decompression of the video. Having loads of low-cost compute available via CPU, Intel Quick Sync, and NVidia is helpful.
2) Milestone users have the option to do motion detection on the camera or on the server. Most choose on-the-server as it allows a variety of benefits such as super speedy Smart Search, set-up of motion areas in the Management Client, etc.. This also requires decompression of the video stream.
3) Super powerful, low-cost, and/or low-power, recording servers that can be built using either Core Series or Xeon builds.
4) A way to increase the power of existing recording servers without having to rip-and-replace with new servers. One current use-case might be the Intel Spectre Meltdown flaw in which every Intel system in existence will likely see some form of decreased compute power as patches are released. If you have a server that is already loaded down, the next Windows update might result in the server that is actually overloaded. An easy, low-cost fix might be to drop in an NVidia card or two.
A GPU is not like a CPU, it's generally dedicated silicon for video trans coding operations only. CUDA is an programmatic exception, but why in a record to disk server.. One possible answer is that Milestone is now trans coding to a common media format prior to disk, but this was not the case in previous releases, I hope someone will put us straight out of pure interest :) cheers.
"how is acceleration benefiting the recording server?"
decoding the video to perform the centralized motion detection with metadata
I have been duped!
Thanks for setting me straight... ;)
Question: can you explain what 'transcode prior to disk' means?
Thanks!
Oh yes, you just reminded me that there should have been a #5 which would be the ability to do adaptive, on-the-fly, real-time transcoding for various purposes/needs coming in the future.
Milestone does not, nor any others I know of, do any type of transcoding on the streams before going to disk. These are always native.
OK, so analytics is being performed in the recording server, understood, as you must decode the frame before performing and motion detection :) thanks.
To answer previous post, if you wanted to say record in a common format such as H264, any other format, lets say MJPEG would have to be trans coded prior to disk, decoded then re encode to the new format. But for clarity, this is not whats happening in hardware on the Milestone server, thats about analytics.
Any more VMS vendors using hardware acceleration that we know of?
"if you wanted to say record in a common format such as H264, any other format, lets say MJPEG would have to be trans coded prior to disk, decoded then re encode to the new format."
yes, I understand what transcoding is.... my request for an explanation should have been more clear.
I don't understand the logic of encoding in format 1 and then transcoding to format 2 'prior to disk'. What value can this possibly have vs simply encoding in format 2 to begin with?
I don't see any logic in that either. No video surveillance solution that I am aware of transcodes before going to disk. That would be costly in many ways.
Of course video intended for broadcast or web streaming (e.g. YouTube) is always transcoded into various formats before going to disk, but that is typically not a real-time transaction by any means.
Not all encoders (eg ip cams) may support the desired common format in the recording server- why have a common format? - forwarding to web & mobile under a common open protocol perhaps, I can think of a few reasons, but lets not go there! I was trying to second guess the answer before being put straight that it was for in fact analytics :) sorry for any confusion.
Aside, I'd of thought analytics would be done on either the client or dedicated analytics server, interesting that it's done on the recording server.
Cheers.
Analytics can be done anywhere there is a video stream, and the push in recent years (and currently as well) is towards the edge device. This makes a lot of sense because we can perform the analytic before the video is compressed reducing the typical CPU overhead of decompression. The metadata created at this point can be used for real-time applications, or stored for future use and analysis.
However, there are many future use-cases in which analytic activities (translated into, "We need a decompressed stream here in the chain") may need to take place outside of the camera and much later in the lifecycle of the videos existence. In addition, the exact metadata being sought may not be known at the time the video is captured, or we may not know it is useful at the time of capture, or a solution to capture it at the edge in real-time may not exist.
A few particular use cases might include; "stacked" analytics by various 3rd-parties operating in a chain, analytics taking place on stored video, comparative analytics that operate on more than one stream at a time, and even analytics taking place on video being sent out to clients. The main thought being that the extraction of all different types of metadata on demand, and not having to be predetermined, opens up a lot of new possibilities.
Thanks Richard for the idea, there have been some good responses already.
Client Decoding:
Milestone - supports NVIDIA now (in addition to the OP mention of Intel)
Exacq - Supports NVIDIA and Intel
HikCentral - Supports NVIDIA and Intel
GeoVision - Supports NVIDIA and Intel
Not for Client Video Viewing:
Avigilon supports NVIDIA CUDA compute for Appearance Search
Milestone - NVIDIA and Intel for Server-based Motion Detection video processing
We have a recent breakdown of the NVIDIA GPU lines and where they fit into the surveillance market. (IPVM NVIDIA for Surveillance)
I will add more platforms that support GPU decoding to the list as they come up.
For clarification, on the server-side with Milestone, NVIDIA and Intel GPU are still used for decoding, it is just decoding the video for motion detection purposes.
Also, client decoding also supports Intel GPU.
AxxonSoft
Herta Security
Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.
