NVIDIA Updates Video Encode and Decode Matrix with Reference to Ampere GPUs

[Mouth of Sauron]

Basically, neither decode nor encode absolutely *needs* hardware support - 3090, with hardware support, still doesn't allow real-time encoding for higher resolution - perhaps maybe 720p, there isn't lots of tests online.

Same goes for Intel CPUs, which have hardware support for both encoding and decoding.

As for decoding, so far hardware acceleration on GPU just lessens the burden on CPU side - virtually anything can decode file in real-time, except load is bigger if there isn't hardware support (but then again, both AMD and NVIDIA have support for decoding, we just don't know how much - the only results I found were for 3080 and I can't remember which CPU - utilization dropped from ~90% to some 50-70%, but not all cores were utilized).

Perhaps it would be interesting to test if 16 core 5950x can decode faster then Intel offering with HW support, or how many FPS which offering makes (though I doubt anything will allow real-time).

VP9 is considerably faster, and ffdmpeg supports both.

I realize this isn't something that everybody needs, but it means something to streamers or to people who compress raw video material - AV1 gives considerably better quality than VP9 at lower bitrate, and same goes for VP9 and x.265. Currently, most real-time encoding is done by x.264, which is rather old (VP9 is from 2013th, I think, and AV1 is from 2018th, x.264 is ancient).

Just an idea, encoding is a highly CPU-intensive test, HW does help - but only to the point.

Back in the days, I did some testing on VP9, but hardware was much less advanced (I mean just CPUs, I don't have a number of CPUs to test, and HW support was non-existent at that time).

I don't know how much is ffdmpeg optimized for either Intel, AMD or NVIDIA, too...

 

[mtcn77]

Basically, neither decode nor encode absolutely *needs* hardware support -

You have to understand, client side encoding practically kills your mouse trigger reflex. For the recent while Nvidia has been demonstrative since their good encoding eliminated the use of cpu, thus saving encoding latency for the user. You have to run a seperate server side pc just to offload the encode latency, otherwise, which isn't practical, nor sane.

 

[Mouth of Sauron]

You have to understand, client side encoding practically kills your mouse trigger reflex. For the recent while Nvidia has been demonstrative since their good encoding eliminated the use of cpu, thus saving encoding latency for the user. You have to run a seperate server side pc just to offload the encode latency, otherwise, which isn't practical, nor sane.

Yeah, but which CPU? We have advancement for both Intel (HW encoding) and AMD (raw horsepower, many cores) + NVIDIA GPU encoding. I know that currently most encoding is done on separate PC, but we still have question 'how fast', which is important.

At some time, single PC configuration for video encoding will become reality.

Besides, many CPU-burn tasks (rendering, for example, I could name some others) is not exactly multitasking-friendly...

 

[mtcn77]

At some time, single PC configuration for video encoding will become reality.

Well, you did miss the point on hardware encoding - it works like a seperate pc. It is quite handy, latency doesn't go on par with encoding latency.

 

[Mouth of Sauron]

Well, you did miss the point on hardware encoding - it works like a seperate pc. It is quite handy, latency doesn't go on par with encoding latency.

Nope. You don't know what you're talking about, sorry to say. Very small amount of either CPU or GPU is dedicated to hardware encoding and it effects results accordingly.

Because I don't want to be misunderstood again, I'll stop this pointless discussion now - I suggested a thing, it's upon the staff to decide is it worthy or representative, I'm fine either way.

 

[mtcn77]

Nope. You don't know what you're talking about, sorry to say.

I'm notmaking such a case. I just read what hardware.fr inquired about the subject. Hardware encoding deters video compression lag.