Hardware NVEnc Test for low bitrates.

Today I’m testing NVEnc video encoding quality with low bitrates. ‘Why?’ you ask? Well, I have some interest about video encoding and video entertainment. Not that I am capable of producing them— I just want to learn a little more about them, and maybe someday I can make videos or stream video for fun.

Notice: If you don’t want to read everything/checking the results on web, you can download every test files, be it videos or images, at the end (because it took way too much space, I’m deleting them). I am not professional, nor knowledgeable about the subjects, please do share your thoughts in the comments section politely. If it ends up bad, and wasn’t worth it I might just delete everything though. :p

My home internet connection is shit. I’ll say it again, it is shit. Well, it is usable for watching stuff, as the download speed is adequate for most things (around at least 10Mbps and up, but limited to maximum 20Mbps), the shitty thing is the upload speed (they say it’s 0.2x of the download speed, equals to maximum 4Mbps. Most of the time I get 1.5~3.5Mbps, but very jittery for some reason). So, if someday I’m going to upload or stream videos, I need to do something about it.

I wanted to upgrade the connection to something better, but, well, for economic reasons, it’s too difficult. There’s also the option to change the ISP to a better one, with equal down-up rate, but for some reason they force two years contract when you sign up. If you breach the contract, they’ll make you pay ten times the monthly cost. And of course, my family protested heavily and kept complaining about the risk. “What if the connection is unsatisfactory and we can’t change back to the former ISP because of the contract? No way we’re going to risk our daily needs!” or something similar, they said. Oh, too much out-of-topic, let’s stop right here.

So here we are, testing video encoding for low bitrates. At first, I consider x264 or x265 encoders, but they are heavily CPU based which is not optimal for me. So I decided that I will use NVEnc because the CPU load will be much lower in theory, and I also heard it won’t impact gaming performance very much too.

・ I am using Handbrake for the test (a free video encoding/compression software, you can check that here, very simple and easy to use).
・ I used video source from NVEnc encoded h264, 1080p60, at around 50,000kbps—clean and superb quality in my opinion. It is a clip I captured from Destiny2 game. You can download the sample video here (20secs, 120MB) or with the same file plus some other download options at the end of the post, below.
・ As my internet upload rate hovers around 1.5~3.5Mbps, I will be testing the bitrate at 960Kbps and 1536Kbps in both 720p and 1080p, also in both 30 and 60 frame per second.
・ I took the snapshots at 8 seconds in and 17 seconds in. The first one is when there’s quite a lot of motions, the second one is when there’s not much motions going on. Both have quite a lot of details to be pixel peeped if that’s jingle your bells.

Here it is, in this sequence: all at 1536kbps (720p30, 720p60, 1080p30. 1080p60), all at 960kbps (720p30, 720p60, 1080p30, 1080p60). We’ll compare it motion screenshot first, then the still one after. For better and easier comparison, viewing in horizontal 1080 monitor/screen is recommended.

● Motion 1536kbps

Mot 720p30 – 1536kbps
Mot 720p60 – 1536kbps
Mot 1080p30 – 1536kbps
Mot 1080p60 – 1536 kbps
Stop 720p30 – 1536kbps
Stop 720p60 – 1536kbps
Stop 1080p30 – 1536kbps
Stop 1080p60 – 1536kbps

● 960kbps

Mot 720p30 – 960kbps
Mot 720p60 – 960kbps
Mot 1080p30 – 960kbps
Mot 1080p60 – 960kbps
Stop 720p30 – 960kbps
Stop 720p60 – 960kbps
Stop 1080p30 – 960kbps
Stop 1080p60 – 960kbps

It is all. Starting with the motion captures.
As you can see, in most cases 720p30 has the most details, with less pixelations/blockiness. 720p60 might be alright, but the pixelations and blurriness became really distracting, as you can see when comparing mot720p30 and mot720p60 specifically at the weapon model texts, and the turquoise stripes at the ground. That descriptions apply to both 1536kbps and 960kbps, although it is fairly obvious the 960 has a lot less details than the 1536 version.
In 1080p, the same is basically can be said. But again, as the bitrate is very low, it’s kind of confusing whether it is worth it to bump the resolution to 1080p or just stay in 720p. I would say 720p is better though, because even if the general lines in 1080p might turned to has less jagged edges, the 720p details is much better (you can see around the gun stock for the details, also around the fallen tower background and grey parts of the gun for the jagged edges and pixelations differences)

In the captures of the more still part though, 720p quality is very, very acceptable either in fps and in bitrate. Even when comparing 720p30-1536kbps with 720p60-960kbps, you can barely see difference, probably the second one wound appear to be barely softer detail than the former one, but it still looks quite good.
At 1080p, the 960kbps-30fps is way too blurry compared to 1536kbps-60fps at the same resolution, or even compared to 720p60-960kbps. At least, you should make it 1080p30-960kbps or better 1536kbps at least, if you really need 1080p.

Personally, I think for motions, 720p30-1536kbps will be the best choice, followed by 1080p30-1536kbps. Any videos in 60fps is way more blurry and pixelated, and pretty ugly to be honest, even though it still felt very smooth. Kinda like watching worse youtube videos in this regard. Not that 720p30-1536kbps does not look ugly, it’s just the fairest of the ugly ones.
For stills though, 720p30-960kbps is the best choice, hands down, as there’s not that much detail loss, at very great data savings.

That’s it I guess. I can be possible very wrong, because I mostly focused on the screenshots. I have the original files, so check it out yourselves below. Thanks for checking my tests out!

Below here is the download for your personal, super intimate, pixel peeping ;p Download are no longer available.

Click here for the original video only.

Click here for every test results, including videos and images (original video not included).

Click here for 30fps test results only, including videos and images.

Click here for 60fps test results only, including videos and images.

Click here for every result screenshot images only.

Click here for every result videos only.

Note though, as I am on a shared web server, if the downloads got too intense, I might need to stop hosting it on my own space. I am considering cloud like G-drive, Mega, or anything other, but it’s just seemed to be so troublesome so for now I’ll host it here. A few hundreds of gigabyte bandwidth should be enough for a month. The files are no longer available, as they took a huge chunk of my small web space. I need the space for other projects, sorry.

Any broken links or questions? Please leave whatever is in your mind below at the comment sections. Or you can also go ahead and ask me by sending it to pepo at pepo dot work, if I totally forgot about the damn blog and never responded.

I think I’m going to do similar tests with H264 codec next time, but no promises. Streaming services that uses RTMP (I think it was RTMP) only support realtime h264 I think, which isn’t like uploading videos, because VOD will be converted to more easily distributed format without the need of fast/realtime process (h264, or vp9, or the heavily anticipated av1)…. I heard Youtube, Twitch, or Mixer uses RTMP for the streaming services. I know for sure Youtube will process any regular non-h264 videos to whatever codec they use though.

Leave a Reply

Your email address will not be published. Required fields are marked *