YouTube's Existence is Insane: How Video Compression, Encode, & Decode Work (Basics)

This video goes through the very top-level basics of how videos work. Most of the discussion is hardware-agnostic, talking about video encode, decode, and compression. At GN, none of us are experts in these topics (and they exit our usual coverage spectrum), so graphics engineer Tom Petersen will be joining to help provide the foundational knowledge as a part of our educational series of engineering discussions. Towards the end, he talks about the hardware-level choices that affect media processing. This is the last of our series of 3 videos with Tom Petersen. Check the others below, and check back for videos with other engineers later!



TIMESTAMPS

00:00 - Video is Complicated
02:07 - Basics of a Block Diagram & Media Hardware
04:51 - Basics of Colors, Pixels, & Media Format Formula
08:08 - Encoding, Compression, & The Human Eye
12:32 - Spatial & Temporal Redundancy
15:00 - Frequency Quantization
18:07 - Symbol Coding & Bit Reduction
20:18 - Intel's Specific Hardware for Media

** Please like, comment, and subscribe for more! **

Links to Amazon and Newegg are typically monetized on our channel (affiliate links) and may return a commission of sales to us from the retailer. This is unrelated to the product manufacturer. Any advertisements or sponsorships are disclosed within the video ("this video is brought to you by") and above the fold in the description. We do not ever produce paid content or "sponsored content" (meaning that the content is our idea and is not funded externally aside from whatever ad placement is in the beginning) and we do not ever charge manufacturers for coverage.

Follow us in these locations for more gaming and hardware updates:


Steve Burke: Host
Vitalii Makhnovets: Video Editing
ft. Tom Petersen, Intel Engineer (Guest)

[GamersNexus]

Hey everyone! This is our third and final installment of educational deep dives with Tom! You can watch our previous two below. We'll have some other industry engineer videos from NVIDIA and case manufacturers coming up. I'm working on booking something technical with AMD hopefully in the near future as well! Aside from the big 3 silicon companies, what other engineering professions within the industry would you like to see on this channel? Even if I'm not familiar with the subject matter, I can study enough of it to at least interview someone for the basics like this!

[Dudummeskind]

TAP is the perfect example why vendors should let their engineers talk to the buyers. The way he makes you understand horribly complex topics is awesome. And it makes you appreciate their products more than any marketing BS.

[panagiotischagias]

Please don't give up on those technical interviews. They are what we need.

[Anton]

Small correction: YUV 4:4:4 / 4:2:2 / 4:2:0 doesn't describe bits, it describes how many chroma samples are stored. The first '4' says that we are talking about rows of 4 luma samples, the second number describes how many chroma samples are stored in the first of two lines and the second number how many chroma samples are stored in the second line.

That means a 4×2 block of luma samples contains

• 8 pairs of chroma samples in YUV 4:4:4
• 4 pairs of chroma samples in YUV 4:2:2
• 2 pairs of chroma samples in YUV 4:2:0

Another way to think about it is that in YUV 4:4:4, each luma sample has its own pair of chroma samples, in YUV 4:2:2, each 2×1 block of luma samples shares a pair of chroma samples (the chroma planes have full vertical but half horizontal resolution) and in YUV 4:2:0, each 2×2 block of luma samples shares a pair of chroma samples (the chroma planes are half horizontal and half vertical resolution).

[Lishtenbird]

Switching to 144p for increased immersion.

[Julian-bkff]

As others have pointed out, these videos with Tom have been fantastic. I think the information is presented in a way that is not only valuable for gamers, but also for many ComSci students as well.

Thanks to everyone involved, and hopefully we can see Tom back in the channel on another occasion!😄

[luizarthurbrito]

13:00 that's why it's so hard to compress confetti, snow, or other super small moving parts in a video. There's even a term called "compression nightmare" for these scenarios. Videos appear to be at a low bitrate, internet usage spikes, as well as cpu utilization.

[Michael-uojj]

Tom is such an awesome guy he deserves his success 100% really appreciate him doing things like this.

[Capanel]

Just sitting here watching this, eating some dinner, and half way through it just becomes immediately apparent to me that real, tangible people figured all of this stuff out and continue persevering and innovating on greater ideas and technologies. It just blows me away how intelligent the people were who designed and produced this stuff. I guess it's just very impressive. I mean, not even 100 years ago did we even have the first television.

[u-ux]

These discussions and presentations have been fantastic. Thank you (everyone involved) for producing this.

[Shiny_Dragonite]

I could watch videos of you and Tom all day and not get bored, and learn many new things. These technical series are fantastic.

[gigaherz_]

WRT frequency domain on images.
Picture it like this: the corner (0, 0) is 0 oscillations -- constant value. the pixel at (0, 1) has zero horizontal frequency, but 1 oscillaiton on the vertical, meaning it starts at 1, goes to 0, then back to 1, sine wave style (well, cosine actually but you get the idea, it's smooth and connects cyclically end to end). the pixel at (0, 2) is the same but has two oscillations vertically, and so on. This step is usually performed on small blocks, 8x8 or 16x16. So on a block of 8x8, the frequency pixel at (8, 8) is a checkerboard, and (0, 8) is a series of 8 horizontal lines black white black white etc. the bottom right pixel (N, N) on any resolution ALWAYS coincides with the pattern that gives you a checkerboard.

[justanotheraccounthere]

Computerphile did an introduction on DCT a few years ago going into more details of the math and intuition of the algorithm, in their 3 part series of covering JPEG compression. For those interested, it is a series worth a watch, as video compression seems to be very similar to JPEG compression on differences between frames.

[leonardocaetano]

It is so great to see a technology channel that actually talks about tech (instead of making funny, reality show-esque videos with graphics cards).

The videos you guys made a while back about latency/input lag and GPU drivers were amazing as well.

[markdeckard]

What makes GN interviews with specialists and engineers so engaging is that Steve can keep up. Brilliant communicator that can translate the info into ELI5 for us idiots.

[adreto]

Video compression + ffmpeg is a modern marvel that powers so much without users knowing.

[Mrbones]

This was incredible. I've always wanted to know more about compression and although I knew the basics, the step-by-step process overview was super helpful to get a greater understanding of how cool compression is. It's one of those mostly invisible technologies that most people don't know exist but are absolutely essential to keeping everything functioning.

[CyricFTW]

This guy is great. Thanks for collaborating…

[michaelmoses]

It's almost impossible to get bored with GN.

Also starting with the bandwidth YouTube would need is crazy.

[AgentSmith]

I can't wait for AV1 to truly take off, so that 8K, 120 fps, HDR, 12 bit colors, rec2100, 444 chroma subsampling and all that jazz can become common.