It’s such a shame as h265 is such an amazing codec breakthrough. I’m in the process of converting my library for space saving and the h265 files are literally 50% of the original size (give or take), with imperceptible quality difference. I can reencode around 100-200GB/day typically, using a 3090

I really suggest not using GPU encodes for this purpose unless it's mostly worthless archival content. You can save way more space using AV1 on CPU via things like av1an. If you really like H265 using it on CPU also gets much better quality/size tradeoff.

The encoding speed with my CPU was less than real time. This was unacceptable performance to me.

Maybe my CPU settings weren't right?

If you have enough cores, av1an can do it pretty quickly by splitting the video into chunks.

The CPU is an 11700k it’s not the beefiest. Maybe the high end AMD chips make this a more viable choice

Assuming you’re using FFMPEG?

Do you have a good guide for balancing quality and size? I’ve searched but never found something that really nails it for me. I have until now just been keeping everything as it streamed off the dvd or bluray in mpeg4 or h264 in an mkv and yeah, time to re-encode in to something more reasonably sized.

Yes, I wrote a python script that uses FFMPEG and detects the bitrate of the file and determines approximately what CQ to use. If the original file has a low bitrate, by reencoding it with a high CQ you can actually increase the file size (lol).

Normal CQ = 28 Aggressive CQ = 34

  # Thresholds to detect ultra-low-bitrate inputs (use more aggressive encoding)
  # (width, height) → min sensible bitrate (bps). If stream <= threshold, use aggressive profile.  

 LOW_BITRATE_THRESHOLDS = [
       ((3840, 2160),  13_000_000),  # 4K  ≤13 Mbps → use aggressive encoding
       ((2560, 1440),   8_000_000),  # 1440p
       ((1920, 1080),   5_000_000),  # 1080p
       ((1280, 720),    2_500_000),  # 720p
       ((854, 480),     1_200_000),  # 480p
   ]

Thanks!

Do you notice a difference if you encode using your GPU vs CPU? I've heard people say NVENC doesn't encode as efficiently as x265 but I haven't actually tried it out.

NVENC can obviously encode much faster, but for archive I'd probably prefer a better compression ratio if thats on the cards.

I use “slow” GPU encoding. I optimized my pipeline for minimal quality loss, not maximum file size savings. I wouldn’t recommend using consumer grade CPUs to reencode because to get a reasonable speed you’ll need to use “fast” presets which tank the quality. Anecdotally fast CPU method saves closer to 60-70% file size.

Mac can run larger models due to the unified memory architecture. Try building a 512GB nvidia VRAM machine. You basically can’t.

Fully aware, but who the heck wants to spend nearly 10 grand, and that's with just a 1TB hard drive (which needs to be able to fit your massive models mind you). Fair warning not ALL the RAM is fully unified. On my 24GB RAM Macbook Pro I can only use 16GB of VRAM, but its still better than me using my 3080 with only 10 GB of RAM, but I also didn't spend more than 2 grand on it.

Good on you! Writing assembly is a good way to understand how computers work, which can help you further up the stack.

Assembly will not help you further up the stack which is working with agents, not writing code (obsolete skill). Apparently my /s was needed

I got your poor attempt at sarcasm. I just don't think it's a good argument.

The person who understands how lower levels of abstraction work, will always run circles technically around those who don't. Besides, "AI" tools are not a higher level of abstraction, and can't be compared to compilers. Their goal is to replace all cognitive work done by humans. If you think programming is obsolete, the same will eventually happen to whatever work you're doing today with agents. In the meantime, programmers will be in demand to fix issues caused by vibe coders.

And I got your cheeky, dismissive attitude which completely misses the forest for the trees.

> In the meantime, programmers will be in demand to fix issues caused by vibe coders.

Yes, I agree. They’ll be lower on the totem pole than the vibe coders, too. Because the best vibe coders have the same skill set as you - years of classical engineering background. So how can one differentiate themself in the new world? I aspire to move up the totem pole, not down, and leaning into AI is the #1 way to do that. Staying a “bug fixer” only is what will push you out of employment.

This is a very flawed analogy.

I don’t think Ralph is worthwhile, at least the few times I’ve tried to set it up I spent more time fighting to get the configuration right than if I had simply run the prompt. Coworkers had similar experiences, it’s better to set a good allowlist for Claude.

> Teams is the defacto IM app for enterprise now.

Slack has more mindshare

For the 1000+ headcount companies who sit outside the Silicon Valley webdev/software dev world, it doesn't. Silicon Valley looks at these as "products". Purchasing managers see these as "commodities" that need to be interoperable with the rest of their stack first.

> Society was been better without the internet

I don't even follow the reasoning of arguing this counterpoint, you are literally only able to make this argument because the internet even exists.

That's a "yet you participate in society" argument. It's not at all contradictory to use this communication medium to describe my perception of its negative impacts.

It’s actually worse than slop, because they aren’t capable of producing slop, because they don’t understand the basics of version control or even how to talk about code concepts, so they require additional training and onboarding before they can even start producing claude-slop…

If they're that incompetent and talking to customers, you are losing a lot of customers that are turned off by the incompetence.

> Projects like Zig may never allow ai contributions

Good luck enforcing that.

This extension is solving for the wrong problem and is actually only useful as some kind of ideology cudgel, it literally can only create friction. Nobody important cares if code is ai generated, they care if it solves problems correctly.

This is one of those foods in my world which just kind of “exists” without consideration that it’s actually eaten by some people as a preference. Other candidates include: Campbells tomato soup, pork rinds, and ranch dip.

Two of these things are not like the others. Pork rinds and ranch both vary in quality from mass produced crap that I would actively avoid to delicious products that I would seek out. In particular a really tangy buttermilk ranch (or even a really lemony ranch) with lots of black pepper and freshly minced herbs is supremely tasty.

They can submit those resumes to other positions.

This type of recruiter often is paid based on # of resumes furnished, not actually filling the position, so the incentives are completely misaligned.