Author: aura.villivirta.music

Take on ram inflation and hardware dreaming.

Computers are getting more expensive. In the manner that they appear jumping to price categories of cars and that feels surprisingly ok.

AI takes computing muscle and tons of memory. Ideally you are not needing any subscription based cloud services for your AI needs, for example media generation. Open models that are sufficiently good for image, music and video generation already exists and are free to use if you have hardware (and skills) to run them on your own computer. Bottleneck is memory and amount of it – hardware is there: Nvidia DGX Spark and Apple Studio M5 ultra, starting from 96GB of memory, going up to 512GB (later in 2026) per device and you can combine several of them, with rumours of upcoming Macs that have up to 1.5TB of unified memory. 96GB has been mentioned as treshold point at which you’re reasonably comfortable running most of the current local AI models if you seek future proofing, more is better and maybe waiting for next generation of processing that updates the memory bandwidth to be able to utilise it all with even less lag.

In this frame, the current norm of 16-24-32GB seems rather low and utterly outdated, and fair in the sense that most people rather pay the 20€ monthly subscription to computing intelligence needs they have.

However, hobbyists who value sovereignty over their computing are feeling the prices. Current entry level AI focused workstations start at 5k and go up to 15k with reasonable capacity of specs, while you can go way higher stacking them up to build an 100k AI rig. Utter overkill monsters for common daily use even with single device, but after having watched bunch of benchmarks on video production, this feels similar to launch of Apple M1 processors. They’re amazing but not yet as lighting fast as you could expect based on the hype. Back then I hoped M1 macbook would do blender rendering great – did it ok, but only at M3 performance started to match what I intuitively expected it to be. M5 Ultra and image generation still appears more sluggish than what MidJourney v5/6 was back in 2024. And that is the level of performance is what feels truly empowering to have at hand in home – to be able to run 2025 MidJourney/Suno/Higgsfield level of output and performance, at home, with single piece of simple “it just works”-hardware and only paying for the electricity to run it.

This will be possible – already is if you are willing to get into server hardware, but the price of that computing is comparable to decent cars. 5k-15k + taxes gets you going without feeling too much frustration or need for under-the-hood knowhow – and you can get it done with less, and this is ridiculous in light that 3k-6k used to be the ceiling for absolutely top of the line best computer you can reasonably get. But looking onwards, it seems that price range is turning into starting point, the cheapest decent new daily driver you can buy. Meanwhile, there is expectation of computers lasting for longer and longer – and I’m happy with that, why not have macbook pros that last 10-20 years and still work as sturdy hand-downs to younger generations like old cars able to run everything else than latest AIs – because that everything else can be done already on your phone. (Slap a keyboard and monitor to your new iPhone and you have computer that puts most consumer laptops to shame by raw performance. )

Got a saving target for now. The devotion to the artistic sovereignty and privacy of creative process come to be pricey principles to live by. This is backed by my personal intution and experience co-creating with AIs so far – ideas that happen in cloud services feel to lose their spark faster than work that is done privately – exposing a book or new film with whole plot, every twist and full lore to cloud servers comes across similar as leaking hyped movie out to pirate sites well before the launch, stealing the thunder from the launch day making the best possible impact by no-one, not a human nor any AI analysing data sitting on server clouds having any idea what it is about. This is the senseation I seek to preserve and cherish, to be able to convey my creative works with full impact, might and surpirse, defined solely on my terms.

Wasting time with the eternal struggle of tuning algorithms to show you something unexpected (while not revealing too much of true yourself to the big data profiling) today took me over to Chinese channels showing hours long anime series / movies all generated with AI during past year. Have to admit I’m impressed by the sheer quality/quantity they’ve reached. Technically, they’re about watchable, could bring myself to sustain interest through one serie, partially because of the novelty of not knowing what “cultivation”-genre is and having the luxury of seeing it for first time without being overburdened by the repeating tropes. This came also from not having watched anime in decades.

Cinematography was good, chracter consistency held, there was a long season arch and shorter episode archs. Technical hiccups didn’t distract too much, writing was passable, pacing worked ok and action scenes looked great holding it together. Overall felt inspired and elevated after gotten invested enough to go through one of them – while clearly the edge of those series happens in platforms like tiktok.

However what really stroke me was the abundance of them. I trust I saw only an peak of an iceberg. Nothing like that in West, even best AI filmmakers I know are not producing on comparable pace, tho best out there appear to have higher ambition, I have no doubts that Asia has their share of more serious and artistic AI-cinema brewing. Now I want to know their production processes and budgets, what tools and workflows they’re using to up my game on this regard.

Meanwhile, been reading some old comics and books off from the library to freshen up and build library of ideas for projects to come. Not their time yet, but I know the spark won’t be there without batteries recharcted with refreshed resources to knit accessories of an story together by hand.

Also, as cultural differences go, how much gaming has influenced these animations is staggering. Of course you have quest logs and sentient AI assistant guiding the adventures – reincarnation is but an portal to next level. Cultivating immortality takes lifetimes.

After frustrated evening, a rant calling for simple solutions.

Apparently what I look for earbuds ain’t on market becase I want them to do less.

All I want are simple earbuds that play music.

No microphone. No touch controls.

Only bluetooth connection to phone that does all the rest. Simple wireless device for listening music without any extra features because those features create problems and by principle, microphones you don’t need are privacy inconvenience. I don’t want my headphones to be communication device, I want them to play audio coming from source and that’s it.

Bonus if they are ip-rated for heavy rain.

apparently that is impossible to get. Simple earbuds without hardware for touch controls or microphone don’t exists because that seems illogical from business sense having less features to market. While for me, this simple device – ideally with physical on-off switch to get rid of unusuability issues of touch controls – would be perfect for daily use and worth the premium price.

Even if willing to pay more for device that does less, would be simple and cheap to make, it is not product thing I can buy.

If someone knows them, let me know!

(Sheesh! microphones and touch – on earbuds! What utter crap is that, we deserve better!)