Friday, June 05

Geek

Daily News Stuff 5 June 2026

Lilo Pelekai Multipass Edition

Top Story

  • AI token costs are becoming a meme.  Here's why that's a good thing.  (Tom's Hardware)

    It's not.  I mean, not for the AI companies, and not for anyone else unless it pops the bubble, and not even then as trillions of dollars of virtual money suddenly disappearing would cause a certain amount of drama.
    Despite that, Altman projects that AI token usage will continue to increase.  He said that six-and-a-half years ago, the top token spender at the startup used 100,000 tokens a month - today, that is the global per capita average token usage, and that OpenAI’s token leader uses about 100 billion a month.  The OpenAI chief also admitted, to his own embarrassment, that someone else uses even more. So, if token usage were to grow linearly, then he would expect the global per capita token usage to hit 100 billion monthly.
    Somehow I don't think that will happen.  At OpenAI's current rates, that would cost the average user over a million dollars a month and provide the company with a quadrillion dollars in monthly revenue.


Tech News

Musical Interlude





Disclaimer: Big bada boom means no one gets left behind.

Posted by: Pixy Misa at 06:05 PM | Comments (3) | Add Comment | Trackbacks (Suck)
Post contains 347 words, total size 3 kb.

1 I ordered a Minisforum DEG1 and PSU yesterday to go with my AI X1, now that Google's released a model that will fit on a 16GB GPU (my RX 6800).  Be interesting to see how that works.  If I understand correctly I'll be able to download one of two apps (LM Studio or ollama, or something like that) and the weights (18GB) and then I'll be able to run an LLM locally?  The way I look at it is it'll only cost me $280 or so and even if it's not what I want...I'll be able to game on the eGPU.  Stuff I found last night suggests this combo will run the GPU at about 90-95% of it's full bandwidth.

Posted by: Rick C at Friday, June 05 2026 11:16 PM (1zWbY)

2 I haven't tried it myself (I have a DEG1 but it's still in its box) but that should work fine.  The bandwidth limit of OCuLink would reduce game performance by something like 10% (depending on the title, but in that ballpark), but once an AI model is loaded it would run at very close to full speed - 95% or better.

Posted by: Pixy Misa at Saturday, June 06 2026 08:23 AM (PiXy!)

3 Yeah, that 10% figure feels like what I've seen.  Early afternoon my son mentioned that someone in our building was going to the local rodeo that started yesterday and was inviting people to come, so my plans got cut short.  But I did run the latest Unigine, which ran well, and played about half an hour of Guild Wars 2, same.

Normally you have to turn on an eGPU and then turn on the mini PC, but Minisforum's done something so that if you pair a DEG1 with a Minisforum mini PC, you leave the power on in the dock, and when you power on the mini PC, it tells the dock to wake up, so it saves you the possibility of accidentally starting up wrong.

I'm on it right now.  It actually works really well, except I'm not positive the fans are spinning when gaming--the system did freeze up on me once.  Looking into that today is on the table, as is starting to get an AI environment going.  I do have some errands to run tho, too.
Coincidentally, Microcenter's sale flyer yesterday mentioned Arc Pro B70s (with 32GB of VRAM) on sale but I'm not ready to drop $950 yet.

Posted by: Rick C at Sunday, June 07 2026 01:47 AM (1zWbY)

Hide Comments | Add Comment




Apple pies are delicious. But never mind apple pies. What colour is a green orange?




[f#^_^sfleeb#f.stats]