If you want to look at it another way, if you assume every single square inch of silicon from TSMC is Nvidia server accelerators/AMD EPYCs, every single one running AI at full tilt 24/7/365…
Added up, it’s not that much power, or water.
That’s unrealistic, of course, but that’s literally the physical cap of what humanity can produce at the moment.
I’m not sure what you’re referencing. Imagegen models are not much different, especially now that they’re going transformers/MoE. Video gen models are chunky indeed, but more rarely used, and they’re usually much smaller parameter counts.
Basically anything else machine learning is an order of magnitude less energy, at least.
Yeah, that is so perfect.
Imagine a sedan or hatchback. It would be light as a feather (in terms of curb weight) and still feel spacious being so ‘clean’ inside.
On the contrary, custom engines have been bombing.
Look at Starfield or Cyberpunk 2077 or basically any custom engine AAA. Look at what happened to things like ME Andromeda.
…Then look at KCD2. It looks freaking fantastic, looks like raytacing with no raytracing, runs like butter, and it’s Crytek.
Look at something like Satisfactory, rendering tons of stuff on a shoestring budged and still looking fantastic thanks to Unreal Lumen.
There’s a reason the next Cyberpunk is going to be Unreal, and its because building a custom engine just for your game is too big an undertaking. Best to put that same budget in optimizing a ‘communal’ engine, polish, bugfixing and such.
Borderlands 4 is slow because the botched optimization, not because its Unreal.
The sad thing is ‘smartphone on wheels’ is a slur.
Smartphones don’t have to be soulless and uniform and enshittified and subscription based and completely inaccessible and straight up anti-consumer/designed to fail, but here we are.
I really hope Slate takes off though (and they make a nimble hatchback for their next chassis). It feels like the antithesis of all this.
+1 for immutable in general.
Yeah, that’s problematic, heh.
To be fair, I do wish more privacy-friendly browsers took DDG mobile’s approach, namely torch all sites but make it really easy to (and prompt you to) whitelist frequently used ones.
You can permanently disable the chatbot in full DDG search. Click the little gear.
It does make me wonder what API they use. I thought it was huggingface (which would be less bad), but they don’t say it explicitly.
Problem is our billionaires are all in. And also, their fortune comes from the propaganda machines (social media).
It’s not great.
It would help if we become the next North Korea though.
And any divergence from that is “ruining games” or “being woke” to the point that we don’t even GET those games outside of the rare case of a game nobody cared about becoming popular
I would argue the origin is sales. E.G. the publisher wants the sex appeal to sell, so that’s what they put in the game. Early ‘bro’ devs may be a part of this, but the directive from up top is the crux of it.
And that got so normalized, it became what gamers expect. And now they whine like toddlers when anyone tries to change it, but that just happens to be an existing problem conservative movements jumped on after the fact.
TL;DR the root cause is billionares.
Like aways.
Run a giant LLM in RAM, heh:
https://huggingface.co/zai-org/GLM-4.5
Seriously though, I had 32GB, and was having problems with all sorts of media work, complilation, stuff that would grind my system to a crawl constantly from the paging. So I wanted to replace them with 64GB sticks, but the 128GB modules were low voltage, low latency and not that expensive in terms if $/GB… and I realized it was enough space to basically host a reasonable quantization of Deepseek at home, at (as it turns out) close to reading speeds.
I like having a RAM drive to put certain games on, too. Not that I game a ton these days… Hence dangrous’s meme.
Yeah. But it also messes stuff up from the llama.cpp baseline, and hides or doesn’t support some features/optimizations, and definitely doesn’t support the more efficient iq_k quants of ik_llama.cpp and its specialzied MoE offloading.
And that’s not even getting into the various controversies around ollama (like broken GGUFs or indications they’re going closed source in some form).
…It just depends on how much performance you want to squeeze out, and how much time you want to spend on the endeavor. Small LLMs are kinda marginal though, so IMO its important if you really want to try; otherwise one is probably better off spending a few bucks on an API that doesn’t log requests.
In case I miss your reply, assuming a 3080 + 64 GB of RAM, you want the IQ4_KSS (or IQ3_KS, for more RAM for tabs and stuff) version of this:
https://huggingface.co/ubergarm/GLM-4.5-Air-GGUF
Part of it will run on your GPU, part will live in system RAM, but ik_llama.cpp does the quantizations split and GPU offloading in a particularly efficient way for these kind of ‘MoE’ models. Follow the instructions on that page.
If you ‘only’ have 32GB RAM or less, that’s tricker, and the next question is what kind of speeds do you want. But it’s probably best to wait a few days and see how Qwen3 80B looks when it comes out. Or just go with the IQ4_K version of this: https://huggingface.co/ubergarm/Qwen3-30B-A3B-Thinking-2507-GGUF
And you don’t strickly need the hyper optimization of ik_llama.cpp for a small model like Qwen3 30B. Something easier like lm studio or the llama.cpp docker image would be fine.
Alternatively, you could try to squeeze Gemma 27B into that 11GB VRAM, but it would be tight.
How much system RAM, and what kind? DDR5?
ik doesn’t have great documentation, so it’d be a lot easier for me to just point you places, heh.
What’s interesting is they file this in the US.
Is there a reason they don’t file patents in Japan instead?
Original source:
I hate to be nitpicky, but try to link the original, and post where you found it (GameRant) in the description instead.
At risk of getting more technical, ik_llama.cpp has a good built in webui:
https://github.com/ikawrakow/ik_llama.cpp/
Getting more technical, its also way better than ollama. You can run models way smarter than ollama can on the same hardware.
For reference, I’m running GLM-4 (667 GB of raw weights) on a single RTX 3090/Ryzen gaming rig, at reading speed, with pretty low quantization distortion.
And if you want a ‘look this up on the internet for me’ assistant (which you need for them to be truly useful), you need another docker project as well.
…That’s just how LLM self hosting is now. It’s simply too hardware intense and ad hoc to be easy and smart and cheap. You can indeed host a small ‘default’ LLM without much tinkering, but its going to be pretty dumb, and pretty slow on ollama defaults.
Agreed.
I think cars should be more individualistic, which is part of Slate’s pitch. So should men’s fashion expectations.
That being said, a lot of truck lifting is very similar and more “I’m signicaling I’m part of this culture,” than doing it for the sake of being cool. Some people do it as a form of expression, but that’s the exception in my experience.
You know what’s missing there?
An apology.