1 - So GLM 5.2 is 700b parameters (ish)2 - 4x DGX Sparks can supposedly handle up to 700b parameters (give or take)3 - GLM 5.2 is supposedly in striking distance of the performance of GPT 5.5 and Opus 4.8. In my brief tests, it's really not shabby at all.4 - So for $20k, you can get near the frontier on your table.5 - Extrapolate the trend, and you could have mythos/5.5 pro - class models in your dining room for the cost of a cheap car less than five years from now. Even without extrapolation, we're already the near frontier running locally.6 - Paying real api costs, I could easily blow through $3,000 per month coding and running agents. The machine pays for itself in 6-7 months conservatively.7 - In 3-5 years, most power users of AI will self-host.8 - Am I missing something?
>>109078190The cost and availability of electricity. By availability I don't mean the grid running dry but the connection to and wiring in your house not being sufficient.
>>109078257poorfag
You are correct. Despite best effort, mainframes and dumb terminals wont return. The reality of the datacenter rollout will hit in the next few years (too many, too overpriced, cost not worth it as local models become preferable in many circumstances). The, for a while, we'll have org/business wide deployments, before consumer AI hardware will become common place. The AI financial bubble is almost going to be identical to the dot com one: overfinancialization during development into widespread cheap practical adoption. Could take a decade though.
AGI will be a 500 trillion param model (it'll be done in around 2 weeks)
>>109078190so wheres your order for 4 DGX sparks then?You will make your money back early 2027
feet
>>109078190>Extrapolate the trend, and you could have mythos/5.5 pro - class models in your dining room for the cost of a cheap car less than five years from nowYou will have mythos class models running on a 3090 in a year.
>>109078190For me, it's armpit.
>>109078190Nobody will be buying dedicated hardware for that, it will be scaled to run on phones and laptops. You simply can't convince most people to pay for AI. The use cases for AI will crystalize and become local, yes. It will be as ubiquitous as syntax highlighting but it won't change the world.
>>109078190i think it's a good idea, OP anon. post pics when you get it running for the rest of us plebs to see.
>>109078214Fpbp!!!!!!
>>109078190>2 - 4x DGX Sparks can supposedly handle up to 700b parameters (give or take)You can't run GLM 5.2 at Q8 on that.
>>109082003And I don't think you'll have enough context for this one either.
>>109082010lmao q1
>>109082010You could run this on 4, it's an MoE so the inference memory and KV cache will be small
>>109078257"DGX Spark's peak total system power is 240W." So it's less than 1kW for four of them, that's less than a toaster, you'd have to live in some third world country to not have wiring that supports that.
