IBM Granite 4.2 Fits 128GB. That Is Not a Bench.
IBM posted Granite 4.2 on Aug 25: Apache 2.0, dense 3B/8B/30B, thinking switch, 128K native. Official 30B BF16 is 58.55 GB of weights. That fits a GX10. I have not timed it.

> I run open-weight models on boxes I own — NVIDIA Spark, ASUS GX10, Grok, local agents — then write what worked. No brochure. Games included.
Local AI leads: open-weight models on hardware I actually own, plus Grok, vibe coding, and the games that come out of that stack. Marketing archives stay in the hubs — they are not the front door.
Open-weight models, local inference stacks, VRAM planning, and homelab setups for running AI on your own hardware.
Open hub From the labSmall browser games shipped from the same stack — including Moon Trail, built with a local agent.
Open hubIn-depth reviews, comparisons, and guides for the latest AI tools reshaping how we work and create.
Open hubSEO strategies, content marketing insights, and growth tactics for the AI-powered marketing landscape.
Open hubBreaking developments in AI, cloud computing, data centers, and the future of technology.
Open hubOpen WebUI tagged v0.11.1 on Aug 25. Admins can pause each tool call for allow or deny. Automations and temp chats stay out. Their 1000x stream claim is theirs.
Read full articleIBM posted Granite 4.2 on Aug 25: Apache 2.0, dense 3B/8B/30B, thinking switch, 128K native. Official 30B BF16 is 58.55 GB of weights. That fits a GX10. I have not timed it.
ggml-org tagged llama.cpp v0.3.0 on Aug 25. It adds dots3-note, GLM-4.5-Air MTP, DeepSeek 4 -sm tensor, and a moov-at-EOF video fix. A GitHub tag is not a GX10 number.
NVIDIA posted Aug 24: SpaceXAI will deploy Vera CPUs, expand Grok on Vera Rubin toward gigawatts, and plans a Starmind satellite on NVL72. SpaceXAI did not post it. A press release is not a receipt.
SpaceXAI posted Grok 4.6 on Google's Model Garden Aug 21. Google's card says Preview, grok-4.6, global endpoint, 13 QPM. Same $2/$6 as Bedrock. Still not a bench.
Local AI, vibe coding, real hardware
Open-weight models on hardware I own. Grok and local agents when they earn it. Skeptical of the brochure.
Read Full BioHardware unboxings, local-model runs, Grok experiments, and the occasional cookie bet.
Follow @wikiwayne