{
  "id": 21046286,
  "title": "Homelabber’s second-gen Optane build uses two RTX 3090s to run 2.8-trillion-parameter Kimi AI model at 3.11 tokens per second",
  "url": "https://urgent.news/2026/11/21/homelabbers-second-gen-optane-build-uses-two-rtx-3090s-to-run-2-8",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-21T12:00:00.000Z",
  "source": {
    "name": "Tom's Hardware",
    "slug": "tom-s-hardware",
    "url": "https://www.tomshardware.com/tech-industry/artificial-intelligence/homelabbers-second-gen-optane-build-runs-2-8-trillion-parameter-kimi-k3-at-3-11-tokens-per-second-after-adding-two-rtx-3090s-moving-nearly-half-the-model-into-a-ram-drive-boosted-performance-by-41-percent"
  },
  "original_language": "en",
  "account": "A homelabber shared updates on their second-generation Optane AI build on Reddit, adding two RTX 3090 graphics cards to run the Moonshot Kimi K3 model with 2.8 trillion parameters at a speed of 3.11 tokens per second. By transferring 45 out of 94 shards of the model to a RAM drive, generation speed improved by about 41% and prompt evaluation by 26%. However, the extra fans were necessary as the persistent Optane memory got quite hot. The build, which used llama.cpp and DAX for the Optane portion, showcased that the GPUs were not holding the entire model, as they had only 48GB of VRAM between them, less than 5.5% of the 865GB model file. The model used a mixture-of-experts (MoE) architecture, where only a few experts ran per token, with 16 chosen per token, two shared, and an active 104B parameter count. The RTX 3090s handled attention and dense feed-forward network layers, while the Optane memory managed the routed-expert tensors. The new configuration, with the model partly in a RAM drive, resulted in faster generation and prompt evaluation speeds. Despite improvements, the room for further optimizations remained, such as customized quantization, which could potentially enhance performance even further.",
  "summary": "A Reddit homelabber’s Optane PMem 200 server now runs Moonshot AI’s Kimi K3 at 3.11 tokens/s with two RTX 3090s and a RAM drive.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}