KUN Field Notes
How far can a 6 GB-VRAM laptop push local LLMs?
Every video has a companion write-up with the exact commands, flags, hardware and raw results.
Latest video

24B on a 6GB laptop: tune -ncmoe, three tasks all pass
Most popular

A 35B MoE on 6 GB of VRAM? 64 GB of RAM Made It Work
With enough system RAM, a 35B-class MoE model can be practically usable with 6 GB of VRAM, provided its expert weights stay on the CPU.

24B on a 6GB laptop: tune -ncmoe, three tasks all pass
On an RTX 3060 Laptop (6GB) with 64GB RAM, Liquid AI LFM2.5-24B-A2B (Q4_0) runs locally. Default -ncmoe 40 uses ~1.3GB VRAM; -ncmoe 30 is faster end-to-end and still passes all three tasks—including the bookkeeping case a 35B run once failed. Thinking mode will not stay off.
APEX + MTP on a 6 GB laptop: a 39% median generation gain, at what cost?
Across 12 tasks using identical APEX weights, MTP gave a 39% median task-by-task generation-speed gain on my RTX 3060 Laptop (6 GB VRAM, 64 GB RAM). Median peak VRAM rose by 688 MiB; changed answers and retrieval failures are included below.
Latest articles
All articles
763B DeepSeek vs a 6 GB laptop: Strata's planner reports no usable GPU
Strata, a third-party installer for DeepSeek models (Niko1221/Strata), ships a `--plan` mode that computes a run plan and exits without downloading anything. On this 6 GB laptop all three quantization tiers (1.75 / 2.00 / 2.25 bpw) returned the same verdict: no usable graphics card (the floor is 8 GB; docs were measured on 12 GB) and does not fit, with zero experts placed on the GPU. Downloads would be 239.3 / 256.8 / 273.8 GB. The plan is a soft refusal — it does not block startup, but expects the run to be very slow or to run out of memory. No speed was measured here; the body only cites second-hand references.

Flint 40B on 6 GB of VRAM: on a hard prompt, no budget means reasoning only
On the same RTX 3060 laptop (6 GB VRAM), Flint 40B-A2.5B ran against Qwen 3.6 35B. Speed and benchmarks were re-measured on the same day and build (b11889): chat 21.6 vs 22.2 tok/s, essentially tied. Quality comes from the previous run-004 (older build, thinking off); on the four comparable tasks both models scored 3 passes and 1 partial, with Flint 1.2–2× slower per task. On one hard prompt, without `--reasoning-budget`, Flint produced 13,364 characters of reasoning and an empty answer before being cut off at 439 s; with 1024 it answered in 207 s at 11.9 tok/s decode.

A 6 GB VRAM laptop test: NVIDIA's Nemotron 3.5 30B scores 14/21 against Qwen 3.6 35B's 18/21
On the same RTX 3060 laptop (6 GB VRAM, about 64 GB RAM), NVIDIA's Nemotron 3.5 Lightning 30B-A3B took on Qwen 3.6 35B-A3B. On 21 fun-v1 capability tasks Nemotron passed 14 with 2 partial and 5 failures; Qwen passed 18 with 2 partial and 1 failure. llama-bench generation: 16.1 vs 19.2 tokens/s (3-round mean). Nemotron's Chinese commonsense results included a book-author mix-up and an invented source. One round per task — these results describe this single run.
Like hands-on tests you can reproduce?
New videos land on YouTube first; the notes follow here.