Unverified50% confidenceFactExact time
Strata organizes three tiers of storage—VRAM, system memory, and NVMe SSDs—into a pipeline, enabling consumer GPUs with 12GB to 16GB of VRAM to run a 125B-parameter MoE model.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Running 125B MoE on a 12GB GPU? Inside Strata's Three-Tier Cooperative Engine
bilibiliNicksxs11/3/2026
Related Entities
Related Claims
UnverifiedWith a 12GB VRAM graphics card and a model download of about 70GB, the 125-billion-parameter Strata large model can be run locally.79% similarUnverifiedThe author of Strata explicitly gives three hard hardware requirements: graphics card VRAM of 12GB or more, system memory starting at 32GB, and at least 80GB of disk space reserved (about 70GB for the first download).76% similarUnverifiedStrata converts the entire computer into a three-tier memory hierarchy: VRAM holds what each token touches, system memory pins all approximately 24,000 groups of experts, and the SSD stores a 29 GB N-gram lookup table.73% similarUnverifiedOn a 16GB GPU with an 8K context and multiple concurrent streams, the KV cache alone can eat several GB of VRAM.70% similarUnverifiedThe Strata engine keeps the main body of the model resident in system memory and dynamically loads only the layers needed for the current inference into VRAM, enabling a computer with 12GB of VRAM to run a 125-billion-parameter model.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134843API
curl https://kongchang.com/api/v1/knowledge/claims/1134843MCP
get_claim(id=1134843)