Unverified50% confidenceFactExact time
The GPU resource granularity of the Kubernetes Device Plugin is fixed at whole cards, and cannot express requests for half a card or a specific amount of VRAM.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGPU不支持超卖，一张卡要么整卡分配要么不分配，除非使用MIG或MPS等虚拟化技术75% similarUnverifiedIn GPU sharing schemes, merely placing multiple Pods on the same graphics card at the scheduling layer provides no real isolation at the memory level, and a single process may exhaust the entire card's memory and affect other tenants on the same card.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1105261API
curl https://kongchang.com/api/v1/knowledge/claims/1105261MCP
get_claim(id=1105261)