Unverified50% confidenceFactExact time

The parallel execution model of CUDA within a GPU is structured in layers of Grid, Block, and Thread, with the Shared Memory shared by each Block being more than ten times faster than global VRAM but typically only a few tens of KB in capacity.

1

Sources

50%

Confidence

Long-term

Relevance

10/4/2026

First Seen

Sources

DeepSeek Open-Sources Six Core Components: Laying the Foundation for Domestic AI Chips

Related Entities

CUDAtechnology GPUtechnology

Related Claims

UnverifiedGPU上下文切换需保存完整GPU状态（包括寄存器文件、共享内存和L1缓存），数据量可达数百MB，而CPU上下文切换仅需保存几KB79% similar VerifiedThe GPU memory hierarchy from fastest to slowest is: registers, L1 cache/shared memory, L2 cache, global memory (VRAM).77% similar Verified统一内存架构使CPU、GPU与NPU共享同一块内存池，避免数据在处理单元间反复搬运的开销76% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/973404

API

curl https://kongchang.com/api/v1/knowledge/claims/973404

MCP

get_claim(id=973404)