[KongchangAI]
Unverified50% confidenceFactExact time

In the GPU memory hierarchy, SRAM is fast but has extremely limited capacity, while HBM has large capacity but relatively limited bandwidth.

1
Sources
50%
Confidence
Long-term
Relevance
9/17/2026
First Seen

Sources

Why Does KV Cache Blow Up Your GPU Memory? A Deep Dive into LLM Inference Optimization

Related Entities

SRAMconceptHBMconceptGPUconcept

Related Claims

Unverified片上SRAM提供了远超HBM的访问速度和带宽，消除了芯片到外部内存之间的数据搬运瓶颈76% similarVerifiedThe GPU memory hierarchy from fastest to slowest is: registers, L1 cache/shared memory, L2 cache, global memory (VRAM).74% similarUnverifiedWSE-3的片上SRAM容量达到44GB，SRAM访问延迟约1ns，比HBM约100ns低10-50倍73% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/930573
API
curl https://kongchang.com/api/v1/knowledge/claims/930573
MCP
get_claim(id=930573)