Expired50% confidenceFactExact time
When scaling to tens of thousands of GPUs, overall compute efficiency can plummet below 40%, with more than half of compute power consumed by data transfer and waiting — a phenomenon known as the 'communication wall.'
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/30/2026(expired)
Sources
Marvell: The Hidden Giant Behind AI Infrastructure — Four Technical Aces Explained
bilibili中天评论6/6/2026
Related Claims
UnverifiedWhen AI clusters scale to tens of thousands of GPUs, overall efficiency can plummet to below 40%, with more than half of computing power consumed in data transmission and waiting.80% similarUnverified对于MuJoCo、Atari等环境，单个环境的采样速度往往跟不上GPU计算能力，导致GPU利用率长期停留在20%–40%甚至更低68% similarUnverifiedDuring the autoregressive decoding phase, GPU compute utilization is often less than 10%, making it a memory-intensive operation.65% similarUnverifiedGPU utilization can rise from 20%–30% during low traffic to 70%–80% during high traffic, because larger request volumes make it easier to aggregate into larger compute batches64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/49977API
curl https://kongchang.com/api/v1/knowledge/claims/49977MCP
get_claim(id=49977)