Verified75% confidenceFactExact time
在H100 NVLink架构中，单台服务器内8块GPU可通过NVSwitch全互联，总带宽达900GB/s，相比PCIe 5.0的64GB/s有数量级提升
3
Sources
75%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
TensorRT多设备推理：突破单GPU显存瓶颈的工程实践
rss6/25/2026
Related Claims
VerifiedNVLink 4.0的双向带宽高达900GB/s，而传统PCIe总线带宽约为64GB/s83% similarUnverifiedNVLink 4.0 (used in the H100) can achieve a unidirectional bandwidth of 900 GB/s.83% similarUnverifiedThe GB200 NVL72 rack interconnects 36 Grace CPUs and 72 Blackwell GPUs via NVLink 5.0, with a bidirectional NVLink bandwidth of 1.8 TB/s.82% similarVerifiedNVIDIA's NVLink chip-to-chip high-speed interconnect bandwidth can reach over 900 GB/s, building an interconnect ecosystem barrier through NVLink and NVSwitch.81% similarUnverified单台DGX H100服务器内8块GPU的双向带宽可达900GB/s80% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/461710API
curl https://kongchang.com/api/v1/knowledge/claims/461710MCP
get_claim(id=461710)