Unverified50% confidenceFactExact time
On-chip SRAM access bandwidth can reach hundreds of TB/s, whereas the most advanced HBM3e can only provide approximately 8 TB/s of bandwidth.
1
Sources
50%
Confidence
Long-term
Relevance
8/8/2026
First Seen
Sources
软件优化让单张B200媲美专用推理芯片
Related Claims
VerifiedHBM3e single-stack bandwidth can reach 1.2 TB/s, which is 4–9 times the bandwidth of conventional GDDR6.79% similarVerifiedA100 GPU的HBM带宽约2TB/s，而片上SRAM带宽可达19TB/s，相差近10倍79% similarUnverifiedGPU HBM bandwidth is typically on the order of 2–3 TB/s (3.35 TB/s for H100), while the effective bandwidth of Host DRAM connected via PCIe 5.0 is only about 64 GB/s—a difference of nearly 50x.78% similarUnverifiedNVIDIA's H100 is equipped with 80 GB HBM3, delivers memory bandwidth of 3.35 TB/s, and NVLink 4.0 provides 900 GB/s bidirectional bandwidth.76% similarUnverifiedNVMe SSD顺序读取带宽通常为7-12 GB/s，而GPU HBM带宽为2TB/s量级76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/707600API
curl https://kongchang.com/api/v1/knowledge/claims/707600MCP
get_claim(id=707600)