Unverified50% confidenceBenchmarkExact time
On the B200, a single 3-question guardrail check was reduced from 11.6 ms to 2.8 ms, and Laya's loading time was shortened from approximately 17 seconds to approximately 9 seconds
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/1/2026
First Seen
Valid until: 12/30/2026
Sources
Related Entities
Related Claims
UnverifiedOn a T4 GPU, Laya takes 33 milliseconds per single question, and batch processing can compress that to 7 milliseconds75% similarUnverifiedAfter switching to Plan B, scene loading time was reduced from 3.3 seconds to approximately 0.5 seconds, approximately 6.8 times faster.74% similarUnverifiedAfter changing the VAE chunked frame count from 1024 frames to 512 frames, the processing time for the same request dropped from 613 seconds to 393 seconds, a 35.9% speedup, with VAE decoding reduced from 313 seconds to 107 seconds.70% similarUnverifiedOn the B200, the per-step time for LoRA SFT training of Qwen3.5-9B decreased from 0.84 seconds to 0.77 seconds.69% similarUnverifiedText-to-image generation is noticeably slow on the first run; in testing, the first image took 188 seconds, the second dropped to 103 seconds after the cache took effect, and the theoretical fastest speed can reach one image every 14 seconds.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/964987API
curl https://kongchang.com/api/v1/knowledge/claims/964987MCP
get_claim(id=964987)