Unverified50% confidenceBenchmarkExact time
Relying on a compact MoE architecture with 12B activated parameters and optimized NVFP4 kernels in TensorRT-LLM, Nemotron generates 185 tokens/second on a single Blackwell B200 node, up to 3.2 times faster than dense competitors.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/16/2026
First Seen
Valid until: 10/30/2026
Sources
Related Entities
Related Claims
UnverifiedSGLang using official NVFP4 weights with D-Spark speculative decoding achieved an average of approximately 173 tokens/second on Blackwell GPUs, approaching 200 for short tasks, while loading a 262k context window.74% similarUnverifiedllama.cpp通过纯CPU推理和4-bit GGUF量化格式，使7B参数模型在无GPU的普通服务器上达到每秒20-40 tokens的生成速度68% similarUnverifiedWhen running 4-bit DeepSeek R1 on the M3 Ultra, a 1,000-token context generates approximately 19 tokens per second using 392GB, dropping to approximately 15 tokens per second using 410GB at a 16,000-token context.67% similarUnverifiedA 70B-parameter model (140 GB of weights) has a theoretical maximum generation speed of approximately 7 tokens/second at 1 TB/s bandwidth.67% similarUnverifiedOpenTPU 运行 LFM 2.5 230M 模型可达到约 59 tokens/秒，在 4-bit 量化下提升到 85.8 tokens/秒67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047848API
curl https://kongchang.com/api/v1/knowledge/claims/1047848MCP
get_claim(id=1047848)