English(EN) Similar to the panic over DeepSeek R1, some uneducated people think Kimi K3’s use of linear attention (KDA) is bad for NVIDIA, HBM, DRAM, and networking because

SemiAnalysis称Kimi K3模型的规模和效率将提振AI硬件需求

作者 PulseAugur 编辑部 · [8 个来源] · 2026-07-17 22:49

SemiAnalysis认为，尽管Kimi K3模型采用了线性注意力和较低的KV缓存需求，但它对NVIDIA和更广泛的AI硬件生态系统是有益的。该模型拥有庞大的2.8万亿参数，需要大规模的基础设施，包括NVL72等专用机架，其WideEP优化虽然增加了网络带宽需求，但却是为这类系统设计的。此外，杰文斯悖论表明，AI效率的提高最终将导致更广泛的应用，从而增加对GPU、HBM、DRAM和网络基础设施的需求。 AI

影响表明AI模型效率的提高将推动对包括GPU和网络基础设施在内的AI硬件的更广泛应用和需求。

排序理由该集群由SemiAnalysis对Kimi K3模型对硬件需求影响的分析和观点组成，而非直接的产品发布或公告。

在 X — SemiAnalysis 阅读 →

AI 生成摘要 · Google Gemini · 来自 8 个来源。我们如何撰写摘要 →

报道来源 [8]

X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

Lastly, Jevons’ Paradox means that making attention more efficient will drive wider AI adoption, which will ultimately require more GPUs, HBM, DRAM, and network

Lastly, Jevons’ Paradox means that making attention more efficient will drive wider AI adoption, which will ultimately require more GPUs, HBM, DRAM, and networking—not less. 8/8🧵
X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

Kimi themselves have stated that optimal K3 inferencing will require a rack witha n large scale up domain with at least 64 chips. 7/8🧵 https://t.co/UpFo0yIG01

Kimi themselves have stated that optimal K3 inferencing will require a rack witha n large scale up domain with at least 64 chips. 7/8🧵 https://t.co/UpFo0yIG01
X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

Furthermore, since the weights occupy more than 1.5 TB of HBM capacity, the KV cache for K3’s KDA and Gated MLA will need to be offloaded to CPU DDR5 and NVMe,

Furthermore, since the weights occupy more than 1.5 TB of HBM capacity, the KV cache for K3’s KDA and Gated MLA will need to be offloaded to CPU DDR5 and NVMe, even at relatively low user concurrency, because little space remains in HBM. 6/8🧵
X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

The unfortunate downside of the WideEP optimization is that it consumes a tremendous amount of network bandwidth. WideEP is highly optimized for rack-scale syst

The unfortunate downside of the WideEP optimization is that it consumes a tremendous amount of network bandwidth. WideEP is highly optimized for rack-scale systems like the GB200/GB300 NVL72, whose copper backplane provides 18× more bandwidth than comparable DGX B200 systems. htt…
X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

WideEP distributes the 896 experts across many GPUs so that each GPU’s HBM contains only a small number of experts, optimizing per-token memory usage and comput

WideEP distributes the 896 experts across many GPUs so that each GPU’s HBM contains only a small number of experts, optimizing per-token memory usage and compute utilization. 4/8🧵 https://t.co/f6v16blXDQ
X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

Secondly, although Kimi Delta Attention has up to 10× lower networking requirements for KV-cache transfers, its large weights require even more network bandwidt

Secondly, although Kimi Delta Attention has up to 10× lower networking requirements for KV-cache transfers, its large weights require even more network bandwidth to implement an optimization called WideEP, which spreads the weights across different GPUs. 3/8🧵 https://t.co/68Y0pmw…
X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

Kimi K3 is actually quite positive for NVIDIA, as large-model inference is where the NVL72 shines. Because K3 has more than 2.8 trillion parameters, it requires

Kimi K3 is actually quite positive for NVIDIA, as large-model inference is where the NVL72 shines. Because K3 has more than 2.8 trillion parameters, it requires a large scale-up domain to store its weights. 2/8🧵 https://t.co/ykZZZFGLWp
X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-17 22:49

Similar to the panic over DeepSeek R1, some uneducated people think Kimi K3’s use of linear attention (KDA) is bad for NVIDIA, HBM, DRAM, and networking because

Similar to the panic over DeepSeek R1, some uneducated people think Kimi K3’s use of linear attention (KDA) is bad for NVIDIA, HBM, DRAM, and networking because it has relatively lower KV-cache requirements. The opposite is true, and we explain why below. 👇️ 1/8🧵 https://t.co/ih3…

报道来源 [8]

相关实体

相关话题