SemiAnalysis称Kimi K3模型的规模和效率将提振AI硬件需求

报道来源 [8]

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

   最后，杰文斯悖论意味着提高注意力机制的效率将推动更广泛的AI采用，而这最终将需要更多的GPU、HBM、DRAM和网络

   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🧵

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

   Kimi自身表示，最优的K3推理需要至少配备64颗芯片的大规模扩展域机架。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

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

   此外，由于权重占用超过1.5TB的HBM容量，K3的KDA和Gated MLA的KV缓存需要卸载到CPU DDR5和NVMe上，

   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🧵

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

   WideEP 优化不幸的缺点是它消耗大量的网络带宽。WideEP 对机架规模系统进行了高度优化

   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…

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

   WideEP 将 896 位专家分布在多个 GPU 上，使每个 GPU 的 HBM 只包含少量专家，优化了每 token 的内存使用和计算

   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

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

   其次，尽管 Kimi Delta Attention 的 KV 缓存传输网络需求最多可降低 10 倍，但其庞大的权重需要更高的网络带宽

   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…

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

   Kimi K3 对 NVIDIA 实际上相当有利，因为大型模型推理是 NVL72 的用武之地。由于 K3 拥有超过 2.8 万亿个参数，它需要

   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

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

   与DeepSeek R1的恐慌类似，一些无知的人认为Kimi K3使用线性注意力（KDA）对NVIDIA、HBM、DRAM和网络不利，因为

   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…