在最新一期Dwarkesh Podcast中,MatX CEO Reiner Pope 为听众带来了一堂从零开始的“芯片设计黑板课”。他从最基础的逻辑门讲起,逐步构建出乘加运算单元、脉动阵列,并解释了时钟周期、流水线寄存器等关键概念。通过对比GPU、TPU和FPGA的架构设计,他揭示了芯片设计中“计算与通信权衡”这一核心主题,为理解现代AI芯片的工作原理提供了清晰的路径。
"This is the single reason low-precision arithmetic has worked so well for neural nets." ——这是低精度算术在神经网络上如此有效的唯一原因。
"Almost all your cost becomes synchronization or communication cost compared to the actual logic." ——与实际逻辑相比,你几乎所有的成本都变成了同步或通信成本。
"A GPU is just a bunch of tiny TPUs." ——GPU本质上就是一堆微型TPU。
"Make this one bigger somehow while keeping this at the same size. That's the goal." ——想办法把这个(计算单元)做大,同时保持那个(数据移动单元)不变。这就是目标。
"It hurts your throughput, in fact, because the throughput of your chip is the product of how much you get done per clock cycle... times how many clocks you get per second." ——它实际上损害了你的吞吐量,因为芯片的吞吐量是每个时钟周期完成的工作量乘以每秒的时钟数。