When analyzing the memory subsystem architecture of high-performance GPUs versus enterprise CPUs, which statement accurately characterizes their primary design differences regarding memory bandwidth and bus architecture?

The fundamental divergence between CPUs and GPUs is latency optimization versus throughput optimization. CPU memory systems are optimized for low latency. CPUs execute serial tasks where speed depends on how fast a single thread fetches data from cache or RAM, so they use sophisticated cache hierarchies and narrow memory interfaces, typically 64-bit to 128-bit per channel. GPU memory systems are designed for high throughput to sustain thousands of arithmetic cores operating in parallel. To feed these cores, GPUs require massive memory bandwidth, achieved through extremely wide memory interfaces. For example, GPUs using High Bandwidth Memory can have interfaces wider than 2048 bits and deliver more than 1-2 TB/s, versus typical enterprise CPU bandwidth of 200-300 GB/s. Therefore, the option saying GPUs implement significantly wider memory interfaces and deliver higher memory bandwidth, prioritizing massive parallel data throughput over low-latency serial access, is correct. The option claiming CPUs maintain higher raw memory throughput at all levels to compensate for lower core counts is wrong because CPUs prioritize low latency and use narrower interfaces, not higher raw throughput. The option saying both processors use identical bus widths, with differences only from clock speed, is wrong because their bus widths and memory architectures differ fundamentally. The option asserting GPUs sacrifice memory bus width for logic cores and therefore have lower memory bandwidth than CPUs is wrong because GPUs use much wider interfaces and achieve higher bandwidth.