NCA-AIIO: AI Infrastructure and OperationsDemo

By NVIDIA · Question Mode

QuestionQ100

Data Center Infrastructure

A simulation is constrained by memory-transfer speeds.

Which GPU architectural feature addresses this limitation?

  • A Large shared memory and high-bandwidth buses.
  • B Increase number of IO ports for PCIe devices.
  • C Dedicated and proprietary inference ASICs.
  • D Direct wiring of GPUs as main disk controllers.
Explanation

Large shared memory reduces reliance on slower memory accesses for data reused by threads, while high-bandwidth buses increase the rate at which data can be transferred. Together, these features address memory-transfer bottlenecks in GPU workloads.

Community Discussion

No comments yet. Be the first to start the discussion!