From the course: NVIDIA Certified Associate AI Infrastructure and Operations (NCA-AIIO) Cert Prep

Unlock this course with a free trial

Join today to access over 25,300 courses taught by industry experts.

Quick comparison

Quick comparison - NVIDIA Tutorial

From the course: NVIDIA Certified Associate AI Infrastructure and Operations (NCA-AIIO) Cert Prep

Start my 1-month free trial Buy for my team

Quick comparison

“

Let me do a quick comparison so that you know clearly about use cases of each of these aspects. So let's talk about quick comparison here. We will focus on comparing GPU direct RDMA and GPU direct storage. Scope of GPU direct RDMA is across host. So GPU of remote can be accessed, so GPU can communicate to remote GPU NIC via RDMA that is a cross host. GPU direct storage is work within a host, GPU is accessing local or network storage. Main use cases, if you are looking for low latency GPU to GPU or GPU to NIC data transfer for HPC or AI cluster, you would use GPU direct RDMA. If you are looking for high throughput for your data loading from NVMe or RAID or parallel storage, GPU direct storage would be important, would be used, so be aware about this. Then data path, your GPU memory is accessed over RDM and NIC. It bypasses operating system, memory, and CPU. Here in GPU direct storage, storage device directly accesses GPU memory or GPU memory can directly access storage devices. Again…

Contents