Previous | Next --- Slide 26 of 54
Back to Lecture Thumbnails
kayvonf

Question: Why do you think CUDA uses the term "device code" rather than "GPU code"?

aznshodan

I think they arbitrarily decided to call CPU as host and GPU as device. If they were to use the term "GPU code", they would have referred host as "CPU code." Also the word device insinuates that it is physically separated from the CPU.

fgomezfr

It's also a bit of a marketing ploy; calling it "GPU" (graphics processing unit) code doesn't exactly help sell the idea of a general-purpose parallel processor. Referring to it as the 'device' attached to a 'host' is less specific, but as @aznshodan noted it maintains the idea of physical separation, which is critical to understand CUDA's memory model.

solovoy

Maybe one day, CUDA will run on some new NVIADA multi-core devices which are not called GPU any more.

abist

Adding to @solovoy's comment, I think they want it to be a programming interface, and GPUs just happen to be one of the more possible devices that follow this interface.

sanchuah

I think the reason maybe that NVDIA want to separate CUDA abstraction and implementation. CUDA code maybe not necessary just run with GPU in the future. NVIDIA can invent new fancy hardware to implement CUDA code.