1

From a history of graphics hardware:

Indeed, in the most recent hardware era, hardware makers have added features to GPUs that have somewhat... dubious uses in the field of graphics, but substantial uses in GPGPU tasks.

What is the author referring to here?

Justin Meiners
  • 10,754
  • 6
  • 50
  • 92
spraff
  • 32,570
  • 22
  • 121
  • 229
  • 1
    Let's wait for @nicolbolas to answer that one. :) – Bart Aug 27 '13 at 17:40
  • Simple 2D graphics might not need all the fancy features used by e.g. [OpenCL](http://www.khronos.org/opencl/) on GPGPU. – Basile Starynkevitch Aug 27 '13 at 17:41
  • I think I misunderstood the question. I was assuming the author was talking about password cracking, and linked to [this answer](http://security.stackexchange.com/questions/32816/why-are-gpus-so-good-at-cracking-passwords), but after re-reading, I am not confident that that is the meaning the author was intending. – Gray Aug 27 '13 at 17:47
  • @Bart He presumably wrote it so I would guess he has a more accurate answer than me :) – Justin Meiners Aug 27 '13 at 18:26

1 Answers1

5

I would assume that it is referring to the extra hardware features, as well as abstraction to support GPGPU initiatives such as CUDA and OpenCL. From the description of CUDA:

CUDA has several advantages over traditional general-purpose computation on GPUs (GPGPU) using graphics APIs: Scattered reads – code can read from arbitrary addresses in memory Shared memory – CUDA exposes a fast shared memory region (up to 48KB per Multi-Processor) that can be shared amongst threads. This can be used as a user-managed cache, enabling higher bandwidth than is possible using texture lookups. Faster downloads and readbacks to and from the GPU Full support for integer and bitwise operations, including integer texture lookups

These are all features that are relevant when implement for CUDA and OpenCL, but are somewhat irrelevant (at least directly) to graphics APIs such as OpenGL. GPGPU features still can be leveraged in unconventional ways to supplement the traditional graphics pipeline.

The example of "CUDA exposes a fast shared memory region" would be an additional hardware requirement potentially useless to OpenGL.

You can read this detailed document describing the architecture required for CUDA, and the differences between it and traditional graphics exclusive GPUs.

Justin Meiners
  • 10,754
  • 6
  • 50
  • 92