You are viewing the course site for a past offering of this course. The current offering may be found here.
Lecture 23: How GPUs work (5)
sirejdua

Originally, general purpose parallel programming on gpus was done by hacking arbitrary code to be run as a shader to enable parallelization, but all of that has been put into CUDA now to make general purpose parallel programming on gpus a lot easier for the user(at least on nvidia gpus).

kingdish

How are the parts that are responsible for texture different from the shader core?

Michael-hsiu

What are the criteria for work distribution for a GPU? Are certain shaders, such as Blinn-Phong, allocated to less busy cores than shaders such as diffuse, b/c they're more computationally intensive?

Staffrishiu

@kingdish, from what I can understand online, the texture units are specialized to perform texture access and operations such as large matrix multiplications. They also have a personal cache for nearby texture accesses to speed them up.

kavimehta

How much faster is a gpu at computing graphics than a typical cpu?

zehric

@kavimehta I feel like your question can be interpreted in different ways. The speedup of a GPU comes from doing lots of arithmetic operations in parallel (kind of like having a bunch of CPUs all doing compute) not from any individual core being quicker.

Carpetfizz

When buying a GPU, the amount of "Video Memory" is an important factor to consider. When changing settings in PC games, the texture quality usually increases the amount of VRAM being used. Are the orange "Tex" blocks representative of the GPU's RAM?

henryzxu

I'm not confident, but the "Tex" blocks could be texture mapping units if it's just a representation of the processor itself. VRAM is definitely an important consideration when purchasing a GPU, however. On a related note, is there a reason why VRAM is GDDR5 versus the standard DDR4 for regular RAM?

You must be enrolled in the course to comment