VISUAL COMPUTING SYSTEMS

This page contains lecture slides and recommended readings for the Fall 2016 offering of 15-769.

Required Reading:

The required reading for this class is not an academic technical paper, but a whitepaper from Intel describing the architectural geometry of their latest GPU. This processor is particularly notable because it is the integrated GPU that will be in most mid-2016 and later Core i5 or i7 processors -- the marketing name is HD Graphics 530 (or larger).

I'd like you to read the whitepaper, focusing on the description of the processor in Sections 5.3-5.5. Then, given your knowledge of the concepts discussed in lecture (such as superscalar, multi-core, multi-threading, etc), I'd like you to describe the features of the processor (using terms from the lecture, not Intel terms).

Pro tip: Consider your favorite data-parallel language, such as GLSL/HLSL shading languages, CUDA, OpenCL, ISPC, or just an OpenMP #pragma parallel for, and make sure you can think through how an embarrassingly parallel for loop can be lowered to these architectures. (You don't need to write this down, but you could.)

Students wanting to go farther might also be interested in also reading the NVIDIA P100 or GTX 980 whitepapers also linked below. Then you could make a table contrasting the geometry of: a modern AVX-capable Intel CPU, Intel Integrated Graphics (Gen9), NVIDIA GPUs, and any other processor you might be interested in, such as Intel Knights Corner, AMD GPUs, etc.

Further Reading:
There is no required reading for this lecture.
Further Reading:
  • The Stanford CS448A course notes are a very good reference for camera image processing pipeline algorithms and issues.
  • The interactive demos on the Stanford CS178 course site are very well done (some were shown in class)
  • Clarkvision.com has some very interesting material on cameras.
Required Reading:
Further Reading:
There is no required reading for this lecture.
Further Reading:
Required Reading:
Further Reading:
Required Reading:
Now that we've covered the basics of what a DNN is, and discussed the basics of forward evaluation and training workloads, we'll take a look at more recent DNN topologies that are designed to be evaluated/trained more efficiently and/or yield higher accuracy.
Further Reading:
There is no required reading for this lecture.
Further Reading:
There is no required reading for this lecture.
Further Reading:
Lecture 13: Hardware Accelerators for Deep Neural Network Evaluation
There is no required reading for this lecture.
Further Reading:
There is no required reading for this lecture.
Further Reading:
There is no required reading for this lecture.
Further Reading:
Further Reading on Texture Compression:
Required Reading:
Further Reading:
Required Reading:
Further Reading:
There is no required reading for this lecture.
Further Reading:
Required Reading:
Further Reading:
Required Reading:
Further Reading:
Lecture 25: TensorFlow, XLA, and Emerging IRs for Machine Learning
There are no lecture slides from this lecture. The students led an in-class discussion about the papers/links below..
Required Reading:
Further Reading: