GPU Computing at the Core
The primary benefit to GPU computing is that it is massively parallel and for the right kind of code it is like moving from a 4 cylinder engine to a jet. The performance benefits are massive. The kind of code that lends itself to this trends toward multi-media and analysis.
Strangely enough, it is actually more similar to how our brains work to interpret what we see and hear that makes it fascinating to see develop. Most existing code is written to perform tasks linearly however which better suites a CPU and much of this will never work well on a GPU due to the difference in architectures but where it works it works blazingly fast.
The Software Edge
They key problem that had to be overcome was that even if GPUs were capable of doing this work someone had to develop the tools to make it possible. NVIDIA's Cuda and OpenCL efforts were funded to solve this problem and it is highly unusual for a hardware firm to develop a significant software practice.
The Results
The primary benefit to GPU computing is that it is massively parallel and for the right kind of code it is like moving from a 4 cylinder engine to a jet. The performance benefits are massive. The kind of code that lends itself to this trends toward multi-media and analysis.
Strangely enough, it is actually more similar to how our brains work to interpret what we see and hear that makes it fascinating to see develop. Most existing code is written to perform tasks linearly however which better suites a CPU and much of this will never work well on a GPU due to the difference in architectures but where it works it works blazingly fast.
The Software Edge
They key problem that had to be overcome was that even if GPUs were capable of doing this work someone had to develop the tools to make it possible. NVIDIA's Cuda and OpenCL efforts were funded to solve this problem and it is highly unusual for a hardware firm to develop a significant software practice.
The Results
Some of the medical imaging work that Siemens showcased was amazing. They are now apparently, using NVIDIA's technology, able to model a beating heart to determine internal failures like blood leakage. They can also compile real time from ultra-scans the highly detailed image of a baby's face to determine of there are any birth defects early on or just show the parents what their child looks like.
Wrapping Up
This feels like Nvidia's year, perhaps their decade, even their problems with Intel seem to be getting resolved as this week Intel submitted a draft of their consent decree which should help limit what Intel can do against them and benefit both Nvidia and AMD. In the end, I think, we are seeing a solid movement in Nvidia's direction and this could just be their time to shine.
The end result is impressive we were shown advancements in imaging from a variety of CAD vendors that showcased how designs could be relatively easily used to create photorealistic images in a fraction of the time it used to take.
No comments:
Post a Comment