You guys have seen my work, and I've heard some people talk about some interesting subjects...
So what's everyone working on, how is the SDK working for your applications?
I'm working on my master thesis.
I have almost finished my application that calculate Dense Disparity Map from Color Stereo Images.
My algoritm is simple because I want only to prove quickness of the GPU
My GPU calculate it about 20 - 50 x faster than CPU.
Here is my output image: http://img399.imageshack.us/my.php?image=out000mj5.png
Neural networks now in debugging state and planing to look at another Computer Science thing, which I won't tell now.
Nice zimoch. Kos, I really think this technology would be great for servers to procedurally generate and control masses of bots, for any purpose really, I'm hoping to get into a lot of that when I have more time to code.
I working with a group of mechanical engineers on NASA's Orion project, the crew-launch replacement for the shuttle. They are doing the vibration analyses for the launch phase of the craft. In particular they are trying to dramatically speed-up a single run of model code, which must execute 100's of time to get a statistically significant sample size.
---jski
Ooooh! jski I was thinking about how cool it would be to bring the tech to a place like NASA. Wanna get me a job?
That's quite interesting nberger... Any kind of leading edge physics/science applications really grab my attention. And ryta, I'm sorry to hear your lack of success, keep in mind the new version of the SDK will be released sometime around the second week of december as far as we've heard. Should bring quite a few features that seem like it will bring the SDK closer to a more mature model.
As for me, I'm sort of lamenting the task, but it is time to design a model that will allow me to implement backends for any and all optimized or general GPGPU APIs. As well as begin to implement an interface for interacting with my new "noise modules" through a GUI. Once modularized and optimized it will be time for LOD.
There's a lot of interesting things that I've thought of in the style of SIMD. A friend of mine works for a security company. Every camera for instance could have a stream of pixels tested and outputed to a stream of results. There's just such a leap in efficiency, and we're seeing that across the board!
I am working on a High Energy X-ray Diffraction project at Iowa State University for the Air Force. The software is still in development stage.
Now, this is one nice thread 🙂
I would like to encode video/audio to 720p h264 or x264 High Profile. This is something I would like to try using Brook+. As x264 is open source I should be able to get that code but I have no idea as to how to proceed, or even if its possible. My card is Radion HD3850 (Sapphire).
I'm just trying to find documentation on AMD IL.
With some anticipation I downloaded and installed the v1.2 SDK. A quick examination shows that the cal framework itself is cleanly implemented and fairly usable. However, trying to understand the kernels themselves is almost impossible, let alone writing new ones.
For example, a kernel may start with "dcl_input-Interp(linear)". The IL spec (pg 6-11) listed the syntax as "dcl_input[_usage(usage)]", with NO information on what the usages are or their semantics! Apparently the documentation has been gutted. Googling around the web and this message board only shows that others had expressed similar frustrations, but no answers.
In theory, CAL was a great idea because one could have written some great algorithm targeting ATI chips. However, the level of documentation is calling into question why AMD is investing in the SDK at all. Is publishing some semantic information on an IL revealing too much of ATI's secret? That'd be laughable because people have been publishing architecture models along with assembly programming manual since the first chip was invented.
Or maybe IL isn't stable enough for external use? Unlikely because isn't that how the driver stack suppose to work? Whatever the case, please put out a real IL programming manual, or don't bother telling people to check out your SDK. It's really worse than not marketing it at all.
- a dev