Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Rasterisation patterns of the Haswell GPU (g-truc.net)
64 points by willvarfar on Sept 23, 2013 | hide | past | favorite | 13 comments


I don't understand the significance of this... Can somebody explain it to me like I'm 5 or provide a source for me to understand this more?


No expert here (heck, I've done maybe 5 minutes of OpenGL so far).

It reads like they use a special pixel shader to color each pixel of a texture in a more of less unique color based on when the pixel (or maybe block) was scheduled to be shaded (I'm guessing here based on the atomic counter bit). This yields a texture that shows the order in which pixel shaders are run on a texture. This seems to be different for different GPUs and architectures (unsurprisingly) and also seems to give insights into how certain things are implemented.

But that's all just a bit of guessing. Someone else might be able to give an actual explanation.


Yea well reading it closely, the red's value is what you see with the highest frequency of change. You can see that red increases then increments green, then starts over again at zero, so by looking closely at red, you can see where the computations stop and start over again, thus giving you the reds pattern. As to what it means, who knows.


Are these atomic counters related to the atomic object feature that was promised with OGL 3.0 and then dropped? Is there a template model now in OGL or is it using a different method to create immutable objects across multiple threads?

Please excuse my ignorance as I haven’t looked at OpenGL since 3.0 was released in 2006. Now SteamOS has piqued my interest in OGL again.

Edit: grammar. Thanks jonah! After some googling, it looks like atomic counters are not the same as the promised atomic objects, but seems OGL 4.0+ has nearly closed the feature gap with DX11.



From your second link:

2. To provoke; arouse: The portrait piqued her curiosity.


Oh, jeez. Not my intention at all. (Apologies if it was taken that way.)

So, are we not to help people (in general) come across better by using the proper words?


It seems I misread the original post and thought you were 'correcting' an already correct usage. On reading it again, it seems that I was wrong and I apologize.

Given that HN does not show any historical revisions, it can be hard to tell the order (or even existence) of edits. FWIW, if you had quoted the line you were commenting on it would have been unambiguously clear to me that the OP had taken your advice and corrected whatever he original wrote.


Sorry for the confusion. When I came back to edit my original post, I also changed peaked -> piqued per Johan's suggestion. Not sure of the Internet etiquette here, but next time I'll be more clear.


Thanks for the clarifications. :)

(It's hard to humbly offer suggestions online...)

> FWIW, if you had quoted the line you were commenting on it would have been unambiguously clear to me that the OP had taken your advice and corrected whatever he original wrote.

Good point. I usually do this, but will try to always do so as it helps with the clarity.


It is also interesting to see video of these rasterization patterns. The pattern is not exactly the same from frame to frame.

Nvidia: http://www.youtube.com/watch?v=WFx6StqpRdY

AMD: http://www.youtube.com/watch?v=WnUS8kzA3dI


That's cool, I think. Do you have any links to examples from other CPUs/GPUs?


(Poster, not author)

(There are Kepler and Southern Islands patterns on the same page for comparison)

I am trying to internalise this for CUDA and OpenCL kind of mechanical sympathy optimisation.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: