Proposed Patches Would Enable FP16 Pixel Format Support For Older AMD GPUs

[phoronix]

Phoronix: Proposed Patches Would Enable FP16 Pixel Format Support For Older AMD GPUs

Patches posted on Monday by independent open-source contributor Mario Kleiner would enable FP16 pixel format support for older generations of AMD Radeon GPUs...

http://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-FP16-DCE8-Patches

[pegasus]

DCE? What is this?
It would be oh so helpful to link to some translation table in this kind of news ... You know, one that shows the relation between internal engineering name, marketing family name and product on the market name ...

[Michael]

DCE? What is this?
It would be oh so helpful to link to some translation table in this kind of news ... You know, one that shows the relation between internal engineering name, marketing family name and product on the market name ...

Display controller engine... It's mentioned in the very article what hardware it correlates to.

[microcode]

Also seems like mixing FP16 noise into the buffer could be a basic level of dithering, better than doing it in integer.

[pegasus]

I do remember that Kaveri was an APU and that R9-290 had a Hawaii chip ... But Polaris is a quite large family and I have to ctrl+f through https://en.wikipedia.org/wiki/List_o...ocessing_units to translate these names to actual products on the market.

Suggestion ... it would be few minutes of work to populate some db tables with these names and add some code so that on mouseover on these names we would get a list of the products affected by these proposed changes. What do you think?

[StandaSK]

DCE? What is this?
It would be oh so helpful to link to some translation table in this kind of news ... You know, one that shows the relation between internal engineering name, marketing family name and product on the market name ...

https://www.x.org/wiki/RadeonFeature/ can be quite useful regarding that.

[ezst036]

Will this increase performance in a handful of cases?

[baryluk]

I hope also DCE6.1 will get the support. I believe TAHITI (i.e. 7970 HD, aka GFX6) and other Southern Islands chips, did support 10bpp and 12bpp, even over HDMI 1.4, on Windows. I still use that card frequently.

[agd5f]

I hope also DCE6.1 will get the support. I believe TAHITI (i.e. 7970 HD, aka GFX6) and other Southern Islands chips, did support 10bpp and 12bpp, even over HDMI 1.4, on Windows. I still use that card frequently.

I don't think DCE6 supported floating point display formats. It already supports deep color (e.g., 10 bpp) if you enable DC.

[baryluk]

I don't think DCE6 supported floating point display formats. It already supports deep color (e.g., 10 bpp) if you enable DC.

You are right. I think it doesn't support fp16 output or fp16 to 10/12-bit conversion in hardware.

So all is good.