Prohardver! Interview with Eric Demers

[sireric]

Fittingly, you could probably design your own offspring in Verilog but cannot operate a digital camera.

I probably have 3~5k pictures I took with my SLR on my HD, but I doubt there's a single one of myself! I hate self portraits! Just like hearing myself on the answering machine! I'll just keep on sending out that one, and shiver every time!

 

[Headstone]

Just put yourself on a webcam and let people randomly snap shots of you

 

[Xmas]

Between 0.5 and 1.0, off by 1 bit, though same total number of bits. Yes, it looses to the negative exponents near 0, but it's 4x more than 8888. And it's got the same profile as 8888. It's over 2x faster than FP16 (which usually doesn't compress as much), and can be even faster in some cases for filtering. That's why it's an interesting compromise. That's all.

I don't doubt these advantages, and I think 10:10:10:2 is an excellent format and it's a great thing ATI supports it. But I think of it as a replacement for 8888, FP16 is for different uses.

I don't see why FP16 would be worse for compression. In fact, if anything, FP16 should be more compressible (but for block compression, you probably have to live with the same compression rate).

Edit: Why "almost"? I'm guessing, though I have no proof, that the driver support for that could be a nightmare. I'm sure there's lots of exceptions to forcing that mode, which would make it hard to determine when it can and cannot use it. In the past, on things like that, that's been the case. We shall see.

The only case where I can see replacing X8R8G8B8 with X2R10G10B10 would fail is when the application is actually relying on the quantization for some weird reason.
Anyway, there's the easy way around this: hide it as a registry key and don't provide support for it

 

[Humus]

The only case where I can see replacing X8R8G8B8 with X2R10G10B10 would fail is when the application is actually relying on the quantization for some weird reason.

... or that they actually use the destination alpha. It's not common, but some apps do, and you can't really know for sure beforehand. I guess most games would just work, and Cat AI could cover the rest, but I'd imagine it would be a fair amount of work to test all games, current, past and future.

In any case I guess I should be a quick fix to add it to DXOverride though.

 

[Xmas]

... or that they actually use the destination alpha. It's not common, but some apps do, and you can't really know for sure beforehand. I guess most games would just work, and Cat AI could cover the rest, but I'd imagine it would be a fair amount of work to test all games, current, past and future.

That's why I wrote X8R8G8B8. Which is what everyone should use if they're not using destination alpha.

In any case I guess I should be a quick fix to add it to DXOverride though.

That would be great.

 

[Humus]

Alright, so I just added the format override to DXOverride.