Welcome, Guest. Please login or register.

Author Topic: Scandoubler recommendations  (Read 14740 times)

Description:

0 Members and 1 Guest are viewing this topic.

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« on: September 06, 2004, 11:44:38 PM »
@Patrik

So it's really only any use with pre AGA screenmodes? :lol:

If you (for whatever insane reason) absolutely had to use a 4/4/8 format, surely you'd choose 8 bits for green?

RGB 8:4:4? What a pointless design!
int p; // A
 

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« Reply #1 on: September 14, 2004, 10:37:27 PM »
Well, any pre aga games will look just fine with 4-bits per gun anyway.

I still think RGB 844 was completely the lamest choice imagimable. They really should have used some sort of proper 16 bit 5:6:5, or if such quantization was not possible then at least give 8 bits to green.
int p; // A
 

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« Reply #2 on: November 18, 2004, 11:59:25 PM »
Quote

keropi wrote:
I never saw this 16bit dce thread... I got a DCE flickermagic about a week ago, and it looks fine... very clear, haven't noticed the colors though... but it still beats the native flicker screenmodes


To see the effect, create 3 256 colour images in deluxe paint AGA, one showing a gradient from black to maximum red (ie a full palette spread), one showing the gradient from black to maximum green and one showing the gradient from black to maximum blue. What you should observe is that the green and blue suffer quantization gradients where the 8-bit resolution is axed down to 4 bits. Only the red will show a smooth, unbroken gradient.
int p; // A
 

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« Reply #3 on: November 19, 2004, 01:16:50 AM »
Well 8/4/4 would imply red has 16x the resolution of either of the other two. Each bit per gun doubles the range of values it has.

I suggest you try the trick using a vertical gradient on a 256 pixel tall screen if possible -  a horizontal gradient will always have some bleed in it. Turn off dithering also. If you have a genuine 8-bit gun colour, a 256 tall gradient will have a unique colour every line giving a smooth gradient. If you have 4-bit, it will be approximately 16 bands you will see.

If you see that green has 64 bands and that red and blue each have 32 it suggest an RGB 565 arrangement which is far more sensible (its your basic 16-bit RGB format).

This gives you 32x64z32 = 65536 colours and is not too bad.  A well dithered RGB 565 display is almost as good as your HAM8 (since green is the brightest colour, it makes sense to give it the greater resolution so that the steps are smaller).

If you see red as completely smooth, green and blue as 16 bands, it implies the rather insane RGB 844 format. This format could only have been chosen by some tired HW designer who had no knowledge of colour theory at all. First of all, the green channel should have the greater bit depth (where the total resolution is not divisible by three) and no single R/G/B channel really needs more than 2x the resolution of the others. This is why RGB 565 works so well, but RGB 844 is just bobbins ;-)
int p; // A
 

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« Reply #4 on: November 19, 2004, 02:12:39 AM »
Hmm, this got me thinking

Perhaps we could get a colour calibration testcard. A basic 320x256 HAM8 one consisting of your three basic (undithered) gradients (arranged side by side as 100x256, 100x256, 100x256) a test image for scandoublers as there would be no fringing for vertical gradients.

For perfect HAM8 reproduction, there should be 64 bands to each gradient (each one has a 6-bit basic resolution). Perhaps some calibration points down the edge of the image to help counting :-)
int p; // A
 

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« Reply #5 on: November 19, 2004, 02:35:14 PM »
@Hyperspeed

Actually, believe it or not, there are tricks that can be achieved with the copper that allow static images to be shown in near 24-bit on AGA/HAM8 ;-) It centres around using a different set of base colours per scanline. It's good for image viewers but not suited for realtime displays.

If you have RGB565 that is not too bad. Any 844 variant would just look awful.

Let's see how they compare ;-)
int p; // A
 

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« Reply #6 on: November 20, 2004, 01:20:38 AM »
@Hyperspeed

You could look into getting a graphics card. Your eyes will love you for it. My BVision is faster in 1600 x 1200 x 65536 colours than AGA is in 640 x 256 x 4 colours.
int p; // A
 

Offline Karlos

  • Sockologist
  • Global Moderator
  • Hero Member
  • *****
  • Join Date: Nov 2002
  • Posts: 16879
  • Country: gb
  • Thanked: 5 times
    • Show all replies
Re: Scandoubler recommendations
« Reply #7 on: November 20, 2004, 01:39:25 AM »
That's true. The P-IV is a very good card, once termed "the godfather of amiga graphics cards" :-)

You'd be very ill advised to use a BVision in a desktop 1200 though - the Permedia2 and SGRAM are clocked at least 83MHz and get hot. In my tower, the P2 has a 25mm tall power amplifier heatsink bonded to it and a 80mm case fan mounted over the whole card area to keep it cool.
int p; // A