Blotchy Gradient problem again..need help - Page 2

Sorry I can't suggest any other ideas Nullboy. If the Linux drivers did not get rid of the issue that rules out software as the cause . Which, as Sherlock Holmes would say "that when you have eliminated the impossible, whatever remains, however improbable, must be the truth", so maybe it is a hardware issue after all.

There should be NO banding when looking at this image (for those that missed the earlier link) when viewed at 100% size on a true color display. I am NOT talking about blowing the image up. I have looked at it 100% (1 image pixel == 1 screen pixel) actual size on 2 different LCD monitors and the 8890 16" SXGA is definitlely banding the image. It is broken up into small 5x5ish squares. Extremely *sharp* and beautiful squares, but definitely squares.

So either there is a driver problem or the LCD screens are not really capable of showing 16million colors at once. It is really impossible to tell which it is, due to the fact that pretty much everything you see on the screen is drawn by the driver. I have a hard time believing that the LCDs are defective. That seems unlikely. Perhaps it is yet another of the many BIOS deficiencies that the 8890 has experienced? Who knows..

Ok eludwig let me show you something.

My setup:
8890 SXGA running XP Pro
Attached Samsung 17GLsi 17" CRT
Expand desktop enabled (so I can move windows between displays)
Using IE6

The procedure:
I open the link to the image you posted using IE6.
I adjust IE6 by its edges so it takes up all my LCD.
IE6 brings up the 16 x 16 grid and compresses it to fit into one screen. With me so far?
I leave the Expand/Compress button at the bottom right of the image alone so I am looking at IE6's compressed image of the 16x16 colour grid (ie. all 16 M colours are on the one page.

Now if you look at that image on the LCD can you see banding?
If you have a CRT have a look at the image on the CRT do you see banding on the CRT display?

aussie,

I am with you and in fact The reduced image shows no banding, but remember: you are not seeing all 16 million colors at once. The maximum number of simultaneous fdifferent colors that the SXGA LCD can display *at once* is 1,310,720. That is max for 1280 x 1024, period. No way around that. You can't show more colors than there are pixels, I'm afraid... The reduced image cuts the number of colors in each 256x256 square significantly.

We know the LCD is good, the issue is how many colors can it show at once and what is the minimum *step* between colors.

Good point. I was about to take you thru a second step which is to expand the image to its full size.

Ignoring the technical aspects for a second (I will explain later) when you look at the compressed one page image in IE6 do you see any banding?

Step two, hit the expand/compress button on the IE6 image so now we are looking at larger boxes (I get approx 5 across x 3.2 down). Now do you see any banding within each box?

If you have a CRT repeat the process there and tell me if you see colour banding.

I do get banding in each box at full size. Each 256x256 box is made up of approximately 50 or so vertical and horizontal bands (which, of course, form squares) each band is about 5 or so pixels wide & high. This effect is easiest to see in the upper left corner, where the red moves from bottom to top. I am not my laptop right now, unfortunately, so my recollection may not be perfect.

On my monitor, I do not and on my 1024x768 iMac LCD I do not.

I am not hugely technical regarding this type of technology, but it is my understanding that there is an issue with some LCDs and the maximimum number of *simultaneous* colors that can be displayed at once. So even tho an LCD can show all 16million colors, it cannot do so simultaneously. Therefore, if the maximum number of colors on the screen at once is more than the LCD can simultaneously show, there will be banding. Just a thought.

I understand what you mean by the difference between 16M possible and 16M simultaneous. Having gone thru the Sharp data sheet I can assure you it does 16M simultaneous. Even so that would need 16M pixels on one screen as you already pointed out so we only get about the 1.3M displayed.

Anyway to press on.
Each of the 256 boxes of the 16 x 16 grid contains 64K colours (256 x 256 grid in each box). So on our screen we will have approx 5 x 3.2 x 64K ~ 1M colours present.

If you now set your 8890 display colour depth to 16 bit (ie 5 bits of colour depth per pixel) each of the 5 x 3.2 boxes should show a definite grid of 32 x 32 smaller boxes. This equates to 262K colours (I see this on my Toshiba which has a 5 bit colour depth LCD).

So now you should be seeing 32 x 32 smaller colour boxes within each of the original 16 x 16 grid, agreed? If you do, we are at 5 bit colour depth (ie. 262K possible colours).

One other thing I also discovered is that setting the colour depth on one screen also alters the external screen (I was not sure of this before). So, if I now move the image across to my CRT (staying in 16 bit mode), I still get 32 x 32 smaller coloured boxes within each of the original 16 x 16 grid.

Now set the 8890 LCD back to 32 bit colour depth.

The obvious colour banding disappears on both my LCD and CRT displays, to be left with a smooth colour gradient. However, and this is important, if you look at the image in one spot for more than about 10 seconds (on either screen type) you will start to see bands of colour that appear to be wider/larger than the small boxes present under 16 bit mode. Do you get that effect? Is that what you are seeing as colour banding or are you seeing still smaller boxes (say 64 x 64 small boxes within each of the original 16 x 16 grid)?

If the colour banding you see in 32 bit mode is producing boxes *bigger* than the obvious small boxes displayed in 16 bit mode then that is an optical illusion (the reasons for which are complex and we will get to those later) not a LCD problem.

But if the boxes are smaller in 32 bit mode than 16 bit mode it could mean one of three things, either our LCDs are actually different models (8 bit vs 6 bit colour depths), or your laptop has a hardware fault or our software is different.

Let me know your results from the above tests. I am about to get some shut eye so we will pick this up later.

I will give this a whirl later when I get home.. am at work now (it is about 10:45 am here)

Ok, to continue.. I changed the screen to 16 bit and the bands became much wider.. fewer colors = bigger bands. Switching back to 32 bit left me with exactly what I had before, which is smaller, but very noticable bands. No change, I'm afraid. Just to keep the record straight, 16bit mode, which = r, g & b at 5 bits each, will wield 32,768 colors max. a 6 bit LCD, which would be r, g & b at 6 bits each will weild 262,144 colors max. From a pure eyeball standpoint, it looks like that may be the bit depth, for better or worse, of my 16" LCD.

As I mentioned earlier, I didn't really expect a true color LCD when I ordered the 8890. A 6 bit display is about what I expected. Why some of these displays are higher bit depth than others, I do not know. They certainly never promised a true color LCD in any literature I saw.

Hm... people seem to forget about adam's picture of the blotchyness...

http://pctorque.com/misc/8890review/drivers.jpg

Thanks eludwig. If you see smaller boxes then there is obviously something different between our systems. When I go 32 bit mode I get absolutel smooth colour gradients.

That leaves the other options - different LCD's, system problem or graphics driver.

James - I know about Adam's photo. What we are trying to do here is narrow down the effect. If you follow the same procedure I would be interested in your results too. It would appear that eludwig can see *smaller* boxes when going from 16->32 bit mode which I don't.

As to the 5bit vs 6bit stuff you are right. What confused me is that how do you stuff three 6 bit values into 16 bits or is that just a M$ way of describing a colour scheme when they say 16 bit it really should say 18 bit?

I'll see what I can do when I get my 8890 back... It recently died (as I'm sure you've already read about...) but when I do get it back, I have a good digital camera, and I'll take some pictures...

Background theory

Google is a wonderful thing.

At least now I understand 16 bit colour in PC land. Coming from a Sun background we have a much simpler world to work with.

Anyway - 16 bit colour - there are three, count them three different formats:

5650 = Red 5 bits, Green 6 bits, Blue 5 bits, Alpha 0 bits
5551 = Red 5 bits, Green 5 bits, Blue 5 bits, Alpha 1 bit
4444 = Red 4 bits, Green 4 bits, Blue 4 bits, Alpha 4 bits

For our purposes 5650 and 5551 formats are the most common for older PC's. Since few PC's supported transparency we can forget 5551 format. Thus we are left with older PC's in 16 bit mode being able to display 2^5 * 2^6 * 2^5 = 32 x 64 x 32 ~ 65K colours. This is what Windows calls 16 bit mode. My old Toshiba 2710 runs this mode.

Thus when you put the 8890 SXGA display into 16 bit mode you can display a max of 65K colours.

Now if we go to 24 bit colour we have one format:
6666 = Red 6 bits, Green 6 bits, Blue 6 bits, Alpha 6 bits

Thus you can display (ignoring alpha/transparency bits) 2^6 * 2^6 * 2^6 = 64 x 64 x 64 ~ 262K colours. Most current low end LCD's are spec'd to be 6 bit colour depth. Windows does not have this mode so it fudges it and calls it 32 bit mode. However, the hardware is only working with 6 bits of colour depth even though windows thinks it is 8 bit.

Now with 32 bit colour we have one format:
8888 = Red 8 bits, Green 8 bits, Red 8 bits, Alpha 8 bits. Windows of course calls this 32 bit colour. This is the spec for the iMac, the SXGA in the 8890, Samsung 191T and other top end LCD's.

(There were no 7 bit displays AFAIK - hardware guys just don't like odd bit counts. All the driver chips and RAMDAC's were either 4, 6 or 8 bit).

Note that for CRT's colour depth does not apply as a CRT gun is analog and it can display any colour level (within the limits of the phosphor and amplifiers of course). However the video cards that produce the signal for a CRT include what are called RAMDAC (Digital to Analog Converters) and they are limited by the accuracy (or number of bits) that the DAC can put out (normally RAMDAC's are 8 or 12 bits).

background theory pt2

Since we know that the PNG image contains 16M different colours and they are arranged in a 16 x 16 matrix we can deduce that the full image is 4096 x 4096 pixels, thus each of the smaller 16 x 16 squares must be 256 x 256 pixels.

From that we get that each smaller 16 x 16 square contains 256 x 256 = 65K colours.

On a system that only has 5 bit colour depth it has to decimate the low 3 bits of the 8 bits of colour information. Thus when viewing on of the smaller 16 x 16 squares a 5 bit display will have to do 256/(2^3) = 256/8 = 32 colour steps. You can check this on an 8890 SXGA screen by using a millimeter ruler. Each box is approximately 64 mm across. If you look at the screen in 16 bit mode and measure it, each little colour step is about 2 mm across which is what you would expect. Note this will not be exact because the Green signal is actually 6 bits so the display driver will compensate for the extra bit in its colour depth by being able to display more greens than reds and blues.

Now the 8890 SXGA display is rated at 16M colours so switching it back to 32 bit colour mode (8888 format or 8 bit colour depth) the image should go from 32 x 32 small 2mm boxes within the 16 x 16 grid to a smooth colour gradient as it is using 8 bits per red green and blue.

We have already said that 7 bit colour depth does not exist so we either have 6 bit or 8 bit colour depth to work with. In eludwig's case if he sees smaller boxes than those evident in 16 bit mode and I don't see the boxes it would imply that either due to hardware or software he is running at a 6 bit colour depth. Theoretically at a 6 bit colour depth there will be 64 x 64 small 1mm colour step boxes within each of the 16 x 16 grid.

If this is the case then his is a 6 bit system.

Now the big question: Why is his 6 bit and mine 8 bit?

Different LCD?
System hardware fault?
Software driver difference?

I'll see if I can get a photograph of the banding tonight with my digital camera. It is going to be a bit tricky, but I'll give it a shot. I will use the 16million image and see what I can get.

nullboy,

At full size (100%), on a true color display (which we are not even sure the 8890 has, IMO), you should see *NO*, I repeat *NO* banding of any kind when looking at the test image in 32 bit color. The big squares are made up of 256 x 256 boxes. Each pixel is a different color. No banding should be visible.

This can easily be proven by finding a CRT display to check it on. I have looked at it on like 5 true color displays (one was also an LCD) and the 8890 is the only one that shows banding in 32bit.

Now, whether or not we should or should not have a true color display is really a question for Sager. I was not promised a true color display when I bought the machine, so I feel very up in the air about the whole thing. I will attempt to take a picture of this effect to send to Sager.

Why hasn't anyone mentioned the BIOS fix?

Well, I have a pretty darn late BIOS. 889bvr03 was sent to me last night to attempt a fix of an unrelated video issue. This BIOS did not solve the banding "issue", if it even is an issue!