VOGONS


First post, by elianda

User metadata
Rank l33t
Rank
l33t

After the ongoing discussion here I thought I try to measure some actual values how the signal of a few old 2D cards look.

How I did it:
I exploited my capture card (Epiphan VGA2PCIe) to analyze a digital image the specific 2D card shows on its analog VGA output.
The test image I used is this: http://mail.lipsia.de/~enigma/neu/graphics_comp/test.bmp
It is a 24 bit BMP file that has in the upper area colored squares with red / green / blue / white and 50% gray.
Below are vertical white lines alternating with black lines. Then a black area and a black to white bar.
I used Win98SE and put the BMP as background image and set each card to 640x480 at 60 Hz.

First the sampling PLL of the capture card was aligned that it samples exactly per pixel and then the image was shifted vertically and horizontally that the captured 640x480 matches the 640x480 pixel output of the card.
The capture card also allows to set offset and gain for R,G and B (one setting for all). This is identical to brightness and contrast setting.
To get this correct I used Virtualdub as shown here: http://mail.lipsia.de/~enigma/neu/graphics_co … ase_correct.png
You can see at the bottom the histogram of the image. I tuned the offset such way that the black peak touches the left end of the histogram scale then tuned the gain that the whole range fills the range and the white peak touches the right end of the scale. This is the same as tuning brightness and contrast of a CRT to get the best representation.
Afterwards I tuned the sampling phase to hit exactly each pixel when the voltages are stable. A wrong sampling phase introduces brightness variations over the image and can be seen as a splitting of the white and black peak. As example if the stable white has some additional brightness variation it changes between some gray value and white, which results in a "near white gray peak" and a white peak in the histogram.
This can be seen here for wrong sampling phase: http://mail.lipsia.de/~enigma/neu/graphics_co … e_incorrect.png
There a broad peak near black and a broad peak near white are shifted inwards. (compare to the first image)
The sampling phase was tuned such that the peaks are sharp and do not split.

This was done for each card then an image was captured. I used Igor Pro from Wavemetrics to put horizontal Profile Lines over each area with the (vertical) width of 5 pixels.
Interesting to compare is:
How each component scales to white on the grayscale.
Which value the RGB values reach for the white square in comparison to the while vertical lines.
How stable black and white is, and how stable it is in the area of the vertical white lines.
Please note than lower gain values mean a higher contrast and a lower offset means a lower voltage is black.

RESULTS

Matrox Millenium
http://mail.lipsia.de/~enigma/neu/pics/matrox … y_expansion.jpg

Used values: Offset 31, Gain 191, Phase 15

Image captured:
matrox_millenium.png

matrox_grayscale_profile.png

matrox_black_profile.png

matrox_lines_profile.png

matrox_colors_profile.png

SPEA V7 Mirage P64 (S3 Vision 864)
http://mail.lipsia.de/~enigma/neu/pics/graphi … mirage_p-64.jpg

Used values: Offset 31, Gain 172, Phase 31

Image captured:
spea_v7_mirage_p64.png

spea_grayscale_profile.png

spea_black_profile.png

spea_lines_profile.png

spea_colors_profile.png


Biostar S3 Virge

http://mail.lipsia.de/~enigma/neu/pics/s3_virge_dx.jpg

Used values: Offset 31, Gain 172, Phase 0

Image captured:
biostar_s3_virge.png

biostar_grayscale_profile.png

biostar_black_profile.png

biostar_lines_profile.png

biostar_colors_profile.png

Diamond FireGL 1000 (Permedia 2)
http://mail.lipsia.de/~enigma/neu/pics/diamon … _1k_8mb_pci.jpg

Used values: Offset 36, Gain 191, Phase 31
I didn't hit the white point precisely with this one it's a bit too dark.
Still I didn't wanted to drop these data.

Image captured:
diamond_firegl1000.png

diamond_grayscale_profile.png

diamond_black_profile.png

diamond_lines_profile.png

diamond_colors_profile.png

Data: http://mail.lipsia.de/~enigma/neu/graphics_co … /comparison.pxp
You can load this and play with the graphs using Igor Pro (it's 30 days fully functional) from http://www.wavemetrics.com/support/demos.htm

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 1 of 15, by Oerg866

User metadata
Rank Member
Rank
Member

I see my devious plan to get some fame by infiltrating your screenshots has worked 😜

Great work, as usual.

Reply 2 of 15, by obobskivich

User metadata
Rank l33t
Rank
l33t

Very interesting; out of curiosity is there some sort of "standard" or "benchmark" that these should be (or could be) compared to? Like what is an example of "perfect"?

Reply 3 of 15, by elianda

User metadata
Rank l33t
Rank
l33t

Perfect would be if you take the test.bmp image and do the profile lines there. However the result is obvious.
For the analog signals there is a range where you can tune, f.e. I can increase contrast such way, that white and black is capped, so the fully saturated colors give a straight line at 255.

But this would not be very helpful for such comparison. If I tune the output such way that black is close to 0 and white is nearly 255 then I can see f.e. the stability of the signal. I also have a reference compared to the values from the vertical white lines, where the signal does not completely reaches the values of the filled squares. I also noticed that using lines between the points for the graph from the colored squares is better since you can see the step, which corresponds to the sharpness of the color transition.

For a real capture I would tune it of course a bit different, such that black and white caps slightly, which also removes the visible noise for no and full saturated colors. But this would be very close to the images presented here.

I also did some captures at 1024x768, I am just putting together the results.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 4 of 15, by elianda

User metadata
Rank l33t
Rank
l33t

Here are additional results for 1024x768 resolution at 60, 75 and 85 Hz.
I captured from the Matrox Millenium and the S3 Virge. The S3 Virge allows those refresh rates only at 16 bit color depth.
The procedure was the same as described in the first post. I used a test image with the same structures.
The profile lines are averaged over a width of 5 pixels.

Due to the size of the images I will give the links only:

Captured images.
Matrox Millenium 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x60.png
Matrox Millenium 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x75.png
Matrox Millenium 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x85.png
S3 Virge 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x60.png
S3 Virge 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x75.png
S3 Virge 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x85.png

Profile over color squares:

Matrox Millenium 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x60_colors.png
Matrox Millenium 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x75_colors.png
Matrox Millenium 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x85_colors.png
S3 Virge 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x60_colors.png
S3 Virge 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x75_colors.png
S3 Virge 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x85_colors.png

Profile over vertical one pixel thin white lines:
Matrox Millenium 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_lines.png
Matrox Millenium 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x75_lines.png
Matrox Millenium 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x85_lines.png
S3 Virge 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_lines.png
S3 Virge 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x75_lines.png
S3 Virge 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x85_lines.png

Profile over black area:
Matrox Millenium 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_black.png
Matrox Millenium 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x75_black.png
Matrox Millenium 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x85_black.png
S3 Virge 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_black.png
S3 Virge 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x75_black.png
S3 Virge 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x85_black.png

Profile over gradient, S3 Virge is 16 bit mode:

Matrox Millenium 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 60_gradient.png
Matrox Millenium 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 75_gradient.png
Matrox Millenium 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 85_gradient.png
S3 Virge 60 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 60_gradient.png
S3 Virge 75 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 75_gradient.png
S3 Virge 85 Hz: http://mail.lipsia.de/~enigma/neu/graphics_co … 85_gradient.png

The tuned setting were as follows:
Matrox Millenium 1024x768
60 Hz: Offset 31, Gain 183, Sampling Phase 18
75 Hz: Offset 31, Gain 182, Sampling Phase 30
85 Hz: Offset 31, Gain 184, Sampling Phase 27

S3 Virge 1024x768 (16 bit)
60 Hz: Offset 31, Gain 163, Sampling Phase 31
75 Hz: Offset 31, Gain 162, Sampling Phase 0
85 Hz: Offset 31, Gain 161, Sampling Phase 18

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 6 of 15, by elianda

User metadata
Rank l33t
Rank
l33t

Well that the Matrox has a better quality in 1024 was expected, but in 640x480 there is no such difference visible.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 7 of 15, by retrofanatic

User metadata
Rank Oldbie
Rank
Oldbie

Very nice work indeed elianda. Thanks for this.

Reply 8 of 15, by leileilol

User metadata
Rank l33t++
Rank
l33t++

I feel a Geforce256/2 should be part of this list...

apsosig.png
long live PCem

Reply 9 of 15, by smeezekitty

User metadata
Rank Oldbie
Rank
Oldbie

Here are additional results for 1024x768 resolution at 60, 75 and 85 Hz.
I captured from the Matrox Millenium and the S3 Virge. The S3 Virge allows those refresh rates only at 16 bit color depth.

That's funny. I can set 1024x768 24 bit 75hz on mine.

Reply 10 of 15, by elianda

User metadata
Rank l33t
Rank
l33t

That's funny. I can set 1024x768 24 bit 75hz on mine.

Are you sure it is no DX/VX/GX/GX2 ?
While the picture name suggests a DX the card is actually a normal Virge with the 86C325 chip.

I feel a Geforce256/2 should be part of this list...

I feel that there is a ELSA Gloria III in my K6-2+ system:
http://mail.lipsia.de/%7Eenigma/neu/pics/elsa … 2_pro_front.jpg

60 Hz: Offset 31, Gain 181, Phase 3
75 Hz: Offset 31, Gain 181, Phase 3
85 Hz: Offset 31, Gain 183, Phase 31 <- 31 is the limit of the tunable scale

Captured images.
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x60.png
75: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x75.png
85: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x85.png

Profile over color squares:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x60_colors.png
75: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x75_colors.png
85: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x85_colors.png

Profile over vertical one pixel thin white lines:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_lines.png
75: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x75_lines.png
85: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x85_lines.png

Profile over black area:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_black.png
75: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x75_black.png
85: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x85_black.png

Profile over gradient:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 60_gradient.png
75: http://mail.lipsia.de/~enigma/neu/graphics_co … 75_gradient.png
85: http://mail.lipsia.de/~enigma/neu/graphics_co … 85_gradient.png

Some additional notes:
Of course the capture card itself is not perfect and 85 fps is the maximum frame rate it is able to capture. This may rise the problem that I can not set the fully correct sample phase at 85 Hz modes. (The correct value is beyond the scale) So the 85 Hz image quality may be even a bit better on a real screen.
Some more explanation on this: The incorrect phase histogram image in the first posting shows a broad peak at the left side and a similar broad peak at the right side. After all my testing those are primarily from the brightness of the thin vertical white lines. Due to the alternating lines do no reach the same white level as the filled square, same applies for the black lines. So if I shift the sample phase offset I can hit the position just between the pixels where the white level rises. As a result the alternating lines appear as a 50% gray square. This is the worst case.
If I start from this I can tune the phase and move those peaks towards black and white respectively. At a certain value they do not move further and if I change the phase more they move in again towards 50% gray. This point is the optimum I can hit as it represents the highest and lowest brightness for the white/black lines.
About the width of those peaks: The width represents the signal stability. The capture cards starts at the left of each raster line relative to the sync signal. If the signal is has no rectangle shape but looks more like a triangle I can not hit the plateau. In reality the signal is never a perfect rectangle and the black and white plateaus are not flat. If the signal also jitters in time slightly to the left and right relative to sync then it gets smeared out. So the sampled brightness is not well defined resulting in a broad peak.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 11 of 15, by smeezekitty

User metadata
Rank Oldbie
Rank
Oldbie

Are you sure it is no DX/VX/GX/GX2 ?
While the picture name suggests a DX the card is actually a normal Virge with the 86C325 chip.

I don't think it is. http://postimg.org/image/yiks5nh3x/

Reply 12 of 15, by LunarG

User metadata
Rank Oldbie
Rank
Oldbie
elianda wrote:

Well that the Matrox has a better quality in 1024 was expected, but in 640x480 there is no such difference visible.

I think it was generally agreed back in the days, that most graphics cards delivered satisfactory image quality at low resolutions (i.e. 640x480 and possibly 800x600), but that it was when you started turning the resolution up that you could really tell the different cards apart. Go one step higher, to 1280x960 (or 1280x1024) and you'll probably find that things go even further in favor of Matrox over S3. Yes, there were some higher quality S3 cards as well, that used better RAMDACs and higher quality components, but the majority of S3 cards were budget offerings, that simply couldn't compete with the slightly more professional offerings from Matrox, at high resolutions. If you're after a graphics card only to run 640x480 and below, then an S3 based card can be just as good, with perhaps better games compatability, but if you also wanna be able to run Windows on the system with nice highish resolution and high colour depths, then personally I'd go Matrox.

WinXP : PIII 1.4GHz, 512MB RAM, 73GB SCSI HDD, Matrox Parhelia, SB Audigy 2.
Win98se : K6-3+ 500MHz, 256MB RAM, 80GB HDD, Matrox Millennium G400 MAX, Voodoo 2, SW1000XG.
DOS6.22 : Intel DX4, 64MB RAM, 1.6GB HDD, Diamond Stealth64 DRAM, GUS 1MB, SB16.

Reply 13 of 15, by obobskivich

User metadata
Rank l33t
Rank
l33t

Any chance you could test this with a Voodoo or Voodoo2 "in the chain"?

Also +1 on very nice work; lots of good info!

Reply 14 of 15, by elianda

User metadata
Rank l33t
Rank
l33t
obobskivich wrote:

Any chance you could test this with a Voodoo or Voodoo2 "in the chain"?

Also +1 on very nice work; lots of good info!

Voodoo2 (Gainward Dragon) after the Quadro2 Pro from the previous post. I set 60 Hz which most users with a TFT use.

60 Hz: Offset 31, Gain 179, Phase 31

Captured image.
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 1024x768x60.png

Profile over color squares:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 8x60_colors.png

Profile over vertical one pixel thin white lines:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_lines.png

Profile over black area:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 68x60_black.png

Profile over gradient:
60: http://mail.lipsia.de/~enigma/neu/graphics_co … 60_gradient.png

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 15 of 15, by elianda

User metadata
Rank l33t
Rank
l33t

I made a small program that does the alignment and offset/gain/phase setting automatic. (per color channel)
This results in better quality as on the gradient the components now really are on a single line. (no parallel shift or different gains per channel)
If someone has a epiphan capture card and is interested in such a tool, drop me a pm.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool