VOGONS


Matrox P690

Topic actions

First post, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

In this thread, I'll explore the Matrox P690, the least appreciated member of the Parhelia family.

Is it an interesting card? I think so. One of the few DX8 cards for PCIe, of which there are surprisingly few, and possibly the best 16-bit dithering available for gaming, though the algorithm is seemingly shared with the rest of the Parhelia cards and at least as far back as the Matrox G550. Less expensive than the rest of the Parhelias, and possibly the most energy efficient. But like the rest of the Parhelias, Windows NT only.

By default, the card has issues in games due to its office-oriented software. No vsync control, no 3D image quality settings exposed in the Matrox tools (which also means no fragment antialiasing unlike on the P650), and depth-related glitches in DX7/8 games.

But thanks to AI, issues are now much easier to address if there's any way to address them. I already know that the DX7/8 glitches appear to be due to a driver bug that can be patched, that the card does appear to support fragment antialiasing even though it's not advertised, and that there's native vsync control. Maybe overclocking would be possible too if the stars align, the DDR2 chips may easily be rated for higher clocks than what they're running at.

Reply 1 of 4, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

The P690 needs 2.x drivers, but 2.x drivers drop support for PowerDesk's "Game and 3D Settings" dialog, so no way to enable antialiasing, change texture filtering mode or control vsync, unlike in the 1.x drivers.

There appear to be two ways to control the hidden settings in 2.x: by putting certain magic values into certain keys in the registry, or by patching PowerDesk's settings dialog back into 2.x. Based on a 1.x driver, Claude Code with Opus 5.5 put the dialog back into 2.x, which involved patching the PowerDesk executable to allow the dialog to be opened (hidden by default) and to populate some data that had been taken out, and writing a missing DLL that passes settings from PowerDesk to the driver. Works.

The attachment game-and-3d-settings.png is no longer available

A bare dialog but that's the style of this version of PowerDesk. Even though 2.x PowerDesk doesn't expose this info as-is, the driver reports all features except stereo OpenGL to be available on the P690. I've only tested antialiasing, texture filtering and vsync, but presumably the others work.

Thief 2 with no AA, FAA 16x, and SSAA 4x on the P690:

The attachment thief-2-noaa-faa16-ssaa4.jpg is no longer available

At some point the quality of AA on the P690 needs to be compared with earlier Parhelias. Maybe there's regressions, maybe there's improvements.

Reply 2 of 4, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

Table fog. I used driver 2.05 for these tests.

Out of box, no fog in Thief 2 on the P690. Claude Code had a look into it, the game has a bug that breaks standards compliant drivers that don't have good quirks handling: asking for W fog without providing a projection matrix. By spec default projection values get used and the fog breaks.

But the P690 does appear to support table fog in hardware anyway, and the driver looks to have two paths for it: one that's more standards compliant and one that's more legacy compliant. But Thief 2's non-compliant fog rendering doesn't trigger the legacy path in these drivers. Claude made a driver patch that detects the type of fog data Thief 2 passes in and routes it onto the legacy path. Result: fog works.

The attachment thief2-fog.jpg is no longer available

Star Wars: Shadows of the Empire. No fog with the unpatched driver, fog with the patched driver. So the driver patch is valid in this game also, and the legacy support it provides is broader.

The attachment shadows-fog.jpg is no longer available

As a test, I had Claude modify Monster Truck Madness 2 to use table fog rather than its normal vertex fog. It's a similar game to Thief 2 technically in that it transforms its vertices on the CPU and doesn't set a projection matrix. But it uses DX5 which doesn't support W fog and so shouldn't need a projection matrix. Result: no fog with the unpatched driver, fog with the patched driver. Then I added a call to SetTransform to set the projection matrix, and now the fog worked with the unpatched driver.

The attachment mtm2-fog.jpg is no longer available

So the P690 appears to support table fog in hardware, its drivers provide legacy game support but have less than ideal routing for it , and the drivers can be patched to trigger the legacy path more reliably.

Reply 3 of 4, by Joseph_Joestar

User metadata
Rank l33t++
Rank
l33t++
vvbee wrote on Yesterday, 04:58:

Star Wars: Shadows of the Empire. No fog with the unpatched driver, fog with the patched driver. So the driver patch is valid in this game also, and the legacy support it provides is broader.

Interesting. So by default, using official drivers, this legacy table fog path doesn't activate on the Parhelia, at least in some games. Even though the hardware supports it. If you do end up releasing a patched driver, I would suggest making this a toggleable setting. Like a checkbox for "legacy table fog compatibility" or something like that. Just in case some newer games don't like it.

I'm wondering if a similar patch would be possible for ATi's Catalyst 6.2 driver, thus allowing proper table fog rendering on Radeon cards under Win9x. The hardware on R300 (and newer) Radeons seems capable of it, since those cards can render fully accurate table fog when using Catalyst 7.11 under WinXP. Even in Thief 2 and Shadows of the Empire.

My retro stuff

Reply 4 of 4, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

It's not known for sure what the assumed legacy table fog path is for and whether it's intentional that it doesn't get entered when some older game would benefit from it. Claude thinks this path is fairly similar to how table fog is handled in the G400 driver, unlike the other path, so make of that what you will. But it's a path that looks to be more compatible with older games in any case. I think Claude sorted out the patched selection rule in a way that a game that doesn't need it isn't likely to be affected, but who knows before testing.

I don't know about the Catalyst driver, but maybe.