RabidHaMsTeR.Org

Been running 007 though its paces, and compiled a list of errors or otherwise impaired functionality.

Bubble on Biotunnel - MAJOR slowdown
Cubefield on Water - MALFUNCTION (Blackscreen) (PBuffer shows cubefield)
Emboss Tunnel - MALFUNCTION (Blackscreen fade to grayscreen)
GrooveDot Watermorph - Malfunction? (Background seems to work, figure is black, not water)
Radial Blurred Medusa - Appears simply to be very dim version of standard Medusa, but jiggles around a little. No blur.
Radial Blurred Cubefield - Very dim with slowly strobing lights, tumbling in cubefield. No blur.
Recursive Rectangles - Recursive medusas, not rectangles visibly
SnakeCube - Seems more to spaz out and wrap around itself rather than 'dancing' at all
SnakeCube Points - INCREDIBLY slow. May be related to an earlier problem with R2E Point Torus-Sparkball slowdown when using any Catalyst revision above 3.2
Water BioTunnel - MALFUNCTION (Blackscreen) (PBuffer shows tunnel)
Water Bubble - MAJOR slowdown
Water Medusa - Malfunction (Chrome medusa visible, but 'watery' Medusa visible in PBuffer.. frame referencing error?)
Water Morph On Cube - MALFUNCTION (No textures, just dark grey)
Water Point Morph - Malfunction (Point morph visible, but 'watery' point morph visible in PBuffer)
Water Tunnel - Malfunction (Water tunnel visible in PBuffer, but only black/white 'sketch mode' tunnel is drawn in main render window)

ERROR:
Started R4 with the PBuffer size set to 2048. Loading 'whited out' after 'monkey', but ran once loading completed. First effect ran, crash on transition to next effect. Error message follows.
!H:\R4\RenderTex.cpp:79>Couldn't create the PBUFFER []


Fade/merge transitions are an INCREDIBLE slowdown, sub-10fps at any resolution. Unable to test which transition effect directly (can't figure out how to specify which one to use).

Water effects seem 'broken', not being rendered.

Bubble effects are a MAJOR slowdown, dropping to ~4fps (if not outright measuring in seconds-per-frame). Standard framerate is around 300. Some kind of recursive flaw, perhaps?


System spec:
AMD Athlon TBird 1.4GHz
1.5GB SDR PC133 RAM
ATI Radeon 9500 Pro <Catalyst 3.8>
WindowsXP Professional

Settings:
PBuffer 1024
Resolution 1024x768x32

I believe the standard pbuffer size is 512. Do you still get all these problems at 512?

Can you set R4 down to 512x256 PBuffer size and see if the same errors occur? Everything you describe is exactly as it would happen if your video card is running out of memory.

If i up the pbuffer size over 512x512 on my PC, the same artefacts occur. There's nothing R4 can do about this since the video driver doesn't appear to tell it its not got enough memory (until the 2048x2048 case obviously).

Odd.. how many PBuffers are being created? The R9500 I have is a 128MB model (assuming it can use that for calculations as well as texture reference) so it'd seem like a 1024 or 2048 PBuffer size wouldn't be a problem.

In any case, I've tried it on the other PBuffer modes. The 'Bubble' effects only get to an acceptable framerate at a 128x128 setting (even then, only about 25-30fps if the framerate counter can be trusted).
Same with the SnakeCube Points effect. (I should install Catalyst 3.2 and see if this may be a driver issue, as Point Torus under R2E displayed a similar slowdown after upgrading to any later revision)

Water effects, however, are still dead. They simply don't render. Regardless of PBuffer size, it just is a black background. Perhaps a texture was forgotten in the latest beta package? Which makes little sense, as the 'water tube' and 'water' effects at least render _something_ to the screen. Shame there isn't a screenshot or movie function, to post comparison shots. :b

Though the 'Emboss Tunnel' effect is simple to describe. Blank gray screen.

Did a bit more research (testing on another machine with an R9000 Pro and an earlier catalyst revision) and will need to roll this machine back to Cat3.2 to find out if it's just a driver issue for some odd reason, with the newest Cats. Additionally, the R9000 could run in 2048^2 PBuffer mode with only marginal slowdown... 130fps average down to 100fps, even on the 'bubble' scenes.

Unfortunately, even on the R9000 the water effects did not render. Am I missing something? I recall something about using a video capture device in conjunction with the water, but last I'd read that was not working yet.

May snag a trial-copy of Fraps to grab a demo movie... fifteen seconds should be enough to demonstrate the apparent error.

The pbuffer sizes are deceptive. on an ati dx9 accelerator card, 24 bits are used per color channel, so at a minimum on 1024x1024 you're talking:

1024*1024*((3*3)+1) = 10 Megabytes
Infact it may be more. R4 will usually allocate around 15 of these, so its not too surprising your 128mb card runs out. To be honest at 1024x768 you will notice very little quality difference between 1024x1024 and 1024x512. Even less is quite acceptable.

anyway, 2048x2048 will be over 40mb per texture, and on a 256mb card you can't manage many of them either. I think putting the pbuffer size up you're just asking for trouble.

I have no idea why the water effects wouldn't be displaying. Its extremely unlikely i missed out a texture because a lot of the water effects would still work fine without. There is also a big difference between the water tube and the water effects that wouldn't be working for you. the water effects that aren't working do use a module called DIFF that may not be working on the card i guess.

Did you say that the radial blurred medusa doesn't work too?

If thats the case i'd say the issue is with the new radeon's support of ARB_RENDER_TEXTURE. God knows why that wouldn't go though - i changed one thing to try and reduce the memory used on gfx cards, but i think i may put an option in to turn it back and see if that fixes it.

- Gordon

Through my 'CRASH COURSE' I remember/but don't at the moment seeming to recall that wasnt the case. ** OF COURSE not knowing exactly what I was reading I will dig through and see if I can find that info.Did you look into ATI bits per channel color allocation?

Starting to see a pattern involving ATI getting wet. If I recall wasnt there only 1 person w/a ATI card capable of getting the water scenes to work correctly?

Since your re/pre/render method (does that make sense ?) does R4 only render/allocate whats pre-defined or is there some internal minimum? Because even w/just 1 predefine (WATER SCENE) the result is still black.

Radial blurred medusa and cubefield work fine on the R9000, or at least a blurring effect is visible. The 9000 is using a pre-3.8 version of the catalyst drivers (3.4 comes to mind, off-hand), and everything appears to work excepting the water effects. Even the snakecube point effect runs very snappily.. something that cannot be said for the R9500. As well, any ideas why the R9000 would be able to operate in 2048x2048 mode with no problems then? They are both 128MB models, so I have no idea. :b
Admittedly the R9K is a DX8 part, and the R9500 is DX9... perhaps something in the channel specifications.

Still as of yet to roll the Catalysts back to 3.2 on the R9500 for testing. In the past, it handled R2E just fine at 1600x1200x32, with FSAA and Aniso popped up to maximum. And while that may be a bit excessive, it IS very pretty.

Seems weird that the later drivers would run slower on applications like this, but I suppose it could be counted as 'shader ordering optimizations' or somesuch. :b

I'll post what happens after the drivers have been rolled back.

GUEST (Not that there are hundreds, but could you include a screen name/register just to help when making references)

A while back I had found R2e performed better w/certain older drivers but lately seems R2e/R4 do better w/the latest.

I know its NOT common practive for ANY vendor to publicly release complete driver changes but anyone out there a ATI/nVidia partner or aware of one that has a webpage w/such info?