tincup wrote on 2025-03-27, 20:52:

Quick observations. Carsets seem to play the most significant role in performace. Generally the simpler carsets exhibit negligib […]
Show full quote

Quick observations. Carsets seem to play the most significant role in performace. Generally the simpler carsets exhibit negligible dropoff during formation laps and first corner or two before the pack spreads out. OTOH the whole library of nicely modelled and decaled Pavel cars 1990-2001 (lola1990-95, reyn1996-2000) all suffer a consistent 10-20 FPS dropoff more or less regardless of track: ie mid 40's/ low 50's, with a rare occasional dip into the mid/high30's. But even these sets snap back to 60 pretty soon after it goes green and the cars spread out. Maybe the Reyn99 car suffers a bit more than the others? Tracks *might* contribute about 5 FPS? but then again only with the heavy carsets.

IRL carsets (irl96a-b thru irl99a-b etc.: Pavel/Claudio I beleive): negligible fall off - may be based on the simpler 2k model?

Champcar sets (Blackcat indy05 & indy06, CCWS2k6): much larger ~8mb dats and wonky issues. Even reduced won't always load (or load consistently) some reduced tracks sometimes a save error loading/switching between these cars - and the CTD mentioned before. Also a bigger dropoff to low 30's. But again these aren't true CART era cars.

So right now it looks like carsets have the biggest performance impact pound for pound.

You might be on to something:

Here's Houston with Indy2k08: https://youtu.be/8eVz_2vNuBE

Without physics it runs at a solid 30 fps (60 interpolated) and with physics its like molasses, much worse than 1982cart.

tincup wrote on 2025-03-27, 20:52:

EDIT: Probably the worst combo I've seen is reyn1999 + Clevland (7-23-2012 dat); formation lap never above 30s.

Where can I find these? I think the problem with the carset is that there're a lot of cars and they all have to be rendered at high detail.

I've made a terrible mistake. The slowdown is as you said due to the cars and not the physics. Once "Draw" is set to 32 ahead and 32 behind the slowdown becomes significant with a load of cars in there.

So it's the number of cars. I'll profile the game with these settings. I'm also seeing DMA transfer errors with the game (corrupti0n and lockups).

Cleveland 85 and Houston with the Reynard 1999 carset. Once setting the "Drawing" to 31 ahead and 31 behind ruins the performance:
https://youtu.be/06I5tw3Bf4o

sharangad wrote on 2025-03-27, 21:25:

tincup wrote on 2025-03-27, 20:52:

EDIT: Probably the worst combo I've seen is reyn1999 + Clevland (7-23-2012 dat); formation lap never above 30s.

Where can I find these? I think the problem with the carset is that there're a lot of cars and they all have to be rendered at high detail.

You'll find Pavel's reyn1999 and his other carsets/tracks in this link. I found it in a forum thread at ICR2.net - they should really sticky it in the forum.

https://drive.google.com/drive/folders/1MjSDl … Nh25495Q6DcZBz1

Also I can confirm this carset is taxing on performance - I had to prune the Draw Ahead number a fair amount to even start in mid 40's, say at Gateway98. Still drivable, but starts are also critical to get through cleanly so FPS are your friend.

tincup wrote on 2025-03-28, 00:24:

You'll find Pavel's reyn1999 and his other carsets/tracks in this link. I found it in a forum thread at ICR2.nat - the link shou […]
Show full quote

sharangad wrote on 2025-03-27, 21:25:

tincup wrote on 2025-03-27, 20:52:

EDIT: Probably the worst combo I've seen is reyn1999 + Clevland (7-23-2012 dat); formation lap never above 30s.

Where can I find these? I think the problem with the carset is that there're a lot of cars and they all have to be rendered at high detail.

You'll find Pavel's reyn1999 and his other carsets/tracks in this link. I found it in a forum thread at ICR2.nat - the link should really be stickied in the forum.

https://drive.google.com/drive/folders/1MjSDl … Nh25495Q6DcZBz1

Also I can confirm this carset is taxing on performance - I had to prune the Draw Ahead number of cars a fair amount to even start in mid 40's, say at Gateway98. Still drivavle for sure though.

Thanks for that! I'll check out those later today.

tincup wrote on 2025-03-28, 00:24:

You'll find Pavel's reyn1999 and his other carsets/tracks in this link. I found it in a forum thread at ICR2.net - they should r […]
Show full quote

sharangad wrote on 2025-03-27, 21:25:

tincup wrote on 2025-03-27, 20:52:

EDIT: Probably the worst combo I've seen is reyn1999 + Clevland (7-23-2012 dat); formation lap never above 30s.

Where can I find these? I think the problem with the carset is that there're a lot of cars and they all have to be rendered at high detail.

You'll find Pavel's reyn1999 and his other carsets/tracks in this link. I found it in a forum thread at ICR2.net - they should really sticky it in the forum.

https://drive.google.com/drive/folders/1MjSDl … Nh25495Q6DcZBz1

Also I can confirm this carset is taxing on performance - I had to prune the Draw Ahead number a fair amount to even start in mid 40's, say at Gateway98. Still drivable, but starts are also critical to get through cleanly so FPS are your friend.

It's not the wrapper side texture loading that's killing performance. But I can't be sure of that because recent Windows updates have killed my profilers (visual studio profiler broken for over 2 years, Intel profiler broke with last update) . They don't profile child processes and can't even attach to running processes.

[EDIT] Running as admin seems to have fixed the profiler problem.

tincup wrote on 2025-03-28, 00:24:

EDIT: Probably the worst combo I've seen is reyn1999 + Clevland (7-23-2012 dat); formation lap never above 30s.

The slowdown isn't because of the reduced MIPS, even a fully MIPped Reyn1999, slows down on the built-in Australia circuit when [ALT+X] (demo mode ) is enabled. With Cars 95 demo mode produces 60 fps, with reyn99 it hovers around 36 fps.

Texture load times in the wrapper are 50 % (1.5x total) more for the Oz track (60ms to 90 ms).

[EDIT] Disabling texture loading doesn't improve the frame rate at all.
[EDIT2] Neither does using the lower middle 50% mips or shrinking them, the slowdown in Houston is always there with reyn1999.

1982CART has only front wheels circular, while back wheels are *gonal
reyn1999 has both front and back wheels circular

robertmo3 wrote on 2025-03-30, 02:30:

cars1984 has only front wheels circular, while back wheels are *gonal
reyn1999 has both front and back wheels circular

Is that after reduction?

robertmo3 wrote on 2025-03-30, 02:30:

cars1984 has only front wheels circular, while back wheels are *gonal
reyn1999 has both front and back wheels circular

Where'd you get cars1984 from? I don't seem to be able to find it.

sorry i meant 1982CART

robertmo3 wrote on 2025-03-30, 02:30:

1982CART has only front wheels circular, while back wheels are *gonal
reyn1999 has both front and back wheels circular

Is that in the low detail mips?

original unmodified v2 from here
https://www.icr2.net/forum/showthread.php?tid … hlight=1982cart
you can see *gonal back wheels even in driver info

robertmo3 wrote on 2025-03-30, 06:52:

original unmodified v2 from here
https://www.icr2.net/forum/showthread.php?tid … hlight=1982cart
you can see *gonal back wheels even in driver info

I already have the original 1982 cart v2, never noticed it.

robertmo3 wrote on 2025-03-30, 02:30:

1982CART has only front wheels circular, while back wheels are *gonal
reyn1999 has both front and back wheels circular

Not noticing in carset selection.

hexadecagonal

tincup wrote on 2025-03-28, 00:24:

You'll find Pavel's reyn1999 and his other carsets/tracks in this link. I found it in a forum thread at ICR2.net - they should r […]
Show full quote

sharangad wrote on 2025-03-27, 21:25:

tincup wrote on 2025-03-27, 20:52:

EDIT: Probably the worst combo I've seen is reyn1999 + Clevland (7-23-2012 dat); formation lap never above 30s.

Where can I find these? I think the problem with the carset is that there're a lot of cars and they all have to be rendered at high detail.

You'll find Pavel's reyn1999 and his other carsets/tracks in this link. I found it in a forum thread at ICR2.net - they should really sticky it in the forum.

https://drive.google.com/drive/folders/1MjSDl … Nh25495Q6DcZBz1

Also I can confirm this carset is taxing on performance - I had to prune the Draw Ahead number a fair amount to even start in mid 40's, say at Gateway98. Still drivable, but starts are also critical to get through cleanly so FPS are your friend.

DMA transfers take more than twice (~2.5) as long with reyn1999 compared to Cars95 with the Houston track. I think the named pipe comms and the RRedline command buffer construction is taking too long, they're probably saturating a core each. Need to find some way to optimise it.

robertmo3 wrote on 2025-03-30, 08:49:

hexadecagonal

So they are! My bad.

the easiest way to verify is to start practise, switch view to "close external" with f10, drive right to hit right band (you are rotated 90 degrees now) and move back to hit opponent, wait till camera rotates back to front, and move slightly forward so that you have opponet's wheels at the bottom of the screen