BEEN_Nath_58 wrote on 2023-11-10, 13:19:

sharangad wrote on 2023-11-10, 13:12:

Why is Speedy3D faster in restart to dos mode? Seems a little weird.

1MHz difference?

Sorry, reading comprehension failure 😀

Nominal FIFO performance = 10.6 Mb/sec. ( Equal to DOS , higher than windows)
Nominal DMA performance = 28.1 Mb/sec. (Higher than DOS and WIN)

is raised in RTD against other modes, i didn't find any PCI delay settings in bios, its overclocing.
(some users i saw here , have FIFO performance faster than DMA), so they can start "vquake.exe -nodma" and vquake will uses FIFO instead of DMA access, and vquake will work faster)

Without any overclocks RestartToDOS performs faster, dmatest and vquake timedemos confirm it.

Its common practice of RTD is performing faster than DOS and WIN, RestartToDos get enhanced by Windows . Its hard to find any exact answers now. i somewhere reads about it, very long ago, and used a long time at practice, but its also amazed me on speedy3d performance , i used to think hardware accelerated games unaffected . =0

i believe if someone will have more CPU RAW power they can get better results.

sharangad wrote on 2023-11-08, 06:36:

Would it be possible for someone with a V1000 board to run this:
https://nirvtek.com/downloads/RTest.7z
and post the results?

Here you go, this is my result (Miro Crystal VRX):

1****************************************
2Board # 1
3Verite ID       0x1
4IoBase  0x6f00
5MemBase 0xf7000000
6MemMappedIoBase 0x0
7memSize 4194304
8IOBase  0x6f00
9BusMaster       0x1
10VGA Device      0x1
11****************************************

I noticed that it says BusMaster 1, does that mean it reports bus mastering as enabled? I think the motherboard I'm using does not have bus mastering support. I remember there was some tool from Rendition to test it, and it said bus mastering disabled. I did not try putting it into another PCI slot yet, maybe that will solve it.

asdf53 wrote on 2023-11-10, 18:54:

Here you go, this is my result (Miro Crystal VRX): […]
Show full quote

sharangad wrote on 2023-11-08, 06:36:

Would it be possible for someone with a V1000 board to run this:
https://nirvtek.com/downloads/RTest.7z
and post the results?

Here you go, this is my result (Miro Crystal VRX):

Copy code to clipboard

1****************************************
2Board # 1
3Verite ID       0x1
4IoBase  0x6f00
5MemBase 0xf7000000
6MemMappedIoBase 0x0
7memSize 4194304
8IOBase  0x6f00
9BusMaster       0x1
10VGA Device      0x1
11****************************************

I noticed that it says BusMaster 1, does that mean it reports bus mastering as enabled? I think the motherboard I'm using does not have bus mastering support. I remember there was some tool from Rendition to test it, and it said bus mastering disabled. I did not try putting it into another PCI slot yet, maybe that will solve it.

Thanks for that. Rredline is supposed to return the bus mastering status. It could be wrong.

I think the tool is "rendition dma test". You could run it again just to be sure. You can grab it from here
https://www.vogonsdrivers.com/getfile.php?fileid=79.

It's the same tool which reports higher figures in reboot to dos mode., as raven-05 points out.

One weird thing that I see is your mem Mapped io base is 0. It should have an address somewhere in vram., at least the v2000s do.

If bus mastering isn't available, then rredline is lying, possibly to maintain app compatibility.

RaVeN-05 wrote on 2023-11-10, 16:31:

Nominal FIFO performance = 10.6 Mb/sec. ( Equal to DOS , higher than windows) Nominal DMA performance = 28.1 Mb/sec. (Higher tha […]
Show full quote

Nominal FIFO performance = 10.6 Mb/sec. ( Equal to DOS , higher than windows)
Nominal DMA performance = 28.1 Mb/sec. (Higher than DOS and WIN)

is raised in RTD against other modes, i didn't find any PCI delay settings in bios, its overclocing.
(some users i saw here , have FIFO performance faster than DMA), so they can start "vquake.exe -nodma" and vquake will uses FIFO instead of DMA access, and vquake will work faster)

Without any overclocks RestartToDOS performs faster, dmatest and vquake timedemos confirm it.

Its common practice of RTD is performing faster than DOS and WIN, RestartToDos get enhanced by Windows . Its hard to find any exact answers now. i somewhere reads about it, very long ago, and used a long time at practice, but its also amazed me on speedy3d performance , i used to think hardware accelerated games unaffected . =0

i believe if someone will have more CPU RAW power they can get better results.

Vquake is mostly cpu bound. I think Stefan Podell had something to say about that on usenet. I didn't know about restart to dos being faster. More than 20 years after switching from w98, there's more to learn.

sharangad wrote on 2023-11-10, 19:36:

Thanks for that. Rredline is supposed to return the bus mastering status. It could be wrong. […]
Show full quote

asdf53 wrote on 2023-11-10, 18:54:

Here you go, this is my result (Miro Crystal VRX): […]
Show full quote

sharangad wrote on 2023-11-08, 06:36:

Would it be possible for someone with a V1000 board to run this:
https://nirvtek.com/downloads/RTest.7z
and post the results?

Here you go, this is my result (Miro Crystal VRX):

Copy code to clipboard

1****************************************
2Board # 1
3Verite ID       0x1
4IoBase  0x6f00
5MemBase 0xf7000000
6MemMappedIoBase 0x0
7memSize 4194304
8IOBase  0x6f00
9BusMaster       0x1
10VGA Device      0x1
11****************************************

I noticed that it says BusMaster 1, does that mean it reports bus mastering as enabled? I think the motherboard I'm using does not have bus mastering support. I remember there was some tool from Rendition to test it, and it said bus mastering disabled. I did not try putting it into another PCI slot yet, maybe that will solve it.

Thanks for that. Rredline is supposed to return the bus mastering status. It could be wrong.

I think the tool is "rendition dma test". You could run it again just to be sure. You can grab it from here
https://www.vogonsdrivers.com/getfile.php?fileid=79.

It's the same tool which reports higher figures in reboot to dos mode., as raven-05 points out.

One weird thing that I see is your mem Mapped io base is 0. It should have an address somewhere in vram., at least the v2000s do.

If bus mastering isn't available, then rredline is lying, possibly to maintain app compatibility.

Thank you for the link, I'll try the tool again next week with some different configurations. The 0 address could be due the Socket 4 motherboard I'm using. It has lots of problems with newer graphics cards, some will not boot or Windows cannot assign resources to it. It wouldn't surprise me if some of these values are abnormal. Next time I will put the card in my newer Socket 7 machine.

I don't know whether the null memory mapped port address is an error. Yours is the first v1k board I've seen rtest's output from.

It could very well be normal for that board.

I just noticed something interesting. Someone in this thread mentioned that performance improves when rebooting from Windows into DOS mode. I have tested it:

Vquake, booted straight into DOS: 20.3 fps
Vquake, rebooted into DOS mode: 22.0 fps

I then found out that the "dmatest.exe" tool ran from normal DOS says that bus mastering is disabled and DMA mode will not work. But when rebooted into DOS mode, it suddenly says DMA is enabled. I then benchmarked vquake with the "-nodma" switch, and performance drops again from 22.0 to 20.3 fps.

I had noticed this before when I booted into DOS mode and some games felt a bit smoother, but I thought it was just placebo. Turns out that it really does improve performance, in this case by working around a bug that seems to prevent enabling DMA mode.

And what's weird is that the dmatest benchmark gives me:

"FIFO performance 11.4 MB/s"
"DMA performance 10.4 MB/s"

Shouldn't that mean "vquake -nodma" is faster? But according to the timedemo benchmark, DMA mode wins.

And finally, when I run vquake or dmatest from a Windows 95 dos box, they both crash with a page fault exception, and a few seconds later, the computer freezes with a black screen.

No fifo performance there I think is speed of executing command buffers; like rendering commands.

Dma transfers is about texture load speeds/memory writes.

I'm drilling through the disassembley of dma test and I noticed the app behaves differently depending on the dos version number. It compares int21h func 0x30's return values ax al to 0x4565 or something like that. I'm trying to figure out which version of dos release that and what dosbox returns.

[EDIT] DOSBox staging appears to return 0.0 as the version number (AH=0/minor, AL=0/major, bh =0xFF, bl=0, cx=8)

Oh I also patched r launcher to allow modding the full screen underwater resolution using Raven-5's hack., for a future release. If anyone's interested I can upload it.

[EDIT]
https://1drv.ms/u/s!As-dKk-N73dSlmgkUWeR7nfxbsxE?e=DqUgoC
MD5: 4ffdf997dc1ba2158559cd4673dc21e5

Resolutions up to 800x600 can be selected underwater for vQuake 2 and vHexen 2 for fullscreen mode only (from app specific settings for quake2 and vh2). The game resolution should be at least as high as the underwater res.

The default res for vQuake 2 is 320x240 and vHexen 2 is 320x200.

Oh, okay. I somehow assumed that FIFO is a slower, alternative mode that is used when DMA mode is not available.

asdf53 wrote on 2023-11-11, 10:51:

Oh, okay. I somehow assumed that FIFO is a slower, alternative mode that is used when DMA mode is not available.

It is.
FIFO, first-in first-out. It's how regular commands are transferred I think. In the absence of DMA, fifo will also be used for textures. You're right . This is most probably how fast rendering instructions from a command queue can be transferred to the card and not how quickly it renderers..

For rendering commands, fifo's used always (I think). Someone could probably correct me. I think that's what happens because textures can be copied to vram, rendering commands cannot. Each rendering command gets moved for execution a few at a time.

Could you run this from DOS and tell me what it reports:
https://1drv.ms/u/s!As-dKk-N73dSlmnrUwVfHZMf4Q5s?e=5ImAeR
MD5: b2e417f0637d44b953d93c5e46ffc960

The exe's INT21.exe.

sharangad wrote on 2023-11-11, 11:07:

Could you run this from DOS and tell me what it reports: https://1drv.ms/u/s!As-dKk-N73dSlmnrUwVfHZMf4Q5s?e=5ImAeR MD5: b2e417f […]
Show full quote

Could you run this from DOS and tell me what it reports:
https://1drv.ms/u/s!As-dKk-N73dSlmnrUwVfHZMf4Q5s?e=5ImAeR
MD5: b2e417f0637d44b953d93c5e46ffc960
The exe's INT21.exe.

Sure, here's the result:

1return value 10.10
2bh 255 bl 0 cx 8
3SHR 10h = 0

Thanks. I think I can safely assume that dosbox and win98 dos run along the same codepath.

Dosbox returns 0.0 for the version number, like dos 2.0. But "ver" reports version 5.

Neither match the comparand.

asdf53 wrote on 2023-11-10, 18:54:

Here you go, this is my result (Miro Crystal VRX): […]
Show full quote

Here you go, this is my result (Miro Crystal VRX):

Copy code to clipboard

1****************************************
2Board # 1
3Verite ID       0x1
4IoBase  0x6f00
5MemBase 0xf7000000
6MemMappedIoBase 0x0
7memSize 4194304
8IOBase  0x6f00
9BusMaster       0x1
10VGA Device      0x1
11****************************************

Here are Creative 3D Blaster PCI. Both the V1000-E and V1000L-P gives the same resault:

1****************************************
2Board #	1
3Verite ID	0x1
4IoBase	0xfc00
5MemBase	0xfb000000
6MemMappedIoBase	0x0
7memSize	4194304
8IOBase	0xfc00
9BusMaster	0x1
10VGA Device	0x1
11****************************************
12
13Press [ENTER] to close

Thanks Gona. The MemMapped IOBase is null again. V2k cards have this set in VRAM space.

Quake 2 Res Unlock (Currently capped at 1152x864)

https://1drv.ms/u/s!As-dKk-N73dSlmpZlSqIflqPljGS?e=d2gjvn
MD5: a58a6ba01437899f75adb2db401e7685

Corrects bug with setting vQuake 2 underwate res and unlocks higher native resolutions (but only for vQuake2).

I should really stop pfaffing about with this and stick only to DOS.

Here's results from my Canopus Total3D V1000L-P.

1****************************************
2Board #	1
3Verite ID	0x1
4IoBase	0xe800
5MemBase	0xe1000000
6MemMappedIoBase	0x0
7memSize	4194304
8IOBase	0xe800
9BusMaster	0x1
10VGA Device	0x1
11****************************************
12
13Press [ENTER] to close

Yup once again the null MemMappedIoBase.

The other thing is there's only one card. If you run RTest on a v2k, there's a phantom second board with duff values.

[EDIT] I think I have enough samples for the v1k. Thanks guys. It's very consistent.

Does anyone know what INT 2Fh, function 0x1686 does?

Dosbox ignores the parameters and returns disk params.

And something similar with IN31h.

It's supposed to be used for communicating with a TSR. But there's no TSR involved here.

is there a TSR which needs to be installed for Speedy3D? If you create a fresh Win98 install on a Rendition system and try to run a game does it run or is there a config.sys element to it?