WinUAE 1.5.0 beta series

[Toni Wilen]

Originally Posted by Ian

Here are my test results using all modes.

Again, for the benefit of the tape.

Vista Ultimate x64
Athlon x64 3500
2 x 1gb PC3200 DDR
ATI x1950 xt 256mb


Latest drivers all installed

1.4.6 results for comparison? Is this PCI-e or AGP card?

[Ian]

PCI-E x16

Attached 1.4.6 results.

[turrican3]

agp 7800 gs
cpu e6600 2,4 ghz

good results with beta10 default and ddmode 1 and 2.( all really better than 1.4.6 )
worst reults with beta 10 ddmode 3 and 1.4.6.
good job toni 1.4.6 look really old now.
Look by yourself

[AmigaSurfer]

Originally Posted by Toni Wilen

Oops, it is -ddforcemode not -ddforceram, sorry. (last minute change..)

OK, here are my test results again with -ddforcemode.

The ddforcemodes 1-3 are not faster then without ddforcemode.

Then I changed my Radeon X1950Pro 512MB AGP from AGP 8x to AGP 4x and done some tests again.

With WinUAE 1.4.6 and AGP 4x all test results are the same than with AGP 8x.
With WinUAE 1.5.0beta10 most tests results with AGP 4x are half or less than with AGP 8x.

So the bottleneck at high resolutions seems to be the AGP port. WinUAE 1.5.0beta10 seems to transfer much more data over the AGP port than 1.4.6.

With -ddforcemode 1 with AGP 4x the tests Draw(), BlitBitMap() and BlitBitMapRastPort() are faster? than with AGP 8x, but still much slower than with WinUAE 1.4.6.

[Toni Wilen]

Originally Posted by AmigaSurfer

Then I changed my Radeon X1950Pro 512MB AGP from AGP 8x to AGP 4x and done some tests again.

Do you have AGP fastwrites enabled in bios? (and have latest AGP driver)

So the bottleneck at high resolutions seems to be the AGP port. WinUAE 1.5.0beta10 seems to transfer much more data over the AGP port than 1.4.6.

Unfortunately new method makes emulated VRAM extremely fast (it was very slow previously) but it also needs much more data to be transferred to real VRAM. I am still trying to optimize this more..

I still think P96Speed does not test real word performance (whole screen updates/every frame rarely happens with applications, games are difference but games rarely run at 1000+ resolutions..)

[AmigaSurfer]

Originally Posted by Toni Wilen

Do you have AGP fastwrites enabled in bios? (and have latest AGP driver)

My mainboard is an ASRock AM2NF3-VSTA. This is IMHO the only mainboard with AM2 socket and AGP and five PCI slots. But it have an NVIDIA nForce3 250 chipset for which Nvidia doesn't update drivers anymore. Except in her own graphics card drivers.

AGP fastwrites is enabled.
ATI Catalyst is 8.3.
Controlpanel says:
NVIDIA nForce3 250 AGP Host to PCI Bridge is 4.3.6.0 from 02.04.2004.

The file nv_agp.SYS is 4.12.1.436 from 03.09.2006 15:54.
This must be from the driver package AllIn1_MCE_XP_2K(511)Special.zip.
I don't know a newer driver than this for Windows XP.

PS: Does anyone know newer drivers or does anyone know a pci-e mainboard with five PCI slots?

I still think P96Speed does not test real word performance (whole screen updates/every frame rarely happens with applications, games are difference but games rarely run at 1000+ resolutions..)

I play "Die Siedler - Aufstieg eines Königreichs" in 1920x1200. ;-)

[Toni Wilen]

and another Picasso96 speed test version.. Perhaps this one is faster in very high resolutions.

[AmigaSurfer]

Originally Posted by Toni Wilen

and another Picasso96 speed test version.. Perhaps this one is faster in very high resolutions.

Yes, some test are faster with this version. Thankyou Toni for your continuous work.

[Ian]

Try this.

http://support.ati.com/ics/support/d...estionID=31542

AGP support in the Catalyst drivers has been broken for a long time.

[turrican3]

sorry but it's a little bit slower for me 1280x720

[Toni Wilen]

Originally Posted by turrican3

sorry but it's a little bit slower for me 1280x720

I have noticed results vary at least -+10% between runs. So nothing to worry about.

[Toni Wilen]

Last (?) P96Speed test version, using yet another way to update the display. (it is either this or previous that will be used in final)

Only use with P96Speed, this version will be slow in normal use. (because it only does full screen updates, I'll only implement partial updates if this is faster than previous in P96Speed)

Include also -forceddmode 1 to 3 tests, thanks.

[AmigaSurfer]

Originally Posted by Toni Wilen

Last (?) P96Speed test version, using yet another way to update the display. (it is either this or previous that will be used in final)

Only use with P96Speed, this version will be slow in normal use. (because it only does full screen updates, I'll only implement partial updates if this is faster than previous in P96Speed)

With beta10 2008.03.25 most test are slower than with beta10 2008.03.24.

Include also -forceddmode 1 to 3 tests, thanks.

-forceddmode or -ddforcemode?

[turrican3]

not faster toni, i think you can keep the previous one

[paranoid]

I'm late getting to the party so this may not be useful, but thought I'd check it out to see what kind of improvement my system might see. If I should run additional benchmarks or something though, just say so and I'd be happy to. All in all, a very noticeable improvement with my system.

XP SP2
3.2 ghz P4 with HT
GeForce 7800 GS (latest drivers, 169.21)
Using a rather standard AmigaSYS install, OS3.9, only odd thing that may figure in (though I doubt) is that I just installed some math .libraries supposedly optimized for WinUAE. 040, JIT. CPU Idle full to the left. 1024 window in a 32-bit 1280 Windows desktop.

Basically -ddforcemode 1 was faster in most everything. BitMapScaling() bounced around a bit, may be nothing though. Without a commandline option, the results were some better, some worse than -ddforcemode 1, but BitMapScaling() was highest of all tests (except fullscreen). (To hazard a guess, just normal anomalies with the testing...)

Out of curiosity, I tried it with defaults (no -ddforcemode) and switched it to fullscreen and 1280x1024 (as opposed to a 1024x768 window). It was a bit slower here and there, never faster -- with one (broken) exception: the raster lines were blazing fast (but weren't scrolling, always in just one row/column).

I enclosed the txt files and a copy of the Machine.dat as well.

ETA: Sorry if some of this has been addressed (like perhaps "don't run this in fullscreen").. If so, just ignore it all.. I did read most of the article, got to testing, and after a couple goofs on my part, it's been a few hours and just wanted to post what I'd done. Still a sweet improvement, regardless.

[thomas]

Originally Posted by Toni Wilen

This surely isn't 1.5 specific, offtopic

It is a bug and it happens in 1.5.0, so yes, IMHO it is related.

Here is another one: I tried to delete a config together with its corresponding HDF file from within WinUAE, but it crashed.

I went to Configurations, selected the config and clicked on delete, it was deleted successfully. Then I went to Hard drives, clicked on "add harddrive", on "...", right-clicked on the HDF and selected Delete. It was successfully deleted. Then I tried to return to Configurations, but it detected a crash.

[Toni Wilen]

Originally Posted by thomas

It is a bug and it happens in 1.5.0, so yes, IMHO it is related.

Allowed posts in beta thread (not yet posted):

- bug reports but make sure bug is NOT in previous offical version. (Old bug -> separate thread)

Related yes but still off-topic

Here is another one: I tried to delete a config together with its corresponding HDF file from within WinUAE, but it crashed.

I went to Configurations, selected the config and clicked on delete, it was deleted successfully. Then I went to Hard drives, clicked on "add harddrive", on "...", right-clicked on the HDF and selected Delete. It was successfully deleted. Then I tried to return to Configurations, but it detected a crash.

This is another one that only happens rarely, near-impossible to debug.. (and also happens in older versions)

[Toni Wilen]

Originally Posted by paranoid

Basically -ddforcemode 1 was faster in most everything. BitMapScaling() bounced around a bit, may be nothing though. Without a commandline option, the results were some better, some worse than -ddforcemode 1, but BitMapScaling() was highest of all tests (except fullscreen). (To hazard a guess, just normal anomalies with the testing...)

It seems mode needs to be configurable, on some PCs default (0) is faster but others 1 is faster.. I guess registry-stored setting somewhere in GUI is good enough.

with one (broken) exception: the raster lines were blazing fast (but weren't scrolling, always in just one row/column).

It is bug in P96Speed, ScrollRaster-tests are NOPs with very high resolutions for some reasons.

[Toni Wilen]

winuae.net working again.

http://www.winuae.net/files/b/winuae_1500b11.zip

Beta 11:

Picasso96 rewrite finished, bug fixes remaining..

- uaegfx.card is now WinUAE built-in, libsicasso96/uaegfx.card is not needed anymore (and is ignored if it exists)
- resolution names in display preferences are now named "UAE:<w>x<h>" instead of "uaegfx:<w>x<h>"
- added new Picasso96 options to GUI
- 24-bit display depth Picasso96 blitter routines implemented/updated
- added all Picasso96 supported color depth modes, defaults are old standard PC modes (palette mode, R5G6B5PC and B8G8R8A8)
* only one 8-bit,15/16-bit, 24-bit and 32-bit mode supported simultaneously
* All "new" modes require color space conversion = slower.
* Not all mode color space conversions yet supported (black screen)
* Can be useful with old RTG compatible demos or games that require specific "non-pc" display color modes
- added "Display buffer" option to misc-panel (stored to registry)
- "FS" scale setting is now the default
- refresh whole screen using very optimized routine if >80% of Picasso96 screen needs refreshing (faster than partial update)
- Marvin's Marvellous Adventure clouds/score interference fixed (It was really stupid bug, can only happen with AGA sprites)
- Skid Marks hires mode car sprite problem fixed (b10)

[Marcuz]

just a question: the problems with Picasso mode full screen + filters reported a page ago, are still there, right?

everything is fine in full screen (UAE 800 x 600 16BIT) until i enable filters (any filter, even null filter, > error is "createdevice failed") or whenever inside the emulation i try to change the resolution through prefs.

It is prolly something i do wrong, but i'm not sure what.