@Seiya
2 fps for real hw and gallium also the same strange and wrong as 3600/7200 on vmware. Something wrong or with wazp3d->mesa transfers, or with cow3d , or with dunno what..

Just in sake of interest, can you also check Quake3 on the real HW + Gallium3d ? I.e. usual 1280x1024, VSYNC OFF, "timedemo1 / demo four" ?

_________________
Join us to improve dopus5!
zerohero's mirror of os4/os3 crosscompiler suites

@phoenixkonsole

I was missing a word "NOT", did not notice before now

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@NutsAboutAmiga
No problem. Many people are not aware about this and believe "hosted" to be emulated or virtualized.
It is not. For the host os it is like any other application, aros hosted itself believes to be a native os and acts internally also that way.

So a aros hosted audio.device driver is not talking to the soundcard but to the sound api from wimdows or linux.

Well windows version is lacking network and audio support.
Linux version supports oss-audio, network and 3d with hostgl.

They (aros hosted versions) act lice native windows or linux binaries. * i know i am repeating but maybe someone else reads this as well ; )

Last edited by phoenixkonsole on 13-Apr-2015 at 06:59 PM.

_________________
AROS Broadway - AEROS - Aminux - AmiCloud - indieGO! Appstore - AmiWallet - VAN lossless video codec - AMC Amiga media Center -KrypUnite - LibertyNet - MinX - amigaNX

as far as i know aros native wazp3d calls aros mesa and then it depends if mesa runs softpipe hidd or any of the accelerated ones. still, i dont have much experience with it as im not much into arosx86, except for testing, which i do either linux or windows hosted.

@kas1e

i tested Quake3 in the past with icaros 1.3.1 with 166 fps in 1280x1024

_________________

@OlafS25

On Windows you are using either the built in rasterizer of Wazp3D's, or softpipe in Mesa. Both of them are software rendered. Only the Linux-hosted version can make use of the host OS' OpenGL driver.

Last edited by BSzili on 13-Apr-2015 at 08:27 PM.

_________________
This is just like television, only you can see much further.

@Seiya
Quote:

i tested Quake3 in the past with icaros 1.3.1 with 166 fps in 1280x1024

Remember what kind of HW it was ?

_________________
Join us to improve dopus5!
zerohero's mirror of os4/os3 crosscompiler suites

I strongly suspect that the MiniGL/Warp3D workflow will strangle performance on RadeonHD since it prevents some of the most critical parts of the rendering pipeline from being moved to the GPU, meaning the CPU will be the determining step in all but the very highest of resolutions.

Let's face it, even R200 class systems are already painfully underutilised (they have hardware TCL but there's no software support). That only widens with the Radeon HD cards.

_________________
Doing stupid things for fun...

@Karlos
As we see from tests, even with minigl/warp3d not everythng up to CPU, as on SAM460 ppls have 120 fps wth Cow3D test, and x1000 ppls have about 90 FPS, while CPU in x1000 is cleary faster than CPU in sam460. What mean, that even with minigl/warp3d, another stuff is in deal, and not only CPU (cashes, buses, main memory, pci-e controller differences and all that stuff).

_________________
Join us to improve dopus5!
zerohero's mirror of os4/os3 crosscompiler suites

@kas1e

As I see it Aros Warp3D uses Mesa3D and that directly uses the Win32 API and not directly the 3D card but it is still way faster than the AmigOS solution that in theory can directly use the graphic card. So there must be a bottleneck that makes it much slower than necessary. The difference should be much lower and at least the X1000 + modern graphic card and new drivers and 3D accelleration should be faster than even my Windows hosted environment without 3D accelleration. What do you think?

Last edited by OlafS25 on 13-Apr-2015 at 10:35 PM.

@kas1e

the same that i used today:

Intel Core2 Duo E8500@3,16 Ghz
4 GB DDR2 800
Geforce GTS 250 1 GB GDDR3

_________________

@OlafS25

Quote:

OlafS25 wrote: @kas1e As I see it Aros Warp3D uses Mesa3D and that directly uses the Win32 API and not directly the 3D card but it is still way faster than the AmigOS solution that in theory can directly use the graphic card. So there must be a bottleneck that makes it much slower than necessary. The difference should be much lower and at least the X1000 + modern graphic card and new drivers and 3D accelleration should be faster than even my Windows hosted environment without 3D accelleration. What do you think?

I realise that your AMD A6-4400M APU has impressive specs. but, assuming zero overdraw, cow3dv5 running at 3600 fps is ~1.1 Gpixels/s (G meaning 1e9 here). Seiya says he gets alternating 3600fps and 7200fps; 7200fps is ~2.2 Gpixels/s. If Wazp3D's software renderer really manages that, then thellier has done an amazingly impressive job of optimizing it. Without GART support there is no way that the W3D_SI driver could even feed cow3d's vertices to the GPU fast enough to match that speed.

I do wonder if something is going wrong with cow3d's fps measurement on your machine (and Seiya's), because those results sound fishy. They appear to be out of line with other results.

Hans

Last edited by Hans on 14-Apr-2015 at 03:00 AM.

_________________
http://hdrlab.org.nz/ - Amiga OS 4 projects, programming articles and more. Home of the RadeonHD driver for Amiga OS 4.x project.
https://keasigmadelta.com/ - More of my work.

@Hans
We could run "gears" demo to measure frames.

Last edited by phoenixkonsole on 14-Apr-2015 at 06:05 AM.
Last edited by phoenixkonsole on 14-Apr-2015 at 06:04 AM.
Last edited by phoenixkonsole on 14-Apr-2015 at 06:03 AM.

_________________
AROS Broadway - AEROS - Aminux - AmiCloud - indieGO! Appstore - AmiWallet - VAN lossless video codec - AMC Amiga media Center -KrypUnite - LibertyNet - MinX - amigaNX

@seiya
Quake3 results doesn't looks too good then for such machine imho. Probably on the same HW but under windows/linux (with proprietary driver) it will give about 300-400fps ?

But at least quake3 results looks real, while cow3d with all those 3600 and 7200 vmware and window hosted, and 2 fps on real sounds like something is wrong with cow3d itself.

@Olaf
Quote:

What do you think?

Dunno, its all mess for now :) But what for sure, is that cow3d results with 3600 and 7200 are wrong. You may try to run "gears" demo about which phoenixkonsole says indeed, just to see how it will be.

In end of all we probably need some other test case for all the oses. Maybe indeed "gears" will be better. But , it will show us only opengl speed, while we want to know speed of the 3d drivers only, without involving additional layer (like minigl on os4, which will slow everything down).

Try that "Gears" on window-hosted, how much will you have ?

@Hans
Quote:

I do wonder if something is going wrong with cow3d's fps measurement on your machine (and Seiya's), because those results sound fishy. They appear to be out of line with other results.

It is results from window-hosted (olaf) and vmware (seiya) , while on real hardware (seiya) cow3d strangely give only 2 (!) fps. What mean that indeed something wrong somewhere.

Last edited by kas1e on 14-Apr-2015 at 07:50 AM.
Last edited by kas1e on 14-Apr-2015 at 07:37 AM.

_________________
Join us to improve dopus5!
zerohero's mirror of os4/os3 crosscompiler suites

@Karlos

I have to agree. The relatively high CPU usage of MiniGL (10-15%) shows that doing all the vertex transformations and clipping on the CPU takes its toll.

_________________
This is just like television, only you can see much further.

@kas1e

Quote:

kas1e wrote: In end of all we probably need some other test case for all the oses. Maybe indeed "gears" will be better. But , it will show us only opengl speed, while we want to know speed of the 3d drivers only, without involving additional layer (like minigl on os4, which will slow everything down).

I don't think that we're going to find a test that everyone agrees is fair. With Warp3D-only tests some will complain about the overhead of emulation wrappers, while with OpenGL we have the added overhead of MiniGL on OS4.

On the bright side, this'll give people plenty to discuss/argue about... until the cows come home.

Hans

_________________
http://hdrlab.org.nz/ - Amiga OS 4 projects, programming articles and more. Home of the RadeonHD driver for Amiga OS 4.x project.
https://keasigmadelta.com/ - More of my work.

@all

Yes on Aros/X86 Wazp3D use ArosMesa so Gallium3D (if Wazp3D-Prefs is set to renderer:hard)
But I dont know if "Aros hosted" support that feature ...

If Aros/Gallium really support hardware acceleration on this machine then hardware accelerated Gallium3D is used
Else use Gallium3D "softpipe" = slooow software rendering

Cow3D is not adapted to very fast machine as it use clock() for timing = that is not precise enough for very small time interval so that may lead to crazy values

You should then replace in this part FRAMESCOUNT*2 to (say) FRAMESCOUNT*20

if (++FramesCounted >= (FRAMESCOUNT*2)) /* ie after the object turned two times */
{
now = clock();
FPS = ((FRAMESCOUNT*2)*CLOCKS_PER_SEC) / (now-last) ;
last = now; FramesCounted = 0;




Alain Thellier

@Hans

If I understand it right Warp3D (based on Wazp3D) is just a kind of wrapper on MESA3D and that is certainly directly using the Windows API. I do not know if FPS are correctly calculated of course or for some reasons calculated wrong.

I have just tested Gears (that is drectly using MESA without Warp3D) 252 FPS

@phoenixkonsole

Gears should be not using Warp3D on Aros because there is MESA/Gallium there

@Hans

Would it be worth normalising the results against 'some linux benchmark' per machine?
I'm sure there'll still be some variance, but it could be a useful guide and help rule out hardware as a cause (eg bus speeds).

_________________
Troll - n., A disenfranchised former potential customer who remains interested enough to stay informed and express critical opinions.
opp., the vast majority who voted silently with their feet.