Amithlon & GFX-cards: any advice?

[humppa]

Today I finally received my FX5200, hooked it up, booted into Amithlon and only got a black screen.

Then I found out that DVI is not working! Analogue VGA display works well though. Am I really the first one who can't get DVI to work on a Geforce FX?

Do I have to change any settings in order to get DVI working?

Here is some information, maybe that helps tracking down the problem which is hopefully not hardware-related (DVI works in Windows):

When booting, I get this:
"rivafb: nvidia device/chipset 322"
"rivafb: RIVA MTRR set to ON"

At the end of booting, the text "Opening framebuffer: no such device" is displayed. The FX is well detected and HW-acceleration is working:
"xcat /proc/fb    0   GeForce FX 5200"
I can select between the "builtin" screenmodes and "NVIDIA-1" screenmodes.
With "vga=***" I tried different modes, but I think these only have any effect for VESA cards, right?

Do I have to set "video=***" to something? Somewhere I found the setting "video=rivafb:640x480,flatpanel=1" but that only resulted in two black modes (VGA and DVI) when I tested it.

I could also try different "setconfig >NIL: p96_accell" settings, but I don't really know where to start or if those have any effect at all.

Another oddity with "kern310":
"xcat /proc/version" gives me:
"3.0.0-test1(2.4.19-pre4) compiled 19 Dec 2005"
Am I using an old kernel or is it just a version number/date that wasn't bumped?

Thanks for any help in advance!

This is the GFX-card that I bought.

[humppa]

Whohooo! Finally DVI!

The Geforce did not seem to autodetect my TFT panel, so I got it working using
"video=riva:640x480,flatpanel=1".
Before I got confused from reading too much on Linux-boards about kernel options and I must have accidently typed in "rivafb" instead of just "riva". Doh!  :crazy:

[DrBombcrater]

@humppa

Nice to hear you got it working  :-D

You had me worried for a moment there, because the DVI code has always worked on every FX card I've tried it on!

DVI auto-detect is broken on GeForce cards. This is listed as a known issue for 3.00, but also happens under 3.10 (and you are running 3.10, stupid me forgot to update the version string :oops:)

Edit: the kernel agruments stuff is a right pain in the bum. I spent hours wondering why the FX driver was giving me a screen full of random garbage on my laptop, it turns out I typed 'video=riva:640x480, flatpanel=1' instead of 'video=riva:640x480,flatpanel=1'.

Just adding one space caused the flatpanel argument to be ignored...

[humppa]

DrBombcrater wrote:
@humppa

Edit: the kernel agruments stuff is a right pain in the bum. I spent hours wondering why the FX driver was giving me a screen full of random garbage on my laptop, it turns out I typed 'video=riva:640x480, flatpanel=1' instead of 'video=riva:640x480,flatpanel=1'.

Just adding one space caused the flatpanel argument to be ignored...

Indeed. I spent the whole afternoon trying to figure it out. My missing low-level Linux experience wasn't helpful either. The output during boot "rivafb" must have made me assume that the same string could be used as a kernel argument.
Btw, I didn't know that you can even display GFX card mem and AGP aperture size from Amithlon: "xcat /proc/mtrr".
Don't know if that could be used for anything, e.g. relocating the AGP-area to circumvent the 1GB limit. Bah, I'm talking out of my butt again.  :-P

Btw, did anyone ever manage to get 1600x1200 resolutions working? I got a 20" TFT, so it could be nice running at native resolution. At the moment, all resolutions seem to run at 1280x1024 on accelerated drivers. When selecting e.g. 1024x768, a black border is displayed around the view area and my TFT still recognizes it as 1280x1024. I think you wrote in some readme that scaling does not work with DVIs.

[Ratte]

humppa wrote:

DrBombcrater wrote:
@humppa

Edit: the kernel agruments stuff is a right pain in the bum. I spent hours wondering why the FX driver was giving me a screen full of random garbage on my laptop, it turns out I typed 'video=riva:640x480, flatpanel=1' instead of 'video=riva:640x480,flatpanel=1'.

Just adding one space caused the flatpanel argument to be ignored...

Btw, I didn't know that you can even display GFX card mem and AGP aperture size from Amithlon: "xcat /proc/mtrr".
Don't know if that could be used for anything, e.g. relocating the AGP-area to circumvent the 1GB limit. Bah, I'm talking out of my butt again.  :-P

It would help some people (including me) a lot, if we collect all known arguments, commands, processes together.

xcat /proc/mtrr
xcat /proc/pci
xcat /proc/ide/via
xcat /proc/cpuinfo
...

[Georg]

I spent some time working on that but the GF6 family is just too different internally from the earlier GeForce parts.

AROS NVidia driver is based on XFree86 nv driver sources. Here it works with a 6600 GT. With hw accelerated blits, rectfills, line drawing, color expansion. There's not a whole lot of difference between NV_ARCH_40 and earlier gfx cards. Mostly just in NVLoadStateExt() (which in AROS driver is basically just a copy&paste from XFree86 driver). The AROS driver is at the moment in sync with the XFree86 nv driver from some two years ago. Doing a diff shows that only few things changed since then which if you merge them in would give you support for gfx cards upto 7900.

[DrBombcrater]

@Georg

I've tried grafting NV40-compatible loadstate routines into the rivafb driver (taken from the linux 2.6 nvidiafb driver, not AROS, however) without any success. Something is going wrong early in the initial card setup, and I've so far no idea what.

Are the AROS sources publically available? I've had a look but can't find them.

[Georg]

Are the AROS sources publically available? I've had a look but can't find them.

Yes. http://aros.sourceforge.net/download.php

"Core AROS sources"

arch/common/hidd.nvidida/*

[humppa]

The AROS driver is at the moment in sync with the XFree86 nv driver from some two years ago. Doing a diff shows that only few things changed since then which if you merge them in would give you support for gfx cards upto 7900.

Interesting. Would it also be possible to "unlock" the 1600x1200 modes there? After doing an extensive search through various forums I have come to the conclusions that nobody ever managed to successfully define a 1600x1200 or higher screenmode in Amithlon.
Just like others, I had no luck defining such modes in Picasso96Mode although they were specified according to VESA standards. When I press "test", I only get a white or black screen which my TFT recognizes as "1280x1024" and not "1600x1200". Unfortunately PicassoModeTNG doesn't run and complains about lack of memory (huh? maybe it requires a PIV).

[humppa]

It would help some people (including me) a lot, if we collect all known arguments, commands, processes together.
xcat /proc/mtrrxcat
/proc/pcixcat
/proc/ide/viaxcat
/proc/cpuinfo

Have a look here:

Link

You can find descriptions to these commands on many Linux-sites. Only problem is that they do not always apply to the (relatively old) Linux-kernel that Amithlon is using.

[Ratte]

Shadowolf wrote:
Too bad that it's not that the 6600GT is fried...

My wife went to bed early now and I just could
remove the 5900XT from her PC for a little test:


| Resolution....: 1024 x 768 x 24         |
| Depth/Colors..: 16777216 colors         |
| Testlength....: 13                      |
.============= SPEEDRESULTS ==============.
| GF4200TI                          | GF5900XT
| RectFill()..........    3818 op/s |    4676 op/s
| RectFill() Pattern..     651 op/s |     635 op/s
| WritePixel()........ 1658235 op/s | 1477204 op/s
| WriteChunkyPixels().    5964 op/s |    5956 op/s
| WritePixelArray8()..    5972 op/s |    5960 op/s
| WritePixelLine8()...  122431 op/s |  119790 op/s
| DrawEllipse().......   66897 op/s |   62853 op/s
| DrawCircle()........   66385 op/s |   62325 op/s
| Draw()..............   17519 op/s |   17418 op/s
| Draw() Hor/Ver......   43759 op/s |   55315 op/s
| ScrollRaster() X....     365 op/s |     522 op/s
| ScrollRaster() Y....     368 op/s |     526 op/s
| PutText()...........   28155 op/s |   27965 op/s
| BlitBitMap()........   23365 op/s |   26780 op/s
| BlitBitMapRastPort()   21985 op/s |   25180 op/s
| BitMapScale().......     810 op/s |     823 op/s


Now this really looks good compared to the TI4200.
So the top card to get for Amithlon now should be
some FX5900 based one - and I just got one
for me on EBay.  :-D

And while it still has a fan at least the 5900XT
we already have is a lot less noisy than the TI4200.

Have fun - Shadowolf

Radeon9800Pro 128MB / Kernel 3.0.0t1

I used 640x480 8Bit to compare it with A4000 sample benchmarks and so on ..

[Ratte]

humppa wrote:

It would help some people (including me) a lot, if we collect all known arguments, commands, processes together.
xcat /proc/mtrrxcat
/proc/pcixcat
/proc/ide/viaxcat
/proc/cpuinfo

Have a look here:

Link

You can find descriptions to these commands on many Linux-sites. Only problem is that they do not always apply to the (relatively old) Linux-kernel that Amithlon is using.

Any idea, what these commands are for:

"direct_access.c"

and how to edit it in a usefull way ...?

---

i4fff0000

nb 3=13 % serial
nb 4=13 % serial
nb 7=13 % parallel

nb 1=5 % keyboard
nb 12=5 % mouse
#nb 14=3 % IDE
#nb 15=3 % IDE

# N 5=3 % The CVPPC

# THIS IS FOR THE PCI PERMEDIA2
# mef000000:00020000=ef000000
# mee800000:00800000=e0800000
# mee000000:00800000=e0000000

#
L 00f00000:/newdisk/uaefiles/0xf00000.bin
L 00f10000:/newdisk/uaefiles/0xf10000.bin
L 00f11000:/newdisk/uaefiles/0xf11000.bin
L 00f12000:/newdisk/uaefiles/0xf12000.bin
#L 00f13000:/newdisk/uaefiles/0xf13000.bin
L 00f20000:/newdisk/uaefiles/uaehard.img
#L 00f21000:/newdisk/uaefiles/uaesana.img
L 0xf40000:/newdisk/uaefiles/fat95
L 0xf60000:/newdisk/uaefiles/links
L 00f80000:/newdisk/uaefiles/shape.rom
#L 00f10000:/newdisk/uaefiles/myrom.img

[humppa]

@Ratte

I used 640x480 8Bit to compare it with A4000 sample benchmarks and so on ..

Yeah, but we can't compare with our results anymore. :-P
Please do another benchmark in 1024x768-32

I made some more with my FX5200 and it seems that all FX-models perform about the same. Benchmark scores were very much the same like with Shadowolf's FX5900.
Not really surprising though, the main difference of these cards is the 3D section (completely unused in Amithlon), in 2D they should perform very similar.

[Karlos]

@humppa

Code: [Select]


| GFX system....: Picasso96               |
| Resolution....: 1024 x 768 x 24         |
| Depth/Colors..: 16777216 colors         |
| Testlength....: 13                      |

.============= SPEEDRESULTS ==============.
| RectFill()................    647 op/s  |
| RectFill() Pattern........    651 op/s  |
| WritePixel().............. 2252845 op/s |
| WriteChunkyPixels().......   4937 op/s  |
| WritePixelArray8()........   4936 op/s  |
| WritePixelLine8()......... 171056 op/s  |
| DrawEllipse().............  89452 op/s  |
| DrawCircle()..............  87446 op/s  |
| Draw()....................  23326 op/s  |
| Draw() Hor/Ver............  50647 op/s  |
| ScrollRaster() X..........      2 op/s  |
| ScrollRaster() Y..........      2 op/s  |
| PutText().................  40289 op/s  |
| BlitBitMap()..............    219 op/s  |
| BlitBitMapRastPort()......    218 op/s  |
| BitMapScale().............    196 op/s  |

[/font]
Also odd: why are all the blitter based functions so slow? RectFill()/RectFillPattern()/Blit#?() ought to be insanely fast on a graphics card like that. Even emulated in software, with the massive CPU <-> Display memory bandwidth available to the PC itself I'd expect better than that.

I typically got around 4000-5000 op/s for these functions on my bvision (on the 040/25MHz)

[Ratte]

Athlon 2600+ Radeon9800Pro 128MB
32Bit (24Bit not supported)

Code: [Select]


| GFX system....: Picasso96               |
| Resolution....: 1024 x 768 x 32         |
| Depth/Colors..: 16777216 colors         |
| Testlength....: 13                      |

.============= SPEEDRESULTS ==============.
| RectFill()................  13368 op/s  |
| RectFill() Pattern........    586 op/s  |
| WritePixel().............. 2315206 op/s |
| WriteChunkyPixels().......   5185 op/s  |
| WritePixelArray8()........   5188 op/s  |
| WritePixelLine8()......... 140083 op/s  |
| DrawEllipse().............  51293 op/s  |
| DrawCircle()..............  49933 op/s  |
| Draw()....................  58497 op/s  |
| Draw() Hor/Ver............  83084 op/s  |
| ScrollRaster() X..........   1570 op/s  |
| ScrollRaster() Y..........   1439 op/s  |
| PutText().................  29608 op/s  |
| BlitBitMap()..............    690 op/s  |
| BlitBitMapRastPort()......    688 op/s  |
| BitMapScale().............    485 op/s  |

[/font]

[humppa]

@Ratte

Thanks, some interesting differences between your 9800 and the Geforce FX-series. The speed of several functions differs quite a lot between both. Impossible to say which card is "best" or "fastest" however I guess.

Athlon XP2900+, Nforce2, Geforce FX5200:

Code: [Select]


| RectFill()................   6016 op/s  |
| RectFill() Pattern........    667 op/s  |
| WritePixel().............. 2652861 op/s  |
| WriteChunkyPixels().......   5070 op/s  |
| WritePixelArray8()........   5068 op/s  |
| WritePixelLine8()......... 178357 op/s  |
| DrawEllipse().............  91589 op/s  |
| DrawCircle()..............  95400 op/s  |
| Draw()....................  23485 op/s  |
| Draw() Hor/Ver............  75747 op/s  |
| ScrollRaster() X..........    377 op/s  |
| ScrollRaster() Y..........    380 op/s  |
| PutText().................  40151 op/s  |
| BlitBitMap()..............  24647 op/s  |
| BlitBitMapRastPort()......  22814 op/s  |
| BitMapScale().............    245 op/s  |
|--------------- Intuition ---------------|
| OpenWindow()..............   1692 op/s  |
| MoveWindow()..............   4655 op/s  |
| SizeWindow()..............   2428 op/s  |
| CON-Output................   1083 op/s  |
| ScreenToFront()...........     51 op/s  |