I need that this information came to Rudolf's attention (nv)

1

Who am I to tell a think to him, but during his last testing, he achieved these results:

Card under test: GLteapot @ 16bit: GLteapot @ 32bit: Quake2 @ 16bit: Quake2 @ 32bit:
TNT2, original, 32Mb (NV05) 195-205 fps 175-185 fps 40.4 fps 31.4 fps
TNT2 M64, 32Mb (NV05M64) 175-185 fps 135-145 fps 30.9 fps 20.7 fps
GeForce2 MX400, 32Mb (NV11) 200-220 fps 190-210 fps 38.2 fps 25.5 fps
GeForce4 MX440, 64Mb (NV18) 90-100 fps 90-100 fps 10.5 fps 10.4 fps
GeForce4 MX4000, 128Mb (NV18) 90-100 fps 90-95 fps 10.3 fps 10.2 fps

from his 3d weblog that I use as my main page:
http://web.inter.nl.net/users/be-hold/BeOS/NVdriver/3dnews.html

You can see the results going down with the GeForce4 series but IIRC the inner workings of the GeForce 4 and up were totally revamped and that’s why the speeds are going down, so the driver needs to be changed, I mean, the UtahGLX will not work correctly with those new boards. He will only get a progression in speed until the GeForce3 series, then the driver could not be used “correctly” (read “efficiently”) since this internal changing. In practical terms, we all see 3D acceleration is working but something is not OK (at least not OK for GeForce4 and up).

I own these video boards:
GeForce2 MX-400 PCI 64MB SDR with TV-Out (S-Video) from Palit
GeForce4 MX-440 PCI 64MB SDR with TV-Out (S-Video+Composite) from Palit (sold about 4 months ago)
and a GeForceFX 5200 PCI 128MB DDR with TV-Out (S-Video) and DVI from Prolink.

I can assure to you all that the GeForce4 mentioned is about 125% faster than the GeForce2 (I don’t mean memory/processor speeds, I mean rendering power) and way more efficient in terms of heat dissipation (The GeForce2 has a cooler while the GeForce4 has only a termal dissipator). I own these cards for a long time and tested them over and over again. I remember when the GeForce4 series launched you could find something at nVidia’s web site commenting about this.

I will look around to see if I can find more information to backup me up, but I’m really really sure that it is the why those cards are performing badly with his driver. Probably he may be aware of this but, I’ll try to warn anyway.

Hope this come to his attention and please mind you that it’s just my intention to help it anyway I can: I’m not try to put sand on his excellent and hard work!

–
Doca

2

Rudolf reads through the comments posted at iscomputeron, I added a comment with a link to this post so he will probably read through it.

3
tb100 wrote:
Rudolf reads through the comments posted at iscomputeron, I added a comment with a link to this post so he will probably read through it.

Thanks a lot, tb100.

I still haven’t found more info about it. I’ll search today (April 13th) by the noon.

4

To keep the threads in sync, I’ll add Rudolf’s reply at ICO here too:

Hi,

Sorry I have to reply here, but time is limited :-/

Anyway: if people can come up with benchmarking results from similar cards on Windows, and/or nVidia techinfo indicating what we should see as behaviour for speeds, I am always interested for pointers :slight_smile:

Rest assured I am not taking this ‘lightly’ I’ll see what I can do later on (it’s not top prio ATM)
About ‘new’ cards: make sure to distinquish between the GF4MX and other GF4 types: these are totally different GPU’s!!

And: I think GF3, GF4-non-MX, and newer cards won’t work at all currently: indeed: lack of specs…

Thanks for thinking along!!

Talk to you later :slight_smile:

Rudolf.

So it’s a lack-of-specs issue again. I hope it’s one that Rudolf can resolve.

5

Hi,

Thanks for sharing your concerns. I am a bit puzzled myself as well: maybe I can find out more, maybe I can’t: we’ll just have to wait and see.

Anyhow: here are some thoughts:
->Please make sure you realize that there are two entirely different types of GF4:
GF4MX and all others (Ti…)
The MX types are NV17/NV18, while the rest is NV25/NV28.
Now, also bear in mind that the GF3 is NV20, and GF2Ti is NV15.

->The MX type of cards are ‘getting started’ cards, while the rest of them are ‘the real thing’. It’s only logical that a MX type card is slower than a non-MX type card as I see it. The TNT2 I tested was designed to be ‘top-notch’, while the GF4MX are ‘getting started’ cards: maybe that could explain it.

->Of course, something else may be the problem: poorly implemented ‘old’ commmands, slow FIFO access, slow MEM and CORE clock by default (maybe I have to reprogram them upon 3D actication). When you think about it, there may be even more options here.

->About NV20 and higher cards:
If I remember correctly, since GF3 nVidia has programmable matrix-transformations stuff, while before it was hardcoded for much-used setups. It might well be, that I won;t be abe to get NV20 and higher going because of just that fact: combined with the fact that there’s no known info about this new feature.
OTOH: Those cards might well default to ‘the old way’ if you don’t touch that new feature. If so, there’s probably some 3D pre-init stuff not correct/missing currently.

I’ll put up more benchmarks in an hour or so (blog), but not for:
->GeForce4Ti4200 (NV28)
->GeForce FX5200 Go (NV34 laptops)
->GeForce FX5200 (NV34)

Those cards I tested: and they all crash immediately: 3D does not work at all there (apart from the back to front buffer blitting as that’s a 2D function in fact).

Rest assured I’ll do my best to get as much up and running, and in the fastest possible way. But you have to be prepared for the possibility that you will need to use an old card if you want acceleration on nVidia. Unless they finally spill specs of course :wink:

BTW: It’ getting harder and harder to follow up on the messages I am getting: sorry if my responses are sometimes late… :-/

But I want that driver (alpha) out the door asap too :slight_smile:
Note that the first release will probably work on the cards as described, in PIO mode, using Mesa 3.4.2.
After that I’ll try to go to Mesa 6.2 and do DMA…

Rudolf.

6

Hi,

You’ll be very relieved to hear that I could fix 3D rendering ion NNV18 just now. In DMA mode quake2 now runs at 91.5fps in 640x480x16. Teapot is at 400fps in 16 amd 32bit deepth. This is the currently fasterst supported card now (runs at 120% speed comared to NV15/GeForce2Ti).

Hope this info helps :wink:

Rudolf…

7
rudolfc wrote:
Hi,

You’ll be very relieved to hear that I could fix 3D rendering ion NNV18 just now. In DMA mode quake2 now runs at 91.5fps in 640x480x16. Teapot is at 400fps in 16 amd 32bit deepth. This is the currently fasterst supported card now (runs at 120% speed comared to NV15/GeForce2Ti).

Hope this info helps :wink:

Rudolf…

I was just reading HaikuNews and I was coming here to thank you.

:smiley: Thanks a lot! I think that the entire community is grateful.

By the way, the problem was “just” the way the board is initialized? If so, then I was mistaken. The “inner workings” of the GPU still the same but the way the board is initialized is what should be changed… (I imagine that it is not that simple, but is something like it?)

And thanks again! You are doing a very nice job.

Tell us if we can assist you in any way.

–
Doca

8

Hi,

Thanks for the compliment :slight_smile:

Anyway, I just had to reset the engine by setting all registers (known and unknwown) to zero. That did the trick…

I’m curious what newer cards will need: that’s definately more complicated: pipeline init /pgm stuff or so. Unknown of course :-/

Going to try something anyway…

Rudolf.