Re: Linux console / Ruby patch on dual PCI Express

[Aivils <ai...@un...>]

On Monday 29 November 2004 20:33, Zoltan Boszormenyi wrote:
> Magnus Lundstedt =EDrta:
> > This is my first post to this list, I hope everyting works. :)=20
> >=20
> > I am using kernel 2.6.* with ruby on my home computer since december la=
st=20
> > year. I have two nvidia graphicscard - TNT2 Ultra PCI and Geforce4 AGP.=
=20
> >=20
> > Everything works as it should and without problems, now I have a few=20
> > questions:=20
> >=20
> > 1) When will this patch go into the main linux tree? What is left to do=
 before=20
> > it can merge, or is that "out of the question" / not gonna happen in th=
e=20
> > forseable future?=20
>=20
> As I heard, it won't get into the kernel in this form.
> Instead, one of the development directions of Linux-2.7.x
> will be to remove the console system from the kernel and
> create a user space console system.=20

 i have read similar rumors sice 1999 :)
James Simons have political influence. He should know.

> This way, all the past,=20
> present and future races will be removed as well, that plagued
> the development of 2.[56].x and were only solved in late
> 2.6.x.
> > 2) Several vendors (via, nvidia, etc) will soon start selling mobo-chip=
sets=20
> > capable of dual PCI-express or AGP + PCI-express slots.=20
> >=20
> > First (a); will PCI-express cards work "just like" AGP cards in general=
 in=20
> > linux.=20
>=20
> They should.
>=20
> > Second (b); will PCI-express motherboards/graphicscards work "out of th=
e box"=20
> > with linux?=20
>=20
> They should.
>=20
> > Third (c); Does Linux support multiple AGP ports today, which the AGP-s=
tandard=20
> > does.=20
>=20
> The kernel should support it, as all the PCI-like devices are discovered
> through the PCI BIOS calls. But I am yet to hear about such a mainboard
> that implements the "sidecar" AGP that is part of the AGP 3.0 standard.
>=20
> > If it does, will it be possible to use one of these dual PCI-express bo=
ards=20
> > with the ruby patch to make a system with two X-consoles where each has=
 one=20
> > PCI-express graphics card?
>=20
> The answer is "definitely, yes" here. :-)

This is device driver specific feature. Since Nvidia print in their README
"improved PCI Express" i hope about full success as before.
If Xfree86, Xorg do not put PCI Express into drivers, then we can dream onl=
y.

> > A system like that would be very intresting, since I am today having pr=
oblems=20
> > with openGL and other accelerated graphics on the PCI nvidia card. (whi=
ch I=20
> > assume is due to the poor performance of the PCI bus, I only get ca 700=
 fps=20
> > in glxgears on the PCI card, and xv-accelerated video has lower fps (so=
me=20
> > jerkiness) than xv on the AGP graphicscard).=20
>=20
> I had the same problem, like one of our daughters played SuperTux on the
> X server that drove my PCI Radeon and it hogged even IDE transfers.
> I reniced the X server (renice 10 <pid of X>), and it almost solved the
> problem. I say "almost" because now the SuperTux lowered the framerate,
> pretty much visibly. I haven't experimented more.

PCI is slower by default. In my end glxgears schows similar reasults.
Unloaded box - PCI is 2x slower. mplayer takes 2x more CPU on PCI.

Troble hides in the 2.6.X task scheduler which one do not care how evenly
cpu is distributed over a very short timespace, and yes, 2.6 does not do=20
this well. By default 2.6 uses 100msec timeslices for each process. If You
have 2 tasks, then each have 5 moves per second. This case is very
ugly for video games.
As solution is nice +19 usage. In nice +19 case task schedule gives only
10msec for task as maximal timeslice. So if You start
user1>nice -n +19 game1
user2>nice -n +19 game2
both games run smooth, but never as under 2.4.XX . i cannot explain
technical essence, but this is gamer feeling. 2.6.XX bump off WineX,
which becames slower, choppy.

> > So, in summare, my second question is; Does anyone know if the future d=
ual=20
> > PCI-express (or PCI-express + AGP port) motherboards (such as the VIA=20
> > K8T890Pro or nvidia Nforce 4 SLI) will work with linux and the ruby pat=
ch in=20
> > particular to set up a multi-console system using fast gfx cards?=20
>=20
> They should work.

Who knows. For AGP we have hardcoded code in the kernel for each mobo vendo=
r.
In the Nvidia case this code seems exists in the binary driver, i hope so
because Nvidia anounce PCI Express support (which mobo? reference board?).

Aivils