> Geert Uytterhoeven writes:
> > On Sun, 16 Aug 1998, Albert D. Cahalan wrote:
>
> >> He told us all to get better hardware (obviously the Amiga) or just
> >> live in hell with our PC hardware.
> >
> > I don't think Jes wants you to buy an Amiga.
>
> The Mac is too expensive and the Atari is dead. Most of the world
> isn't rich you see, and would like to have computers anyway.
And when you are using older hardware, it becomes doubly important
to make optimal use of the hardware. Linux is famous for resurrecting old
386s and 486s and putting them to good use. Why should video cards be
different?
> Is stable modern video restricted to those who can afford an SGI?
> If so, it is not for any technical reason.
Well, the technical reason is that it costs more to design
hardware properly, and in the cutthroat PC video card market that is a
luxury you can't afford.
> >> The i740 can issue multiple requests for texture data stored in main memory.
> >> Don't think "WinVideo" here, since the chip uses several MB of local
> >> memory while simultaneously fetching from main memory at 533 MB/s.
> >> On a good machine, that could mean hundreds of MB of textures.
> >
> > Please show us the programming specs for the i740 first (without NDA!)
> > so we can write a driver for it :-)
> >
> > It's nice to have such a video chip if you can't even write a driver for it.
>
> I know. Consider it an example of modern video hardware, nothing more.
> Matrox has something similar AFAIK, and I'd expect we get info after
> they feel it won't help the competition too much.
Or we let them use binary drivers.
> Perhaps you can explain how a userspace solution will drive a chip
> like the i740. If you prefer, think "Cyrix CPU with built-in video"
> and assume we have documentation. Don't forget the chipset-initiated
> DMA, perhaps including use of the i386 page tables and an IRQ.
What about the embedded multimedia set-top boxes of the future?
I'd like Linux to be useable in embedded systems like that, and a flaky
userspace video subsystem will not fly in that type of situation.
> >>> (I'm an x86 user, by the way, so your ultimate argument doesn't apply.
> >>
> >> No, you have different reasons than he does. You are willing to give up
> >> stability for a tiny bit of speed on low-end hardware. He has an Amiga.
> >> The only thing you have in common is a "screw you" attitude towards
> >> people who want stable PC video and/or want to use advanced features.
> >
> > People who want a `stable' system should buy `stable' hardware.
>
> With OS protection, most junk is stable. It is much cheaper to
> use a bit of kernel memory than it is to buy proper hardware.
> So far, only one example of proper hardware has been given,
> and I can't find it at www.pricewatch.com.
I think I remember that some Cirrus Logic chipsets had some type
of priv locking and some type of context management, but I dunno for sure.
For damn sure 3Dfx Voodoo doesn't.
> I don't think I could buy a well-behaved modern video board at any price,
> though I could put the whole damn X server in RAM for only $14.41. :-)
RMA is so cheap and CPUs are so powerful these days that the
"ludicrous" notion of putting the whole X server in the kernel is not
actually something that would be that big of a deal. I personally
wouldn't want to do it, but it *could* be done as a module without
bloating the kernel anywhere close to the level of commercial Unix
kernels. Do it right and performance won't be a problem either.
> KGI gives poverty-stricken users good video for less than a dime.
> Compare that to the cost of an SGI or the slowness of a dumb framebuffer.
Precisely. KGI *works*. It find the maximal balance of speed and
safety. If you want to give up the safety for som more speed, use
suidkgi. But don't try to tell the rest of us that want stable systems
that *we* have to be the exception to the rule. Stability should ALWAYS
be the default.
Jon
--- 'Cloning and the reprogramming of DNA is the first serious step in becoming one with God.' - Scientist G. Richard Seed
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html