OT: nVidia driver [was: Wish list]

Thu Jun 9 15:45:26 UTC 2005

From: fedora-devel-list-request at redhat.com
> Actually you are tied to nVidia.  Every time a significant kernel change
> comes out you have to rely on them to produce a new driver for you.

This has _yet_ to happen.  At the most, there was a stack or other
configuration difference not in the stock kernel that required a quick
rebuild, nothing difficult to accommodate.

In fact, you complain about the "zealots" who blindly recommend
nVidia's drivers.  Well, as someone who is not blind to _both_ sides,
I tell you I tire of "zealots" who _overstate_ the "quality" of Freedomware
DRI/GLX solutions versus nVidia.

Yes, I love a solution like an integrated GPU in a chipset that comes
ready-to-use, out-of-the-box.  I love that Fedora Core 3 installed on my
i855GM system and immediately gave me GLX on first boot.  I will
commend all parties involved on that.

But that solution is _not_ going to solve _real_ engineering needs.
Not now, and not likely in the future.

> If they tire of that exercise you are SOL or you'll have to go buy a
> different piece of hardware.

My mitigating factors are the reality that nVidia will not because of the
industry that deploys their solutions, and the continued, strong marketshare
as a result.  This is a _far_better_case_ for risk mitigation that some
_blind_faith_ in ideals.   Just because something is Freedomware does
_not_ guarantee it will be supported.

Many video cards and vendors have either gone by the way-side, or "closed
up."  One only needs to look at the number of drivers that have gone
unmaintained in the 2.6 kernel.

REALITY #1:  As long as open standards are adhered to, sometimes risk
is more about popularity when it comes to maintanence than philosophy.
And even if something goes unmaintained, by adhering to open standards,
it's a matter of changing the intermedia solution, such as hardware.

REALITY #2:  As long as I stick to GLX applications, I have my _choice_ of
hardware with Freedomware or Standardware drivers.

BIGGEST REALITY:  The cost and economies-of-scale of nVidia solutions
are the best "bang-for-the-buck" and if they are not supported some 1-2
years down the road, it will still cost me _less_ than buying an eccentric,
costly hardware solution, or relying on a poorly supported solution _today_.

I'm sorry, but in my industry, nVidia provides _excellent_ support.  Yes,
the new 1.0-7xxx series drivers are flaky, but just like some of the other
series when they first came out.  I've been sticking with the last revs of
the 1.0-6xxx drivers without issue, including supporting some of the latest
products.

And I've _never_ had a kernel incompatibility, sans only one case where
a rebuild of the kernel was required.  But I've _never_ been without working
drivers.

> Why not just start out by buying a different piece of hardware that _is_
> supported by open source and reduce the risk?

Because the risk is _not_ reduced!

They their Linux drivers are often:  
1)  Less reliable
2)  Far lower performing
3)  Far less supported by the vendor
4)  Often lag product release, heavily

> Sure, and there are open source solutions that accomplish the same thing. 
> While nVidia may still enjoy a slight edge in performance and features

Okay, get off this.  Sorry, this is a _bogus_ assumption.
I'm not talking about gaming here either!

> there are _very few_ situations that actually demand the extra capacity
> provided.

But those "very few" situations are a _major_ sustainment of nVidia's
Linux driver efforts.  Not only for other companies, but do you know
how much of nVidia itself is Linux???  It's not just a majority, but a
massive and overwhelming majority.  Linux drivers are a staple to
their own, internal IT need.

> So, if you want all the advantages provided by open source
> software the best choice is to avoid any solution like the one you're
> still promoting.

By "advantage," you are typically talking about "open" v. "proprietary"
in a blanket statement -- proprietary APIs, data in proprietary formats,
etc...  The problem is that it isn't that simple because we are not talking
about "proprietary standards."

In the case of nVidia, the _only_ thing you are risking is that your hardware
investments will not be supported in the future.  Given the cost of nVidia's
typical solution, it is chump change to lose compared to going with more
costly hardware and/or 3rd party drivers.

> No, that's a false analogy.  There are _real_ risks when running binary
> only modules in kernel-mode.   Those same risks don't come into play with
> binary only user applications.   That's a big difference.

The reality is that that are real _needs_ to address issues at the kernel level
that cannot be done in a user-space driver, especially memory control.  This
is not some "infestation" of user-space detail into the kernel, but things that
should probably be controlled by the kernel in the first place.

Now I'm not saying it wouldn't be nice if nVidia GPL'd it.  I would love to see
that.  But I just know it's not legally possible at this time, at least not all
of it.  I'm not "praising" nVidia for it, but understand that as long as people
are aware of the risks, then it's still using "open standards."

Unfortunately, you can't see to see the "other risk" we have in not deploying
nVidia as a solution.  See #1-4 above.

> Not to mention there is a very real risk of you losing support for your
> beloved hardware.

But the _cost_ of using something else is _much_greater_!  So what nVidia
provides me is _value_ in a Standardware product, that does _not_ expense
open standards, _today_.  Not promises, not drivers for obsoleted products
with various product issues and quirks -- _real_, _working_ hardware.

Probably the biggest reality is the fact that nVidia is one of the _largest_
users of EDA on Linux, using their own GLX solution.  They could _easily_
persuade the EDA industry to start supporting only their chipsets, but they
are dedicated to OpenGL.  Maybe not as strict to the ARB that 3DLabs
wants, but fairly good.

So, once again, who is going to give me "less risk":  

A)  A Standardware vendor who is not only the producer of products and
drivers that adhere to open standards, but is also a _user_ in the same
industry as I?

B)  Freedomware--er, excuse me--Commuware zealots that do not take
the time to understand the _real_ risks of going with _less_supported_
solutions.

There's quite a bit of "Hostageware hardware" out there now, because
no one's maintaining the hardware.  With nVidia, I know that they are
going to continuing supporting Linux, at least in the near-term, because
they need it for themselves!

Especially for a product that is obsoleted in 12 months anyway, and
rarely used beyond 3-4 years.

> Again, a problem that doesn't exist in your analogy.
> Sorry, you are in that category if you fail to recognize that the vast
> majority of the time there are perfectly viable fully open source
> solutions.  That's fine if you want to pursue them yourself, just don't
> come here and recommend the practice for others, it's bad advice.

I use nVu, and have since version 0.2.  But I also use Macromedia
Dreamweaver.  If I lay out the risks involved, and let others decide
for themselves, then I'm going to be damned on what I can and can't
say.  Choice is the key.

Furthermore, I didn't even start this thread!  My problem is that you
claim there are these "zealots" out there and what I'm telling you is
that you're doing the same, damn thing!

Especially given the fact that you are so oblivious to thing that no
"Proprietary" vendor could possibly reduce risk over an "Open" vendor.
Sorry, there's a good example in widespread use by doctors and law
firms for decades, Corel.  And there are many others.

Risk is not about zealousy.  It is about careful analysis of the risks
involved.  But deploying nVidia, I not only solve my needs, but I
stick with POSIX/GLX applications on _Linux_.  Without nVidia, many
solutions would have gone to Win32/DX, never to return.  And even
if nVidia drops support -- which I've gone at length here to show
they will not for their own, internal sake -- then I merely switch to
another GLX solution.  Given the increased cost of those solutions
in price/performance, I'm already getting the nVidia solution for
chump change, and benefit from their economies-of-scale.

> Supporting open source as a preferred platform is no more ideological
> than the arguments you're putting forth.

The problem is that "open source" is _not_ always the "preferred platform."
People take some concepts that work well against "proprietary standards"
and then assume they still have the same argument against "proprietary
source" but "open standards."  The game changes there a bit.  ;->

If a Standardware solution offers far greater feasibility and _reduced_
risk than a Freedomware solution, then it should be considered on those
merits, even if only limitedly or for specification applications where they
apply.

> Again, there are perfectly viable fully open source solutions today for
> the vast majority of uses.

Yes.  And in many cases, I don't need GLX, so I use the "nv", "radeon"
or whatever, Freedomware MIT driver comes with XFree/Xorg.

And I have deployed UtahGLX, DRI and even Xi and other, partially
open/proprietary source solutions.

> There is nothing wrong with people promoting them over binary only
> solutions.  Period.

Except when they introduce more risk.

> That's fine if you want to pursue them yourself, just don't
> come here and recommend the practice for others, it's bad advice.

That's the problem -- I did _no_such_thing_!
I merely tried to explain why so-called "zealots" aren't understood
by opposing "zealots" like yourself.

--
Bryan J. Smith   mailto:b.j.smith at ieee.org