[NTLUG:Discuss] Re: pci-e and linux ( and radeon -vs- nvidia )

[Bryan J. Smith]

On Wed, 2004-12-22 at 06:20, Jack Snodgrass wrote:
> I'm thinking of building a new PC using a ASUS motherboard with PCI-E
> and lga775 cpu.

Any reason you wouldn't consider AMD?  As I noted in my other post, the
nForce4 is now shipping in a couple of mainboards.

> Will current versions of the linux kernel have any issues with this?

>From what I've seen -- just like SATA is to ATA, PCIe is to PCI --  it's
a serial channel that appears as a logical parallel implementation of
the same.  So PCIe channels just look like bridged PCI busses.  Don't
quote me on this, but the support is clearly there.

So, as always, only the actual peripheral logic becomes the issue. 
E.g., if the ATA, NIC, etc... logic changes

> I'll have to get a PCI-E Graphics card... do I want a radeon or nvidia
> card... as far as linux drivers are concerned. 

First off, see these posts:  

"PCI-Express now seems to be the price/performance mainstay for Linux"
http://www.matrixlist.com/pipermail/leaplist/2004-December/000117.html

[ Unfortunately, it doesn't appear that the nVidia GeForce 6200 is
officially supported with the latest 1.0-6629 drivers, yet. ]

"NV2x (GF3/4Ti) v. NV3x (GFFX) v. NV4x (GF6K) on PCIe v. AGP ..."
http://www.matrixlist.com/pipermail/leaplist/2004-December/000171.html
http://www.matrixlist.com/pipermail/leaplist/2004-December/000172.html
http://www.matrixlist.com/pipermail/leaplist/2004-December/000173.html
http://www.matrixlist.com/pipermail/leaplist/2004-December/000188.html

ATI is working hard to "catch up" right now.  They should be commended
for their past XFree/Xorg efforts, but reality finally "sunk in."  I
don't know how extensive the card support is yet, but it's not the
entire line.  And they do _not_ support Linux/x86-64 yet.

nVidia has been supporting Linux for a long time, both with paid
developers on the 2D/overlay-only Freedomware "nv" drivers as well as
the 3D/GLX Standardware "nvidia" drivers.  The latter definitely
includes both AGP and PCIe.  In fact, since AGP is an "Intel trade
secret" whereas PCIe is a "PCI Standards Group" standard, PCIe is
actually "more open."  E.g., No more AGPgart crap -- the Linux kernel
implementations _suck_, and for AGP cards, I _always_ used nVidia's
driver integrated support instead.

You can _always_ get the _exact_ card support by model by looking in
"Appendix A" of the README for the latest nVidia Standardware "nvidia"
driver.  

For IA-32 (x86) 1.0-6629:  
ftp://download.nvidia.com/XFree86/Linux-x86/1.0-6629/README.txt

For AMD64 (x86-64) 1.0-6629:  
ftp://download.nvidia.com/XFree86/Linux-x86_64/1.0-6629/README.txt  

Looks like the nVidia GeForce 6200 PCIe isn't supported yet.  Right now
that's the "best bang for the buck" at around $100 for the 128MB
version.  It typically _beats_ even the "highest end" GeForce FX5900
series.  And since nVidia's current "AGP cash cows" are the GeForce
FX5200, FX5500 and FX5700"LE, they won't release an AGP version of the
6200.

The FX5200 and FX5500 can't beat even an old GeForce4 Ti4200 (unless you
are running Doom3 and jacked up all the settings so you get 9fps with
either).  The FX5700"LE" (a massively underclocked FX5700) loses to the
GeForce4 Ti4200 until you use DirectX 9 or Doom3-like OpenGL games and
then it varies on settings.  _All_ still lose to the 6200 though.

But be _wary_ of the new, _cheaper_ GeForce 6200 with "TurboCache." 
What is it?  It's a card that has only 16MB, 32MB or 64MB of _local_
SDRAM.  It uses the increased DTR/priority of the PCIe channels to use
main memory for additional buffer/texturing.  It's cheaper, but for
gamers, it's not worth the $10-40 savings over a _true_ 128MB 6200 card.

-- Bryan

P.S.  Again, I haven't checked to see if/how SLI works under Linux with
GLX, if at all.  SLI is different this time around, but with GLX, I
think it will be supported because it's easier to do than with DirectX
(which is not so tight on specifications).  But that's just assumptions.

-- 
Bryan J. Smith                                    b.j.smith at ieee.org 
-------------------------------------------------------------------- 
Subtotal Cost of Ownership (SCO) for Windows being less than Linux
Total Cost of Ownership (TCO) assumes experts for the former, costly
retraining for the latter, omitted "software assurance" costs in 
compatible desktop OS/apps for the former, no free/legacy reuse for
latter, and no basic security, patch or downtime comparison at all.