[Pc_Support] It's here!

[Austin Denyer (Ozz)]

On Sat, 05 Nov 2005 10:41:59 -0600, "Bryan J. Smith"
<b.j.smith at ieee.org> wrote:

> On Fri, 2005-11-04 at 21:22 -0500, Austin Denyer wrote:
> > Hi Guys.
> > The nice FedEx man delivered my new workstation today.
> > Tyan K8WE MoBo, Dual-Opteron246 CPU, 2GB RAM, 2x200GB SATAII HDD, EVGA
> > GeForce 6600GT, Thermaltake XaserIII V1000A case.
> > It ROCKS!!!!!
> 
> Glad to hear!
> 
> Remember to update to the latest BIOS on those mainboards.  There was
> some issues a few months back with Cool'n Quiet reliability under Linux
> (and Windows Server 2003) until a BIOS update.

It has the latest BIOS.

> > I used a modified version of the Debian AMD64 netinstall CD, as the
> > standard one used the 2.6.8-2 kernel, which has limited hardware
> > detection.  This had the 2.6.12 kernel (thanks to Len Sorensen
> > over on the Debian-AMD64 list), and installed fine with the exception of
> > the network drivers - it only found/loaded drivers for the firewire
> > networking, which I don't use.  By continuing with the install and
> > rebooting, the forcedeth drivers loaded fine.
> 
> The forcedeth drivers seem to be good for GbE performance now.  As
> someone corrected me on the Fedora development list about 3-4 months
> ago, nVidia's legal shackles have been off for the last 12+ months.
> While they still offer the closed source "nvnet," the have had active
> people on the GPL "forcedeth" driver since 2004, including full
> disclosure.

I like them.
 
> > A quick hack of the
> > network config files and I was up and running. I then just upgraded to
> > the 2.6.14-2-amd64-k8-smp kernel and I was good to go.  (There is a bug
> > in the Opteron CPU that causes issues when running in SMP mode, and the
> > work-around was only just included in that kernel.)
> 
> Which bug was that?  I know there many errata coming out all-the-time
> with the Opteron's compatibility modes (i.e., so the Linux kernel can
> treat it like a "front-side bottleneck" processor, instead of its true
> NUMA/multi-point calling).

See my other e-mail for links.
 
> > Two gotchas.  
> > 1. When I installed X/KDE it swapped my NICs around (?!?!?).
> 
> Interesting.  The PCI bus order is defined by the APIC/ACPI, typically
> at POST.  You're going to experience the greatest number of busses
> you've ever seen -- the sucker has 3 different HyperTransport tunnels,
> the nForce Pro 2200 and 2050 (20 PCIe channels _each_) and the AMD8131
> (2 PCI-X channels).

Yep - I can't explain it either.
 
> > 2. I haven't managed to get the nVidia video drivers working yet.  The
> > official nVidia drivers will not load on X startup (although they did
> > appear to compile OK, and modprobe doesn't barf on them), and the nv
> > driver causes display corruption after a few mins. I'm running the vesa
> > driver right now without issue.
> 
> Is the new Debian using udev?  If so, that's the issue.  The devices
> need to be created.

To be honest, I'm not sure.  I'll check on Monday.
 
> What do you see in /proc/driver/?  If the nVidia driver loads
> successfully, then you will see it there.  Cat those files if they
> exist.  In that case, it's definitely the lack of udev devices, which
> you'll need to create.
> 
> As far as corruption on the "nv" driver, what Xorg version?  If it's an
> older version, then yeah, that's your problem.  Although the older Xorg
> release for NV3x (FX) will still drive NV4x (6xxx) series cards, you're
> going to have issues.

Again, without checking the box again, I'm not sure.  It's whatever was
with Debian Sid as of Friday...
 
> > As far as (2) goes, I've heard that the nVidia drivers don't yet work
> > with kernels > 2.6.12, but I need the 2.6.14-2 kernel for SMP.
> > Can anyone confirm/deny/provide pointers for a workaround?
> 
> Interesting.  Do you have links to this information?  Frankly, I think
> nVidia is lax in getting the latest Forceware 75 drivers over to Linux,
> probably because they have been so focused on Forceware 80 for Windows.

Again, links in my other e-mail.
 
> > Anyhoo, the thing FLIES!
> > OpenOffice2 (run from a 32-bit chroot) loads in 3 secs.
> > The Gimp loads in under 1 sec.
> > I've only carved up one disk so far - I'm considering going to RAID-1
> > at some stage.
> > I've tacked a few stats at the end to get the juices going...
> > Many, MANY thanks to all who provided input and convinced me to buy
> > this kit.
> > dev05:/home/adenyer# cat /proc/cpuinfo
> > model name      : AMD Opteron(tm) Processor 246
> > cache size      : 1024 KB
> 
> The design of the Opteron at 2.0GHz with 1MiB L2 (and 64+64KiB L1) is
> like having a 4.0GHz Pentium 4 with 1MiB L2 (and it's measily 16KiB +
> 8Kops L1).

Sure makes Intel look silly.
 
> > flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge
> > mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext lm
> > 3dnowext 3dnow
> 
> No SSE3 for programs that Intel has tricked people into such lossy math.
> Oh well, it looks like the older Rev. D, but that's not really an issue
> in Linux.
> 
> > fdisk /dev/sda
> > Disk /dev/sda: 200.0 GB, 200049647616 bytes
> > 255 heads, 63 sectors/track, 24321 cylinders
> > Units = cylinders of 16065 * 512 = 8225280 bytes
> 
> Yep, the first "nv_sata" device.
> 
> > fdisk /dev/hdb
>              ^^^ typo?

Yep - my bad, should be /dev/sdb

> > Disk /dev/sdb: 200.0 GB, 200049647616 bytes
> > 255 heads, 63 sectors/track, 24321 cylinders
> > Units = cylinders of 16065 * 512 = 8225280 bytes
> > top - 16:49:15 up 56 min,  2 users,  load average: 0.82, 0.70, 0.49
> > Tasks:  97 total,   2 running,  95 sleeping,   0 stopped,   0 zombie
> > Cpu(s): 13.5% us,  6.7% sy,  0.0% ni, 69.1% id,  7.8% wa,  1.2% hi,
> > 1.7% si
> > Mem:   2059056k total,  2047008k used,    12048k free,     4096k buffers
> > Swap:  3903752k total,        0k used,  3903752k free,  1786264k cached
> 
> No games above 1GiB, you have full access to the entire 40-bit space of
> the EV6 address model.

I really do love the machine.  

Regards,
Ozz.