tHog

Nanite

Gentoo Linux

Intro

A silent number cruncher. Those 14-cm Shaman fans at 7 V are truly silent, and the only other moving part is the hard drive. (Stock CPU fan installed on 2013-06-22, a hot summer day, but the fan is quiet at 5 V.)

BIOS/Grub bug -> LILO

As in Hoo, Grub fails to boot but LILO works fine.

Linux drivers

CPU temperature monitorcoretemp, ISA buslm_sensors
CPU undervoltingacpi_cpufreq, phc-intelacpi_cpufreq must be built with the kernel, but not loaded by itself; phc-intel can be installed from Portage
Ethernetr8169
Integrated graphicsi915
IDE/ATAata_piix
SATAAHCIOr emulating IDE if set in BIOS
SoundIntel HDA, Realtek ALC882

Open source and OpenCL on Radeon

As in Hoo.

BIOS settings

This is probably my first motherboard where the BIOS-provided CPU fan control works as advertised. The temperature-driven setting is not quite steady, as it pumps back and forth over a large interval of cooling and heating. The fixed RPM settings are better for my tastes.

I struggled for a while to get PCIe video. At least these Northbridge settings work in BIOS versions 4.9 and 5.0:

priority [PEG/IDG]
internal [Enabled, 8MB]
keep PEG/IDG [Enabled]
PEG port [Disabled]
Indeed, a PCIe GPU only works when it is 'disabled'.

Remote access

This BIOS can be controlled via a serial terminal. After enabling the setting with a monitor and keyboard, it is a very nice feature for remote number crunchers. Besides BIOS settings, you can also access bootloaders; my LILO came up without any extra config, perhaps due to the BIOS access, but some settings may be necessary in others. Linux also provides serial terminal access, after setting it up in /etc/inittab. This is often there already, though commented out, in many distros.

Ethernet problems

There are three Realtek gigabit NICs onboard. My initial setup was to use them as a bridge; not mainly because of actual bridging, but to ensure that I could log in via any port equally, even if their ordering in Linux changes due to udev or something.

Unfortunately, some of the ports fail to initialize occasionally. It seems like a BIOS (v 4.9) or other initialization bug, rather than a hardware problem, because the situation never changes after bootup. The ports that show up as working in Linux remain working. Crucially, the missing ports are equally likely to return to life in the next boot. Moreover, the problematic ones are indicated by a MAC address of FF:FF:FF:FF:FF:FF in the BIOS setup.

This is pretty bad for a headless, bridged system, because the entire bridge fails if even one of the configured ports is missing. There may be options to setup the bridge more automatically, but even that does not help if the cable is plugged in a mute port. My current solution is to network with eth0 only, i.e. the first working port. That way it should not take too many tries, or any software commands, to find a way in.

Western Digital hard drive quirks

idle3 timeout

The WD10TPVT, along with many current WD drives, automatically parks the drive heads outside the platters after a timeout. By default this is 8 seconds. It is not the usual spindown timeout (hdparm -S) though it also does have some power-saving effects.

This is IMHO very useful in a laptop, as it protects against drive head crashes; I recall a similar feature in a Toshiba laptop drive from 1999. However, in a more fixed setting it is somewhat annoying, and can even impact performance and longevity. Strangely enough, it is also found in many 3.5'' drives from WD, probably as a way to market them as "green", even though it introduces needless waste in many situations.

Hdparm has experimental provisions for changing this (option -J), but the safest way is to use a DOS utility from WD. The change is permanent, and in fact requires a power cycle to work at all.

[2012-07-12] There seems to be another dedicated utility for this, but having already fixed the issue under DOS, I have not had the need to test this.

[2013-10-13] Disabled idle3 altogether with the Linux tool, following the problems on Willow.

4-KB physical blocks

Not something to worry about under Linux, but nice to know. Apparently, DOS/XP requires misaligned blocks, but at least this drive is OK by default. In any case, it is a good idea to check that each partition begins at a multiple of 8 sectors.

Nomenclature

Initially, I was planning to use a VIA VB8001-16 board with the Nano CPU. Unfortunately, it has been out of stock at many places. The nano-based name stuck, as this was meant to be a small but fiery performer in a personal "botnet", and nanobots are also known as nanites.


Risto A. Paju