I would like to tender my 2 cents, as I've just gone thru
very similar experience to make an "always on" linux machine
at home.
I started with an Asus A7N8X-VM board + Athlon XP 2600+.
Price of the main board is around USD100+ and CPU is around
USD90 here (Hong Kong). CPU performance is very good, as
the 128kB+512kB cache on the Athlon (Barton) core is very
effective. But the Asus board gave me a lot of headache,
and the main board had been RMA once. The board is very
picky on RAM used, if DDR400 RAM is used, the onboard VGA
(from NForce2-IGP chip) has to be disabled, and there is no
way in the BIOS that one can "downgrade" a DDR400 to be used
as a DDR333 RAM. Only SPD is supported on the Asus board.
In order to install linux (I've tried RH9.0 / Fedora Core 1
/ Fedora Core 2 Test 3), the onboard LAN (also from NForce2
chip) can't be detected, due to no GPL driver available,
though NVidia does have a driver. My quick-n-dirty way to
"bootstrap" the board is to insert a NVidia GeForce FX5200
AGP card and another 3Com 3C905 LAN card into the computer
and disable the onboard VGA/LAN! I've run
llcbench/cachebench on the setup, and RAM bandwidth is like
980MB/s
I think, maybe the Asus is very conservative, so, I go out
and buy another NForce2 based mainboard, which is a
Leadtek/Winfast K8NCR18DLM at USD59. The Winfast board uses
NForce2-400 chipset, so, similar LAN, but without onboard
VGA. Only when after I bought the board and find out, the
board doesn't support Dual-channel DDR RAM! But the board
does allow tunning of RAM speed parameter. One interesting
finding is that the llcbench/cachebench performance is
actually better with DDR333 setting (on a pair of DDR400
RAM) compare with running at DDR400. Net wisdom suggests
that sync operation (FSB@333 / RAM@333) will give a better
performance compare with async operation. To my surprise,
dual-channel operation on the Asus board only gives a < 10%
performance edge (980MB/s vs 910MB/s) compare with the
non-dual-channel Winfast board.
I've also searched the market for "dirt-cheap" CPU price,
and found that there are 1.6GHz and 1.8GHz Duron, which they
are not marked by performance figure but true working speed.
An Athlon XP 2200+ actually runs at 1.8GHz and an Athlon
XP 2000+ runs at 1.6GHz. However, the on-chip cache on
Duron is a mere 128kB+64kB (L1+L2) versus 128kB+256kB on
XP2000+/2200+. Price of a Duron 1.8GHz is around USD42
here, but the cheapest "tray" version of Athlon XP2000+ is
at USD46! All price is without HSF, so, I buy a USD5
aluminum HSF together with another Athlon XP2000+. On
benchmark the Athlon XP2600+ is much faster than Athlon
XP2000+, but on my everyday use, I can't tell the speed
difference!
Well, towards the end part of my story. I'm not comfortable
with the additional AGP card, LAN card solution, and this or
that tweaking to optimize the linux kernel for AMD/NForce2
combination. So, I bought after computer, which compose of
an Intel 865GLC mainboard and a Celeron 2.4G CPU, price here
is around USD92 for board and 70 for CPU (HSF included), the
Intel board comes with Intel 10/100Mbps LAN / on-board VGA.
After a 30-mins assembly of various parts, and installed a
new copy of Fedora Core 1, ALL features on this setup are
recognized and configured properly. And the
llcbench/cachebench figure for RAM is 1700MB/s! It's around
double compare with AMD/NForce2. All 3 setups are rock
solid which runs at 100% CPU load (with SETI@home) 24x7 for
weeks.
In summary:
1. Asus A7N8X-VM + Athlon XP 2600+, price USD190, RAM
perf 980MB/s
2. Winfast K8NCRDLM + Athlon XP 2000+, price USD105,
RAM perf 910MB/s
3. Intel 865GLCL + Cerelon 2.4G, price USD162,
RAM perf 1700MB/s
All run very reliably, but the Intel combination is better
supported in terms of driver installation. All quoted price
is my buy-in price within 2 weeks.
Hope you'll be able to make your wise choice.
Stephen Wong @ Hong Kong.