Re: Seek advice - Upgrading pc.

Follow up on _my_recent_upgrade_ experience_ (server, rather than a
desktop - still a "PC upgrade ;-)


The message <31303030373730364933EC1245@xxxxxxxxxxxxxxxx>
from Johnny B Good <jcs.computers***@xxxxxxxxxxxxxxxx> contains these words:

The message <mab6j454ifko7k3u0nj423u7ulshjnv82q@xxxxxxxxxxxxxxxxxxxxxxx>
from Jaimie Vandenbergh <jaimie@xxxxxxxxxxxxxxxxxxxxx> contains these words:

On Sun, 30 Nov 2008 23:32:29 +0000, John Jordan <junk@xxxxxxxxxxxx>
wrote:

Jaimie Vandenbergh wrote:

However - the extra 4 pins are more power lines (one of each 3.3, 5
and 12 volts) to allow more power to get to the motherboard, and due
to changing power requirements you might run out of juice on one
voltage line from the old-style PSU.

The reason for the extra pins is the potential power requirements of
PCI-E cards, but it's not an issue with high-end video cards because
they draw most of their power over the auxiliary connectors. A pair of
HD4670s could be a problem though.

I'm sure I've seen reference to CPUs using a lot more 5V now than they
used to, which can upset older PSUs that only really expected to run
USB off 5V and won't supply enough amps for modern kit. I could well
be wrong.

You're wrong. It's actually the other way. Some MoBos can run the cpu
vrm from the 5 volts if the 12 v 4pin plug is left disconnected
(obviously, an on board diode bridging the 5v line to the vrm 12 volt
input line in the case of my current Asrock K7S41GX powered FreeNAS
server) and others, such as the new Asrock Alive N7FG-Full HD which I'm
fettling for the FreeNAS server upgrade[1&2], (unconscionably) won't.
:-(

It turned out not to be a problem in spite of that 12v max of 4.2A.

[1] I chose this as the cheapest board with 4 sata ports and built in GB
ethernet[3] which used the cheapest CPU (a 2.2GHz AM2 Semperon) that I
could get from my supplier. I'd prefer to power the VRM off the 5 volt
line if possible (a Gateway 2000 desktop box with a very efficient and
quiet 145 watt PSU and 4 sata drives to be powered is the reason for
this preference), but since I've underclocked it to 600MHz and
undervolted it to 0.85 volt from the 1.35 max, I think the 4.2A [4]
rated 12 volt rail will be able to cope (from past experience, the PSU
seems to have a good short term overcurrent capacity margin).

When I checked the PSU label, it turns out that my memory of the 4.2A
max continuous rating was correct. However, what was a surprise finding
was seeing the 5Vsb limit at a mere 0.1A! This turned out to be a show
stopper as far as the MoBo upgrade was concerned, but I fixed this
(after testing to confirm that the 12v 4.2A limit _wasn't_ going to be
an issue) by adding the innards of a little 5.2v 1.2A smpsu wallwart to
the PSU.

[2] I did try feeding the CPU VRM 12volt from the 5 volt rail after I
discovered it wouldn't boot without the 12v feed plugged in, but the VRM
was incapable of doing its job from the lower voltage (crap switching
regulator design, unlike the expected result with the other board) which
shouldn't have presented any problem for an output voltage that was
still less than a third of the 5 volt rail voltage.

[3] Unfortunately, the current FreeNAS version doesn't support the
RTL8211B chip used on this board, so I've had to plug an RTL8169 based
NetGear PCI card into a slot for now. Hopefully, Open BSD will
eventually support the on board chip one of these days.

[4] Assuming a 50/50 split of power rating between the 5 and 12 volt
rails, I'd have expected a 12 volt rating of 6 amps, but I have this 4.2
amp figure in my mind. I won't know the truth until I lift the lid off
the server box.

I've just raised the Vcore from 0.85v to 1.00v to test the power
difference. The box now takes just under 40 watt (drives spun down)
which is no more than a 1 watt maximum increase on the mains input
power. This suggests a cpu idle condition consumption of a mere 3.6
watts at 600MHz and 1.00v V core from the 12v rail, so it looks as
though it may only be a 300mA draw (which I can test with a multimeter
and the chopped off 4 pin connector I was using to test the 5 volt
option mentioned in note 2 above.

It's not the CPU loading on the 12v rail that concerns me, it's the
spin up surge induced voltage dip effect on the cpu vrm (a start up
induced crash issue). If needs be, I can always knock up a simple DC to
DC converter (95% efficiency) to generate the 12v from off the 5 volt
rail (the original Gateway 2000's 145 watt PSU was so efficient _and_
quiet, it's worth the extra effort to hang onto it :-)

In the event, this wasn't necessary, but I still landed up having to do
more PSU related mods. The first was to address the absence of the 4 pin
12v flylead. The second was to beef up the 5VSB 0.1A line using the
innards of a 5.2v 1.2A smpsu wallwart.

I have almost finished the upgrade, but I'll be opening the PSU yet
again to undo a much earlier mod with the fan thermo regulator speed
controller to remove a 3.3v zenner diode I'd fitted to give the fan's
minimum speed setting a small boost (it's now proved to be a
counterproductive measure ;-).

Upgrading a fileserver box doesn't usually call for any extra
processing power, in fact, for quite a few years now, even the slowest
of the slowest CPUs in current production are usually more powerful by
an order of a magnitude or (even!) two.

In this case, a server upgrade concentrates on storage capacity and
network connectivity speed and minimising the heat production per
terabyte of capacity (ie keeping the electric bill to a minimum ;-).

I've upgraded from a 3 disk setup (it's had four disks previously -
that's why I knew the 4.2A 12v rail could cope) back to a four disk one
(all now 1TB SAMSUNG HD103UJs) from just a pair of SAMSUNG HD103UJs and
a WD5000AAKB.

The previous 3 disk arrangement was a capacity upgrade over its 4 disk
predecessor with a reduction of power from 70 watts to 63 watts. The
current 4 disk setup, I'm pleased to note, now only consumes 58 watts
(once the drive temps climb back up from this morning's 21 degC/60 watts
consumption condition (Ta = 14.6 deg C) to the 30 deg C region they
typically run at when the room temp is at the more comfortable 20/21 deg
level.

I was monitoring power consumption last night (well, TBH, early this
morning ;-) and was a little dismayed to see a slow, but consistent rise
from just below the 58 watt mark to around the 59 watt mark after
fitting the lid. This morning, at 9am, it was showing 60 watts, however,
the amazingly low drive temps of 21/22 deg C tipped me off to a possible
cause for the power increase.

Having now slid the lid back to make the airflow by-pass the drives,
the consumption has now dropped back to the 58 watt mark as the drive
temps have gone back up to the 23, 27, 30 and 31 deg C readings for
drives 1 to 4 respectively ( the last two are stacked one above the
other in the case's original drive bay).

This is the first time that I've ever had to deal with hard disks
running at too low a temperature! The usual problem with most desktop
designs is the woefully inadequate disk cooling provision and drives
(usually Maxtors) running dangerously close to their 60 deg C max even
in moderately warm (circa 25 deg C) room conditions.

I knew that the lower operational limit of 5 deg C for the Samsungs was
a spindle bearing lubricant viscousity issue, but I'm quite surprised to
see the effect over the 20 to 30 deg C temperature range! This phenomina
is probably one of the contributing factors to the absence of a strong
temperature correlation with disk failure that was noted in the figures
published by google on its server farm disk reliability survey recently.

Although I could reduce the consumption by a further 25 to 30 watts by
employing spin down power saving, I'd rather not subject the drives to
such temperature cycling abuse. I can get a a more useful 38 watt
reduction by retiring a couple of my UPSes and consolidating the battery
capacity into my 2KVA Smart-UPS2000 battery bank if I really need to
save electricity.

I'm quite satisfied with the modest 4 to 5 watt saving (once I've
removed the zenner diode fan speed mod from the PSU) and the 1.5 TB
capacity upgrade that I've achieved on a system with a Ta designed
rating of 40 deg C.

For reference, the server is built into an old Gateway 2000 desktop
case with its rather quiet and efficient 145 watt mini ATX PSU (but
please note the extremely low milliamperage rating of the 5VSB line!).
This gave me room to fit another two drives which could be more than
adequately cooled by the whisper quiet airflow generated by the existing
PSU fan alone.

The use of modern drives and undervolting/underclocking techniques
produces an extremely quiet high performance home NAS fileserver which
can operate even under extreme heatwave conditions. The OS is FreeNAS
version 0.69RC1 Salusa Secundus (revision 3991) which is booted off a
pen drive (a 64 MB minimum is required, but I'm using a 1GB unit at the
moment).

As I've already mentioned, the open BSD this is based on doesn't
support the Gigabit ethernet chipset built into the MoBo and I'm having
to use a Netgear PCI adapter for the time being. If support for the
RTL8211B becomes available, I may be able to get a further 2 or 3 watt
reduction and, hopefully, a further boost in connectivity speed
(currently 18.5Mbps write speed and a 28 or so Mbps read speed).

HTH

PS The disks are now (11:33am) at 24, 29, 32 and 33 degrees with the
power down to just below 58 watts (room temperature now at 17 deg). This
pretty well confirms the power consumption variation to be an effect of
drive temperature on spindle bearing lubricant viscousity.

--
Regards, John.

Please remove the "ohggcyht" before replying.
The address has been munged to reject Spam-bots.

.

Loading