Archives

- Beowulf
- Beowulf Announce
- Scyld-users
- Beowulf on Debian

Problems with dual Athlons

Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.

Search

Robert G. Brown rgb at phy.duke.edu
Wed Jul 31 15:20:58 PDT 2002

```On Wed, 31 Jul 2002, Bari Ari wrote:

> One of the most often overlooked and misunderstood problems with PC
> power supplies is Transient Regulation or Dynamic Regulation. Many
> factors play into the transient response of PC AC/DC and onboard DC/DC
> power supplies; switching regulator frequency, component tolerances and
> the series resistance of power supply cables, connectors and PC board
> traces.
>
> Motherboard and PC power supply manufacturers meet cost targets by using
> low-cost, low-ESR electrolytic capacitors. However, if manufacturers
> excessively reduce capacitor values and sizes, they encounter another
> problem. Capacitor life is much shorter for capacitors with a lower
> value and higher ESR. With lower capacitance, higher ESR, or both, the
> ac-current components that heat the input and output filter capacitors
> cause the capacitors to dissipate more power. As a capacitor's
> dissipation increases, its operating temperature rises. An electrolytic
> capacitor's life decreases with increasing temperature.
>
> You can calculate the capacitor life using the formula below. Capacitor
> life depends on operating temperature, ESR, and ripple current:
>
> L=LCX2((TC-TMAX)X0.1).
>
> L is the capacitor life in hours. For a capacitor rated at 105°C maximum
> operating temperature and at 5000 hours life at that temperature,
> TC=105°C and LC=5000. TMAX is the maximum temperature at which you
> operate the capacitor whose life you are calculating. Use the formula
> below for TCAPCASE, which establishes TMAX for the above equation.
> (These formulas apply to United Chemi-Con LXV-series capacitors.)
>
> TCAPCASE=80X(AC-0.7)X(PCINC0.5)+TAMB,
>
> where AC is the capacitor surface area in inches squared;
> PCINC=ESRXIAC2; IAC=IOMAX/(2XN), where N is the number of input
> capacitors connected in parallel, and ESR is the equivalent series
> resistance of the capacitor at operating temperature and frequency; and
> TAMB is the ambient temperature.
>
> The calculated life of the C1 input-filter capacitors at 50°C ambient
> temperature for United Chemi-Con LXV series capacitors is as follows:
>
> Fully compliant circuit=63,000 hours, or approximately seven years;
> Mostly compliant circuit=25,000 hours, or approximately three years.
>
> In today's computer market, a variety of inexpensive motherboards and
> power supplies is available that has dramatically smaller capacitors
> with less capacitance than those in the designs above. For some of these
> inexpensive boards and power supplies, we calculated the capacitor life
> to be approximately three months!
>
> The problems on the other side of the power supplies are the AC power
> line are low power factor, harmonics caused by the large (or multiple in
> a cluster situation) AC/DC switching supplies and poor wiring
> installation or use of hardware store quality power strips. The static
> voltage and current may look fine or be installed to the latest NEC code
> but the dynamic properties of the AC power lines may be inadequate for a
> cluster.
>
> Bari Ari

Wonderful lecture, Bari.  So, granting all the above (and very likely
being in a situation where our dynamic properties are inadequate) how do
we a) test to determine that this is so; and b) fix the situation?  Our
wiring was done by a commercial contractor that is often used by Duke.
One thing that they did that had me a bit concerned from the beginning
was wire the power poles by pulling all three phases to a power pole and
coming back with a single common ground.  Of course if you have three
equal, purely resistive loads on all three hot wires you could in
principle grab the common cold wire in your hand as it would carry no
current (Kids! Do Not try this at home!) but in practice it is not at
all clear that the load currents of a cluster will all be in phase with
the current (or even shifted by a common amount).  In practice, our
racks were popping breakers before they reached 75% of nominal load,
making me wonder if they actually had three DIFFERENT phases sharing a
ground...

Are there published standards for wiring clusters (or similar
environments)?  We will soon be having them redo the power pole wiring
so each line has its own ground.  Is there anything (relatively
inexpensive) we can add or should insist on at that point to condition
the power and isolate the harmonics so that they don't bleed through
from system to system?  I think dual isolation transformers are beyond
our means, but could a high-quality surge strip (perhaps one with
limited UPS capacity) serve the same purpose?

rgb

Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu

```