In conventional tower server chassis and in low-density
rack chassis installations, there is ample real estate in the front
and rear and on the sides of the chassis for mounting intake and exhaust
fans. In contrast, the low-profile 1U and 2U rack space does not provide
adequate room for exhaust fans. Furthermore, there is precious little
room for intake and exhaust vents, which compromises fan performance,
lowering their actual performance to significantly below their "rated"
airflow capacity, which is designed for a minimal resistance (to airflow)
environment. In fact the only exhaust system available on a conventional
1U or 2U chassis is the power supply exhaust fan which is totally
inadequate to provide sufficient exhaust of hot air to allow the system
to cool reliably. The result, of course, is an unstable and virtually
useless system.
The
thermal dilemma becomes even more critical when conventional fan top
or blow through CPU coolers are used. The efficiency of all CPU coolers
is linked directly to the temperature of the intake air. Every degree
rise in intake air temperature to the CPU cooler directly translates
into a corresponding rise in the CPU junction temperature.
In the conventional server chassis the warm exhaust
air from the CPU cooler is re-circulated inside the chassis causing
the CPU temperature to rise until it reaches a thermal equilibrium
which is determined by the chassis's exhaust system capacity. In our
tests at CCSI, we have seen CPU temperature rises of between 25 and
48 degrees centigrade in several conventional 1U and 2U chassis specimens
that we have tested.
This
is easily observed with a simple cover on and then cover off test
while running a CPU loading utility. Another major shortcoming of
the conventional fan top CPU coolers is the reduction of airflow due
to pressure drop resulting from the airflow obstruction of the chassis
cover and the fins of the heat sink itself. Fans performance is rated
in cubic feet per minute (CFM) with 0 pressure drop and performance
is severely compromised with only minimal air flow obstructions from
either the intake or exhaust side of the fan. The CCSI Wind Tunnel
eliminates these problems.
ONLY THERMAL
CORRECTNESS CAN ELIMINATE CPU THROTTLING
As a self-protection feature, some CPU's incorporate
an internal temperature monitoring diode and a throttling circuit.
In simple terms, you can think of this as a thermally controlled governor.
The governor in this case wields a two-edged sword.
On the plus side, the governor will prevent the CPU from overheating
to the point of system failure, and thus provides for a stable system
even when built into a thermally sub-standard chassis. On the minus
side, the governor limits the performance of any such system by automatically
lowering the CPU speed. In other words, while you get a stable system
regardless of the thermal performance of the chassis, you don't "get
what you pay for". The really sad thing about this situation is that
you have no way of knowing if and how much your system performance
has been compromised and performance-monitoring software is no help,
because it will of course report that the CPU is fully occupied. While
there is undeniable value in having a stable system, we at CCSI feel
that insuring system reliability by "hobbling" system performance
is a bit like harnessing a racehorse to a sightseeing carriage or
perhaps in an extreme situation to a plow. While the racehorse certainly
could pull the carriage or the plough, that is not what he was born
and bred for!
CPU's that do not have the throttling feature always
run at full speed, but require a chassis that provides proper cooling
in order to run reliably. The advantage of a thermally correct chassis
is that it will not only allow a non-throttled CPU system to run reliably,
but will also allow a throttled CPU to run without the governor being
invoked, thus optimizing both system reliability and performance.
As can be seen from the above discussion, it is critical
when purchasing a chassis to make sure that it is thermally correct
in order to avoid CPU throttling. Any statement to the effect that
"we have the system running at such and such speed with no problem"
should be taken with at least a "grain of salt". Since the system,
while operating reliably, may not truly be "running", but "walking"
or even "crawling".
Power supply design for 1U and 2U servers is another
extremely challenging task due to several opposing design criteria.
On the one hand we must provide more power for the higher speed dual
CPU processor servers and on the other hand we have far less physical
space available for the power supply plus we have to contend with
the thermal issues described above.
Increasing
power output while significantly reducing the size, and providing
the required thermal management of high output 1U power supplies is
a daunting task, which is well beyond the abilities of the average
conventional designer. In fact, with experienced engineering staff
with many years of experience in designing high density systems and
solving thermal management challenges, CCSI still spent over 2000
man-hours on these solutions and went through three different power
supply designs to arrive at the final dual 300 watt power supply utilized
in our RC 0103 1U chassis.
To solve these thermal obstacles and
build "future proof" thermally correct high-density high performance
1U and 2U solutions, CCSI created a task force of engineering experts
with one objective: to design and develop both 1U and 2U server solutions
that meet and exceed the thermal requirements of not only today's
but also future high performance CPU's.
Our Goals were:
A) To invent a new exhaust system
which allows the maximum cross flow possible in these high-density
server chassis. Mission Accomplished
B) To invent a new "outside the box"
CPU cooling system that draws air from outside the chassis rather
than recirculating warm air from within the chassis. This design
incorporates adequate volume, velocity and pressure of air to not
only cool the CPU's but also the other hot components on the motherboard
as well as peripheral cards. Mission Accomplished
C) To invent a new generation of high
power low profile 1U power supply to meet the power budget of current
and future high performance CPU's. Mission Accomplished
D) To invent an all new server component
layout for optimum configuration, fault tolerance and serviceability.
Mission Accomplished
It took our team over 2000 man-hours
to complete these tasks and finally gain the recommendation of our
most stringent partner, AMD, for all their present and planned future
speed grades of the Athlon MP processors.