Intel Phi 7120A SC7120A Data Sheet
Product codes
SC7120A
Intel
®
Xeon Phi™ Coprocessor Datasheet
Document ID Number: 328209 003EN
18
3.2
Intel
®
Xeon Phi™ Coprocessor Thermal
Specification
3.2.1
Intel
®
Xeon Phi™ Coprocessor Thermal
Management
Thermal management on the Intel® Xeon Phi™ coprocessor card is achieved through a
combination of coprocessor based sensors, card level sensors and inputs, and a
coprocessor frequency control circuit. Reducing card temperature is accomplished by
adjusting the frequency of the coprocessor. Lowering the coprocessor frequency will
reduce the power dissipation and consequently the temperature.
combination of coprocessor based sensors, card level sensors and inputs, and a
coprocessor frequency control circuit. Reducing card temperature is accomplished by
adjusting the frequency of the coprocessor. Lowering the coprocessor frequency will
reduce the power dissipation and consequently the temperature.
The coprocessor carries in it a factory calibrated Digital Temperature Sensor (DTS) that
monitors coprocessor temperature, also called junction temperature (T
monitors coprocessor temperature, also called junction temperature (T
junction
). Data
from this sensor is available to the BMC or other system software via both in-band
(direct software reads) and out-of-band (over the PCI Express* SMBus) interface. Refer
to chapter titled “Manageability” for more information on how to read the junction
temperature. System management software can use this data to monitor the silicon
temperature and take any appropriate actions. Systems that adjust airflow based on
component temperatures must monitor the coprocessor’s DTS to ensure sufficient
cooling is always available.
(direct software reads) and out-of-band (over the PCI Express* SMBus) interface. Refer
to chapter titled “Manageability” for more information on how to read the junction
temperature. System management software can use this data to monitor the silicon
temperature and take any appropriate actions. Systems that adjust airflow based on
component temperatures must monitor the coprocessor’s DTS to ensure sufficient
cooling is always available.
In addition to making thermal information available to system manageability software,
the DTS is constantly comparing the coprocessor temperature to the factory set
maximum permissible temperature called T
the DTS is constantly comparing the coprocessor temperature to the factory set
maximum permissible temperature called T
throttle
. If the measured temperature at any
time exceeds T
throttle
(a state also known as PROCHOT), then the coprocessor will
automatically step down the operating frequency (or Pstate) in an attempt to reduce
the temperature (this is often referred to as “thermal throttling”). Once the
temperature has dropped below T
the temperature (this is often referred to as “thermal throttling”). Once the
temperature has dropped below T
throttle
, the frequency will be brought back up to the
original setting. See
below.
Table 3-2.
Intel
®
Xeon Phi™ Coprocessor Thermal Specification
Parameter
Specification
T
RISE
10°C
Max T
INLET
45°C
Max T
EXHAUST
70°C
T
case (processor)
min, max
5°C, 95°C
T
control
~82°C
1
Notes:
1. T
control
is the setpoint at which the system fans must ramp up towards full power (or RPM) to maintain the
Intel® Xeon Phi™ coprocessor temperature around T
control
and prevent throttling. It is a requirement that
the system BMC use IPMB commands to query the SMC on the coprocessor card for accurate T
control
value
as this value can vary between 80
°C
and 84
°C
.
T
throttle
104°C
2
2. When the coprocessor junction temperature (T
junction
) reaches T
throttle
, the SMC will force thermal throttle
which will drop frequency to lowest supported value and reduce total coprocessor power. It is a requirement
that the system BMC query the SMC on the coprocessor card for accurate
T
throttle
value.
T
thermtrip
~(T
throttle
+ 20°C)
3
3. If the coprocessor temperature reaches T
thermtrip
, the coprocessor OS will take action to shutdown the card
to prevent damage to the coprocessor. This includes shutting down the coprocessor VRs, and the only way
to restart the coprocessor is by rebooting the host system. T
thermtrip
should not be considered a
specification; it can change between SKUs, and is given here as guidance.