RIKEN: “Fugaku” original cooling system: CPU, always 30 degrees or less
-“Fight against heat” unfolding behind the scenes-
RIKEN’s “Fugaku”:
RIKEN’s “Fugaku”, which was the fastest in the world in terms of calculation speed
We approached the “fight against heat” that unfolds behind the scenes.
“Fugaku” is equipped with its own cooling system to cope with the huge heat generated by power consumption.
Proprietary cooling system:
Toshiyuki Tsukamoto, deputy director in charge of the cooling system, explains:
The calorific value (calorific value per unit area) of “Fugaku” is in the nuclear reactor class.
Furthermore, the amount of heat generated also greatly fluctuates depending on the content of the calculation.
Only supercomputers are required to respond to such huge thermal fluctuations.
RIKEN Center for Computational Science:
Cooling system that supports “Fugaku” 432 racks lined up in the computer room on the 3rd floor of the building.
A total of 160,000 high-performance CPUs are included.
When the CPU operates due to the operation of the supercomputer, it generates heat.
Supercomputer heat generation density:
Its calorific value exceeds “up to 100 kW per square meter”.
The electric stove (1kW) is so hot that “100 units are moving at the same time in 1㎡”.
Keep CPU below 30 degrees:
In order to “operate a CPU with a mechanism similar to that of a nuclear power plant efficiently,” it is necessary to keep the temperature below 30 degrees.
If not cooled, it will reach 100 degrees or more in a few seconds.
Cool by running water on the board:
At Fugaku, water is run through the system board equipped with the CPU to cool it.
The mechanism is also large-scale, similar to the nuclear power plant.
Two cooling systems:
The cooling system is divided into two systems, a “primary cooling circuit” and a “secondary cooling circuit”.
By combining these, the temperature is adjusted well.
“Secondary cooling circuit”:
It is the secondary cooling circuit that directly cools the CPU.
Water of about 15 degrees circulates near the 160,000 CPUs that generate heat and takes heat away.
The heat heats the water to 19-25 degrees and returns it to the heat exchanger.
“Primary cooling circuit”:
The water in the secondary cooling circuit is cooled by a total of 11 refrigerators.
It is cooled to 15 degrees with water from the primary cooling circuit and circulates in the circuit again.
Use the primary system and the secondary system properly:
The primary system and the secondary system are separated so that water does not mix.
This is to prevent impurities from clogging the thin pipes near the CPU.
Water, which is a mixture of pure water from which impurities have been removed and a corrosion inhibitor, is used for the secondary system.
On the other hand, the primary system circulates industrial water.
Response to fluctuations in calorific value:
While the supercomputer is in operation, the operating status of the CPU changes in detail depending on what kind of calculation is performed.
Power consumption and calorific value fluctuate greatly accordingly.
In just 1/1000 second, the amount of heat generated jumps up significantly.
Simulate the movement of water:
Therefore, Mr. Tsukamoto and his colleagues simulated the movement of water circulating in complicated pipes.
I thought of “a system that can respond even if something unexpected happens”.
Installation of water temperature monitoring sensor:
On top of that, we installed a “sensor that constantly monitors the temperature of the secondary water going to the CPU.”
The temperature of the secondary water
So that it becomes constant at about 15 degrees
Adjusting the amount of water circulation in the primary system,
We have a monitoring system.
Mainichi Shimbun-Science / Technology
https://headtopics.com/jp/124731249712467538144-18681372
Tour “Fugaku” at Matterport
“Fugaku”:
Two CPUs on one motherboard,
One calculator rack consists of 192 motherboards,
A total of 384 CPUs are installed.
Connect 150,000 computers:
The height of the calculator rack is 220 cm and the weight is as much as 2 tons.
There are 432 racks (384 CPU x 396 racks, 192 CPU x 36 racks).
“Fugaku” with a total of more than 150,000 CPUs is configured.
In other words, “Fugaku” is connected to more than 150,000 computers.
Cooling system:
And on the floor below the floor of the computer room where racks are lined up,
A pump and heat exchanger for cooling “Fugaku” with water are installed.
Pipes through which cooling water flows are stretched around the ceiling.
itok