4 Questions About Freecoolling You Should Ask the Manufacturer

A chiller system is the most expensive, most energy-intensive, and most temperamental part of a data center’s engineering infrastructure. It’s no surprise that more and more operators are switching to free cooling: cooling using outside air. The savings begin immediately, even during the construction phase, and continue every month on electricity bills.

But there’s another side to this coin, and operators face it right after launch. Without chillers, the temperature in the cold aisle is no longer constant. In winter, everything is fine. In the summer, air enters the corridor that the average server simply isn’t designed to handle, and instead of savings, you end up with a drop in performance at the worst possible moment.

To prevent this from happening, servers for such a data center are selected based on different criteria. Below are four questions you should ask any manufacturer, along with the answers for the ITPOD-SR201-S12R-NV-G2 platform .

Question 1. At what temperature is the heatsink’s cooling capacity rated?

This is the most important question, and almost no one asks it.

A heat sink operates based on temperature difference, not absolute temperature. It doesn’t matter how many degrees it is in the room; what matters is how much cooler the air is than the processor. The processor can tolerate approximately 85 °C. With an inlet air temperature of 25 °C, the heatsink has a margin of 60 degrees. At 45 °C, that margin drops to 40. It’s the same heatsink, the same computer, but significantly less heat is dissipated.

Then the protection kicks in: the processor throttles its clock speed. Everything looks great on paper, but on a July afternoon, the system responds more slowly.

That’s why, in the ITPOD-SR201-S12R-NV-G2 datasheet, wattage is listed in conjunction with room temperature, rather than separately. Three heat sinks have been developed for this platform:

Take a look at the second row. On paper, the EVAC is “weaker” than a standard heatsink, although in reality it’s more complex and more expensive. The reason is that its 350 W rating is specified for an inlet temperature of 45 degrees, not for laboratory conditions.

And the main takeaway from the third row: the flagship AMD EPYC with a 500-watt thermal envelope remains on standard air cooling even in a hot environment. No liquid, no additional loops, no new maintenance requirements.

All three designs were simulated in Flotherm—an industrial thermal simulation package—rather than selected by analogy with the previous generation.

Question 2. What happens to the hot air inside the chassis?

When the inlet temperature is 45 degrees, air inside the server becomes a scarce resource, and its distribution must be carefully managed.

One processor instead of two. It’s a single-socket platform: inside is a single AMD EPYC processor from the Genoa or Turin generation. In a dual-processor system, the second chip sits in the aerodynamic shadow of the first and receives air that’s already been warmed up. At 25 degrees at the inlet, this is tolerable; at 45 degrees, it is not. Here, the entire air volume is directed toward a single chip, and a heatsink of such size is placed beneath it that there would be physically no room to fit a second one.

The memory gets its own airflow. Inside the case is a Wind Reflector air duct. It prevents the air from taking the path of least resistance and forces it into the area of the DDR5 modules. The modules themselves are covered by aluminum heatsinks with dense finning. This isn’t just a decorative detail: overheated DDR5 reduces its frequency without a single error in the logs, and you simply lose memory performance without understanding why.

The “one module per channel” layout offers an additional benefit: 12 DDR5 slots running at 6400 MT/s without a second row of modules. This maximizes memory bandwidth while halving the number of obstacles in the airflow path.

The network card isn’t exposed to exhaust heat. The OCP 3.0 module is located in the worst possible spot in the case—right at the hot-air outlet. A modern 100 or 200 Gb/s adapter generates as much heat on its own as a small stove, and you can’t add external heat to that. A deflector isolates it from the CPU airflow.

There are no unnecessary heat-generating components at the front. The storage path contains no SAS expander chip: the NVMe drives are connected directly to the motherboard via MCIO cables. The latency benefits are obvious, but there’s also a thermal advantage. Anything that heats up in the front area compromises cooling for everything further down the airflow path. Here, this heat source is simply absent.

Question 3. What fans are required for a free-cooling project?

This is rarely discussed openly, but it’s worth noting: by doing away with chillers, you’re partially shifting the energy consumption to the servers’ own fans. How cost-effective the switch to free cooling will be depends largely on exactly how the system circulates the air.

The math here is straightforward. The power consumed by a fan increases roughly as the cube of its rotational speed: spin it twice as fast, and you’ll pay eight times as much.

The ITPOD-SR201-S12R-NV-G2 features six 60-mm fans in the central part of the chassis, directly behind the drive cage. You can choose a fan configuration tailored to your specific facility:

• Single-rotor systems are simpler and cheaper. A sensible choice for moderate heat loads and stable room temperatures.
• Dual-rotor—two counter-rotating blades generate the required airflow at significantly lower RPMs. For free-cooling environments, where fans operate at the upper end of their performance range for much of the year, the difference in annual costs is quite significant.

Every watt not spent on fan rotation stays in the computing budget.

And here’s another operational detail. At elevated temperatures, fans work harder and wear out faster. Here, they’re replaced “hot-swap” - without interrupting the load—which isn’t an option but a necessity for a data center like this.

Question 4. How will I see all this in the monitoring system?

With free cooling, the inlet temperature changes every day due to weather, season, and time of day. It’s impossible to manage such a facility blindly.

A dedicated BMC ASPEED AST2600 controller handles out-of-band management: it features a dedicated 1GbE network port, an informative web interface, and support for Redfish 1.15, SNMPv3, and IPMI 2.0. The practical benefit is that metrics are retrieved via standard web requests and can be integrated into your existing monitoring system without the need for specialized utilities.

Here’s what this means in a data center with hundreds of servers:

• you can see which racks have systematically higher inlet temperatures than the rest;
• you can immediately spot servers whose fans are constantly running at high speeds. This is the first sign of clogged filters or hot air being drawn in from the back of the rack;
• you can plan the placement of heavy workloads by season rather than having to put out fires after the fact.

Plus, there’s diagnostic feedback right on the board, including a dedicated memory error LED. A field engineer won’t have to guess which of the twelve modules is at fault.