Case Study | Does Your Server’s KC RRA Still Apply After a Hardware Change?

Does a New Server Configuration Means a New KC RRA Assessment?

With the rapid global growth of AI applications, large computing equipment such as AI servers, GPU servers, IT racks and liquid-cooled servers is becoming an important product category in the data center market.

In our Korean KC RRA projects over the past two years, we have clearly seen a steady rise in EMC certification demand for large servers and rack systems.

However, the nature of AI servers also brings new challenges to certification work.

The equipment keeps getting bigger and heavier, which raises the logistics cost and transport risk of shipping a complete system to an overseas lab for testing. So a very common question from server manufacturers is:

“The whole server already has KC RRA. If we later just swap the SSD, NIC, add-on card or FPGA card, do we need to test again?”

The answer can’t be decided simply by asking whether the part’s model number has changed.

The client’s intuition: same model, same ID, so it should carry over?

We recently handled a real case like this.

One of the client’s FPGA add-on cards had already completed Korean KC RRA EMC assessment as part of an air-cooled L10 server system.

As liquid cooling becomes one of the mainstream designs for AI servers, the client wanted to offer the exact same FPGA card model as an optional part in a new liquid-cooled server system.

For the FPGA card itself:

• The model name had not changed
• The PCB design had not changed
• The function had not changed
• It had already appeared in an approved server configuration

So the client asked a very reasonable question:

“Since it’s the same card, and it has already been used in another KC RRA server system, we shouldn’t need to retest, right?”

This is also one of the most common certification misconceptions among server manufacturers.

Same Part ≠ Same EMC Configuration

Korea’s RRA conformity assessment is not just a check of a card’s part number.

What actually has to be confirmed is this: when the part is used in the configuration of the product as actually sold, does the equipment as a whole still meet Korea’s EMC technical requirements?

Even when an air-cooled server and a liquid-cooled server use the same FPGA card, their actual EMC environments can be completely different. For example:

So even if the FPGA card itself is identical, once it is installed in a different server architecture, the original test results do not automatically extend to the new configuration.

Korea’s RRA Focuses on Configuration Changes

RRA has clear management requirements for changes to products that have already completed conformity assessment.

Changes involving, for example:

• Circuit changes
• Component replacement
• Component addition
• Parts addition
• Optional configurations

may all be treated as changes related to the conformity assessment technical requirements, so it has to be confirmed whether they affect the original test results.

Therefore:

• Same model name ≠ automatically covered
• Same KC ID ≠ all future configurations automatically covered

Whether additional testing is needed still comes back to the actual hardware configuration and the coverage of the original KC RRA test report.

Does that mean the whole server must be retested every time an SSD, NIC or add-on card changes?

No. This is exactly where AI server KC RRA projects need the most professional judgment. RRA itself allows some changes to skip a full retest when specific conditions are met.

For example, the RRA FAQ states that in certain computer component replacement scenarios, if the replacement is a component that has already completed conformity assessment and has equivalent function, it may be handled through the change procedure instead of a full new certification.

RRA’s 2026 regulatory revision simplifies this further in the same direction, allowing qualifying components with the same function that have already completed conformity assessment to be handled as a change in some cases, without additional testing.

So the real question is not: “Did this card have KC before?”. It should be: “In the new server configuration, does this card still fall within the technical coverage of the original KC RRA assessment?”

What happened in our case?

Back to our FPGA card case. The card model stayed exactly the same, but the server changed from air cooling to liquid cooling. That switch brings a different system configuration and different operating conditions from the air-cooled L10 system that was originally assessed.

So we could not decide, based only on “the FPGA card model is the same,” that the air-cooled server’s KC RRA assessment results could be used directly.

We needed to re-examine:

• The configuration in the original KC RRA test report
• The FPGA card’s original installation position
• The hardware architecture of the liquid-cooled system
• The PSU, GPU, accelerators and other high-speed components
• Cable routing and grounding
• The new electronics added by the cooling system, such as the pump and controller
• The worst-case EMC operating mode
• Whether the original test configuration is enough to represent the new liquid-cooled system

After the configuration assessment, if the new liquid-cooled configuration goes beyond the representative / worst-case coverage of the original report, the affected KC RRA EMC items need additional testing or re-assessment. You can’t simply rely on the existing KC ID.

A Configuration Change Doesn’t Always Require a Full Retest

This is also the most important cost-control concept for large AI server projects. When a configuration changes, the first step should be a Regulatory Impact Assessment, not immediately scheduling a full retest. By comparing the original certified configuration with the new shipping configuration, you can first determine which outcome applies:

• No impact: The change is already covered by the original test report, so the existing certification still applies.
• Minor change: The change needs to be reported to RRA, and the documents or registration are updated, but no new testing is required.
• Major configuration change: The change goes beyond the original report’s coverage, so additional testing or a new conformity assessment is needed.

That way you meet Korea’s RRA requirements while cutting the cost of repeatedly shipping and testing large servers.