Undervolt Your i9-13900K: Intel XTU to BIOS on ASUS Z790, Power Limits Explained

ByMarcus Vale

Why the i9-13900K runs so hot out of the box

The i9-13900K has a rated TDP of 125W. In practice, ASUS and most other Z790 board makers ship with MultiCore Enhancement (MCE) enabled by default, which removes Intel’s official power limits and lets the CPU draw as much as it wants under sustained load. That’s how you end up with package power hitting 250–300W and temperatures pinned at 100°C the moment you open Cinebench.

Intel’s own spec puts the Long Duration Power Limit (PL1) at 125W and the Short Duration Limit (PL2) at 253W. Most Z790 boards ignore both and run uncapped. This is the root of the problem, and it’s what Intel’s Baseline Profile update tried to address.

The Intel Baseline Profile: what 253W and 288W actually mean

Intel released a Baseline Profile BIOS update for board partners to address the crashing and instability problems that hit 13th and 14th gen chips. The profile enforces Intel’s stock power limits: PL1 at 125W, PL2 at 253W.

Research from Puget Systems found the i9-13900K averages around 58–59°C with the Baseline profile applied, versus 100°C package temps under unrestricted settings. Multi-threaded performance drops roughly 8–10% in sustained workloads like Cinebench, but gaming is barely affected since games rarely sustain full package power long enough for the limits to bite.

The 288W figure that shows up in some board menus is a non-Intel-standard intermediate: PL1 raised above spec but still capped, sitting between Intel’s Baseline and full MCE-off mode. It’s a compromise value, not an official recommendation from Intel.

One important context: Intel later clarified that the Baseline Profile was primarily aimed at B760 boards with locked CPUs. K-series chips on Z790 were hitting instability for a related but distinct reason, namely excessive sustained voltage from unlocked power delivery rather than a simple profile mismatch. Applying it still helps thermally, but it won’t fully resolve instability on its own.

What undervolting actually does

Every chip off the fab is slightly different. Some hit their turbo clocks comfortably at lower voltage; others need more headroom to stay stable. This is the silicon lottery. Undervolting means finding where your specific chip sits and dialing the voltage down accordingly.

The result is lower operating temperatures, typically 8–15°C less at peak load. Less heat means less thermal throttling, which can actually improve sustained multi-threaded performance compared to a stock system sitting at 100°C with the CPU pulling back frequencies to survive. You’re not slowing anything down. You’re removing a ceiling.

Undervolting does not change clock speeds. It’s not overclocking. The only real risk is instability if you go too far: crashes, blue screens. Not hardware damage.

Using Intel XTU as a testing sandbox

Intel Extreme Tuning Utility (XTU) is the right starting point because its settings reset on every reboot. If you push too far and the system crashes, a power cycle puts you back to stock. No stuck BIOS, no drama.

Step by step in XTU

  • Download Intel XTU from Intel’s official site and install it on Windows.
  • Open XTU and click the Advanced Tuning tab in the left sidebar.
  • Find Core Voltage Offset. This field subtracts voltage from what the CPU would normally request across all cores.
  • Enter -0.050 (negative fifty millivolts) as your starting value. Click Apply.
  • Open Cinebench R23 and run the multi-core test at least twice. Monitor temperatures in HWiNFO64 or HWMonitor at the same time.
  • If stable, drop to -0.075V and repeat. Then -0.100V.
  • Most 13900K chips find a stable floor somewhere between -0.100V and -0.150V. Some reach -0.160V.
  • If the system crashes, step back up 0.025V. That’s your ceiling for now.

Use offset mode, not static voltage. Static voltage locks every core to a single number, which tends to overvolt the efficiency cores while potentially starving the performance cores under burst. Offset mode preserves Intel’s per-core voltage scaling and is the correct choice here.

Once you’ve found a value that holds across multiple Cinebench passes, write it down. That number goes into BIOS next.

Moving the undervolt into ASUS Z790 BIOS (AI Tweaker)

XTU values reset on reboot, so they’re only useful as a testing tool. Once you have a stable number, you want it baked into firmware. Here’s the path on ASUS Z790 boards:

  • Enter BIOS at POST by pressing Del or F2.
  • Navigate to AI Tweaker.
  • Look for CPU Core/Cache Voltage or Global Core SVID Voltage. The exact label varies slightly by board revision.
  • Set the voltage mode to Offset Mode.
  • Set Offset Mode Sign to − (negative).
  • Enter your tested offset value (for example, type 0.100 to apply −0.100V).
  • Save and reboot, then run Cinebench R23 again to confirm the stability you found in XTU carries over.

Some ASUS Z790 boards expose separate P-Core and E-Core voltage offsets. You can experiment with individual values per cluster for a more dialled result. For most people, a single global offset is simpler and gets you the majority of the benefit without added complexity.

Adjusting power limits: the other lever

Undervolting and power limiting are separate tools. You can use one, both, or neither.

To change power limits on ASUS Z790:

  • In BIOS, go to AI Tweaker → Internal CPU Power Management.
  • Find Long Duration Package Power Limit (PL1). On most Z790 boards this defaults to uncapped or a very large number. Setting it to 125W matches Intel’s official spec. A value around 180W is a common middle ground that keeps temperatures manageable while preserving more multi-threaded throughput than full Intel spec.
  • Set Long Duration Maintained to 56 seconds or higher to enforce the limit over time rather than just at burst.
  • Short Duration Package Power Limit (PL2) can remain at 253W or be matched to PL1 for stricter sustained control.

Power limits matter most in sustained all-core workloads: video encoding, 3D rendering, long compilation jobs. In gaming, which is typically bursty and rarely taxes all cores at once, a PL1 reduction usually produces no visible change in fps.

What results to expect

A −0.100V offset alone typically shaves 8–12°C off peak temperatures. Combined with a 180W PL1, most 13900K systems on a 360mm AIO stay under 90°C even in sustained Cinebench runs. With 125W PL1 and a full undervolt, sitting around 70°C under load is realistic.

Single-core and gaming performance stays essentially flat. Multi-threaded throughput depends on how tightly you constrain the power limits; the undervolt itself costs nothing in that department.

Silicon variance is real on Raptor Lake. The stable offset on your specific 13900K might be −0.080V or it might be −0.160V. Don’t chase numbers from someone else’s forum post. Find your own floor.

Sources


Similar Posts