Why dual-CCD X3D chips are the special case
AMD's dual-CCD X3D processors (the 7900X3D, 7950X3D, 9900X3D and 9950X3D) split their cores across two chiplets, and only one of them carries the stacked 3D V-Cache. The other chiplet clocks higher but has far less L3 cache. For most games you want every thread on the cache die, because hopping between chiplets over the Infinity Fabric adds latency.
The CPU can't tell a game from a spreadsheet, so Windows has to supply that signal. According to XDA, the chipset driver uses Game Bar's game detection as the cue to park the cores on the higher-frequency chiplet. With those cores parked, Windows schedules the game on the V-Cache die. A community post on AMD's forum describes the same chain. It says AMD recommends keeping Game Bar enabled and updated, so that the Windows kernel puts most games on the V-Cache die and compute-heavy work on the standard CCD.
That is where the power plan matters. XDA reports that AMD recommends the Balanced plan for dual-CCD systems, and that owners who stuck with High performance have said core parking doesn't engage until they switch back. If parking fails, a game can spread across both chiplets and pay the cross-die latency penalty.
How firm is that evidence?
Treat it as a strong configuration warning, not a measured penalty. I found no controlled benchmark giving a percentage loss for High performance on each of these four chips. Most of the support is user reports. Examples:
- One Steam community post from a 7950X3D owner says SteamVR switches the power plan to High performance when it launches, and that choosing Balanced restores 3D V-Cache use and correct core parking. The same poster reports that switching back to Balanced fixed FPS spikes in the game Vail. That is one person's experience.
- An AMD forum thread about a 7950X3D with no parking shows a helper asking whether the power plan was set to High performance rather than Balanced. The poster eventually reported that a fresh Windows install fixed the parking problem. So the power plan isn't the only possible culprit.
Reports also differ by game and setup. One AMD community user reportedly switched to High performance to cure stutter in a single title. That is an exception, not a rule, but it shows why blanket claims fall apart.
Why High performance rarely helps modern CPUs anyway
The old case for High performance rested on slow frequency changes. XDA says the plan pins minimum processor state at 100%, stops core parking and makes Windows quicker to raise clocks. That helped when the OS managed frequency directly and a change could take about 30 ms.
Newer silicon changed this:
- Intel Speed Shift (Skylake onward) hands frequency control to the CPU and cuts transitions to roughly a millisecond.
- AMD CPPC2 arrived for Ryzen in summer 2019. AMD said it cut clock ramp times from about 30 ms to 1–2 ms.
- AMD's published Precision Boost 2 description (not retrieved in this session, so I'm relying on my own notes) says the processor adjusts clocks continuously, many times per second.
Once a game is running, the cores it uses are already boosting. What High performance mostly changes is idle behavior. XDA says the CPU idles hotter and draws more power, and the fans respond accordingly. That is a small cost for most people, but it buys nothing in return.
The Intel hybrid exception
High performance isn't pointless, and Intel's hybrid CPUs show why. Developer Alois Kraus tested Alder Lake-era and later chips (Core i5-13600 and i7-13700) with a synthetic load tool. He found that on Windows 10, scheduling policy can send below-normal-priority work to E-cores. High performance's defaults favor P-cores. His own caveats matter:
- Kraus says the results are much less dramatic on Windows 11 thanks to Intel Thread Director.
- He also saw erratic scheduling on Windows 10 with High performance, where not all the needed P-cores were used in some runs. He suspects a kernel scheduler issue for many-thread, short-lived workloads.
- His workload was a CPU-bound synthetic test, not a game benchmark. Don't read it as "Intel gamers should switch plans."
Customized plans: don't rely on them
Some X3D owners duplicate Balanced and edit it to mimic High performance, reportedly keeping core parking and seeing no benchmark difference. XDA notes that Hardware Busters reported the opposite on a 9950X3D, where customizing the plan through the legacy Control Panel broke parking. I couldn't independently verify that test. With reports in conflict, stock Balanced is the safer starting point.
What to do
- Dual-CCD X3D owners: set Windows to Balanced and leave it there as a baseline.
- Install the AMD chipset driver for your platform from AMD's support pages. Package contents vary by Windows version and release, so check the notes. Restart afterward.
- Keep Windows Game Mode and Xbox Game Bar enabled and updated. If a game isn't recognized, Game Bar has a setting to remember it as a game. The Vail post describes ticking this option after pressing Win+G and opening Game Bar's general settings.
- Check for apps that change your plan. SteamVR reportedly switches to High performance on launch. Re-check after launching such software.
- Verify parking while a game runs by opening Resource Monitor and looking at the CPU tab. An AMD forum guide suggests this check, but warns that cores may unpark shortly after you switch focus away from the game.
- Everyone else: switching back to Balanced probably won't change in-game frame rates. It should give you a cooler, quieter idle.
Bottom line
High performance isn't dangerous, but "set it and forget it" is outdated advice. It's mostly harmless on conventional CPUs. On dual-CCD X3D chips it can interfere with the game-aware core parking that the chips depend on. Check your plan, and confirm that parking works with a real game running.