I bought a new PC and tuned three BIOS settings that cut heat and boosted performance instantly

Published Sep 9, 2026, 7:00 AM EDT His love of PCs and their components was born out of trying to squeeze every ounce of performance out of the family computer. Tinkering with his own build at age 10 turned into building PCs for friends and family, fostering a passion that would ultimately take shape as a career path. Besides being the first call for tech support for those close to him, Ty is a computer science student, with his focus being cloud computing and networking. He also competed in semi-pro Counter-Strike for 8 years, making him intimately familiar with everything to do with peripherals. I've built enough systems in my life to know how the first hour after assembly usually goes. Get a successful POST, check immediate temps, BIOS update, and then immediately back into the BIOS to undo a lot of the defaults. Out of the box, a motherboard runs your memory below its rated speed, feeds your CPU more voltage than it needs, and pushes more current through the I/O die than the memory controller has any use for, and while these things help make sure your system boots on the first go, they're not great settings to run all the time. Your memory isn't running at the speed you paid for XMP/EXPO should be enabled right away Credit: Source: XDA Despite the fact that I bought a DDR5-6000 kit and installed it correctly, the motherboard runs it at well below that, and it'll do the same for your system as well. Without an EXPO profile applied, the modules fall back to a conservative JEDEC speed that every AM5 CPU is guaranteed to handle. For Intel, these profiles are called XMP, and are functionally the same. It's the single largest performance toggle available to you on a fresh system. DDR5-6000 is the number worth aiming for on X3D AM5 systems specifically, and there's a good reason for that. At 6000 MT/s, the memory controller clock and the memory clock run locked in a 1:1 ratio, and pushing above that means the controller can't sustain the frequency required, so it drops to a 1:2 ratio, where the memory runs at twice the controller's speed. Faster kits are still faster, but you spend the gain on the ratio change unless you go considerably higher and tune manually. Here's the part that isn't discussed about EXPO and XMP: these profiles aren't a promise of functionality. It's a rated profile, and stability at those speeds depends entirely on your specific memory controller. Above the platform's comfort zone, it stops being deterministic, especially with high capacities. At the 64GB capacity that I run, EXPO applied cleanly on my 7800X3D, the system passed every stress test I pointed at it, but games would still crash routinely. I had to step things down to 5600MT/s, which is still a lot better than JEDEC speeds. Enabling EXPO also lengthens boot times, sometimes dramatically, because the board retrains the memory on startup. The usual fix is Memory Context Restore, which lets the firmware reuse previously trained parameters instead of deriving them again. It works, but it causes marginal memory configurations to kick up more of a fuss because you're removing the constant recalibration that happens every boot. Curve optimizer buys speed by taking voltage away It's counter-intuitive, but it works Every Ryzen chip ships with a voltage-frequency curve that's padded to accommodate the worst silicon that passed binning. Your particular sample almost certainly needs less voltage than it's being handed, and a negative Curve Optimizer offset is how you claw some of that back. Less voltage means less power, less power means less heat, and on a chip that's limited thermally rather than by power limits, that reclaimed headroom gets spent on higher sustained boost clocks, which means better performance while running cooler. It almost sounds too good to be true, but there is a bit of a catch, which I'll get to. It's one of the only meaningful tuning methods available on the Zen 4 X3D chips. AMD doesn't permit conventional frequency overclocking on these CPUs because the stacked cache is sensitive to voltage, so the setting that reduces stress is the same setting that improves performance. This was changed with Zen 5, which moved the sensitive cache to the bottom of the CCD rather than the top, which made it easier to cool. One warning that catches people out: instability from an aggressive offset shows up under light load, not heavy load. Undervolting the low-load portion of the curve produces desktop crashes, browser crashes, and idle reboots while something like Cinebench can run clean for an hour. If you only validate under full load, you'll ship an unstable machine and blame your RAM for it. Your SoC voltage is probably higher than necessary Turning it down is recommended Left on Auto, boards have historically pushed SoC voltage (labeled VDDCR_SOC, VSOC, or CPU SoC Voltage depending on the vendor) well past what the memory controller actually requires, particularly once EXPO is enabled. This was well documented when Ryzen 7000 chips began to fail, and AMD distributed firmware fixes that stopped AM5 power rails from going beyond their limits, which included a hard cap on the SoC to keep it below 1.3V. Modern firmware enforces that ceiling, so the danger of a runaway value is largely gone. What hasn't gone is the gap between the ceiling and what your system actually needs. Auto tends to aim high because margin costs the vendor nothing and ensures compatibility, and there's no reward for sitting at a limit your system doesn't need to use. Setting it manually is worth doing because the rail is doing real work, but you want it doing the minimum amount to minimize current and heat wear-and-tear. SoC voltage powers the I/O die, which houses the memory controller, so it isn't a value you can simply drive into the floor. Doing so makes your memory less stable, so what you're looking for is the value that passes testing at the speed you settled on for EXPO. The defaults act as insurance They exist for a reason It's easy to read all of this and conclude that vendors should be better about tuning their boards before they release them, but that's not quite right. A motherboard manufacturer writes one BIOS for millions of machines it will never see. It doesn't know your CPU sample, your memory kit, your cooler, or your power supply. A default that's fast on eighty percent of systems and dead on the remaining twenty generates RMA requests, returns, support tickets, and reviews that cost them very real money, so the defaults are actually very crucial. Not everyone is an enthusiast who can tweak a BIOS to get a dead system up and running; in fact, the majority of people just plug their computer in and expect it to work, and that's why things like JEDEC and conservative voltage values exist.