XMP
Intel's Extreme Memory Profile, a preset stored on DDR memory modules that lets the BIOS run RAM at its advertised speed and timings instead of conservative JEDEC defaults.
XMP (Extreme Memory Profile) is a one-click memory overclock baked into the SPD chip of most enthusiast DDR4 and DDR5 kits. Without it enabled in BIOS, your RAM runs at slow JEDEC fallback speeds — often 30-40% below what you paid for — regardless of what the sticker on the box says.
How XMP Actually Works
When you toggle XMP in BIOS, the motherboard reads a preset profile from the memory module and applies the advertised frequency, timings (CL, tRCD, tRP, tRAS), and voltage in one shot. It's Intel's spec, but most AMD boards also read XMP profiles — though AMD's native equivalent is EXPO. Out of the box, a DDR5-6000 kit will boot at DDR5-4800 until XMP is on. This is the single most common "my new build feels slow" mistake.
Why It's Not a Free Lunch
XMP is technically an overclock, so it sits outside the CPU's guaranteed memory controller spec. On most modern platforms it's stable, but pushing high-speed kits (DDR5-7200+) on boards with four DIMM slots populated can cause boot failures or instability. Higher memory bandwidth also means more heat and slightly more power draw. For a local AI rig, the stability tradeoff matters — a crash mid-inference on a 70B model that took five minutes to load is painful.
Why It Matters for Local AI
For LLM inference, XMP only moves the needle when you're doing VRAM offloading — pushing layers into system RAM because the model doesn't fit on your GPU. In that regime, system memory bandwidth is the bottleneck, and the gap between JEDEC and XMP speeds translates directly into tokens per second. If your model fits entirely in VRAM, XMP is essentially free performance you should still enable, but it won't change inference speed in any measurable way.