Most PCs run RAM at its baseline JEDEC speed rather than its rated advertised speed. Enabling XMP or EXPO in the BIOS is the single highest-impact free performance upgrade for many systems. Beyond that, fine-tuning sub-timings and Windows memory settings can squeeze out additional performance.

Why RAM Speed Matters

RAM speed directly impacts:

CPU-bound gaming performance: AMD Ryzen is especially sensitive to memory bandwidth and latency
Integrated graphics (AMD APUs, Intel Arc, Intel Iris): iGPUs share system RAM and benefit enormously from faster memory
Video editing, 3D rendering: larger datasets benefit from higher bandwidth
General responsiveness: Windows kernel operations benefit from lower memory latency

Step 1: Enable XMP or EXPO

By default, DDR4 runs at 2133MHz and DDR5 at 4800MHz regardless of the kit’s rated speed. Enable the overclock profile in BIOS:

Restart and enter BIOS (usually Delete, F2, or F12 on boot)
Look for: XMP (Intel), EXPO (AMD), DOCP (ASUS), or EOCP (GIGABYTE for AMD) — all accomplish the same thing
Select the highest profile (usually XMP Profile 1 or EXPO Profile 1)
Save and exit

Your RAM will now run at its rated speed (e.g., DDR5-6000 instead of DDR5-4800).

Stability Testing After XMP

XMP profiles push memory beyond JEDEC specs — some systems need additional tuning for stability:

# Windows Memory Diagnostic (basic)
mdsched

# Better: MemTest86 (bootable USB, run for 2+ hours)
# Download from memtest86.com

# Fast stability check: HCI MemTest (Windows app)
# Run 200% coverage minimum

If crashes occur, try a lower XMP profile or increase DRAM voltage by 0.05V in BIOS.

Dual-Channel Configuration

Dual-channel mode doubles memory bandwidth by using two DIMM slots simultaneously. Check your motherboard manual — DIMMs usually need to be in slots A2 and B2 (not A1 and A2).

Verify dual-channel is active in CPU-Z:

Open CPU-Z → Memory tab
Channel: should show Dual (not Single)

Single-channel mode approximately halves memory bandwidth. For a system with only 2 slots, always fill both.

Checking Current Memory Speed

# Check RAM speed in Windows
Get-CimInstance -ClassName Win32_PhysicalMemory | Select-Object Speed, Capacity, Manufacturer

Or use CPU-Z → Memory tab for detailed info: frequency, timings, and channel mode.

Understanding Memory Timings

DDR4/DDR5 kits are rated with four primary timings: CL-tRCD-tRP-tRAS (e.g., 16-18-18-38 for DDR4-3200).

CL (CAS Latency): cycles between column address and data — lower is better
tRCD: time from row activation to column — lower is better
tRP: time to precharge between rows — lower is better
tRAS: minimum row active time — has complex interaction with other timings

Lower timings = lower latency. Higher speed = higher bandwidth. The two are in tension — faster speeds often require higher CL.

Memory Integrity (VBS/HVCI)

Windows 11 enables Memory Integrity (Hypervisor-Protected Code Integrity) by default on new installs. This adds a security layer but costs 5–10% GPU performance in some benchmarks due to increased VRAM-to-system-memory transfers.

To check/disable:

Windows Security → Device Security → Core isolation
Toggle Memory integrity Off
Restart

Note: this reduces security. Only disable on dedicated gaming machines not used for sensitive work.

AMD Ryzen: FCLK and Infinity Fabric

On AMD Ryzen systems, the Infinity Fabric interconnect runs at half the memory speed by default:

DDR4-3600 → FCLK 1800MHz (1:1 ratio — optimal)
DDR4-4000 → FCLK can’t keep up, drops to 2:1 ratio (adds latency)

For AMD systems, DDR4-3600 with tight timings is the sweet spot for Ryzen 5000. For Ryzen 7000 (DDR5), DDR5-6000 hits the FCLK sweet spot (FCLK 2000MHz).

In BIOS: set FCLK manually if using non-sweet-spot speeds.

Practical Upgrades by Use Case

System	RAM Recommendation
AMD Ryzen gaming	DDR4-3600 CL16 or DDR5-6000 CL30
Intel 13th/14th gen	DDR5-6400 to 7200 with XMP
AMD APU / iGPU	Fastest supported — bandwidth-critical
Budget system	At minimum, enable XMP — biggest free gain

Enabling XMP/EXPO is 5 minutes of BIOS work that consistently yields 5–15% improvement in memory-sensitive workloads with zero cost. Do it before any other optimization.