Manual DDR5 Subtiming Tuning: Ryzen Gaming Gains, FCLK Caveat and Safe Windows Testing

Yet beneath those headline transfer rates and primary timings sits a labyrinth of secondary and tertiary subtimings that motherboard auto-rules and factory profiles frequently leave loose. A recent 16-hour hands-on tuning experiment by How-To Geek senior author Ismar Hrnjicevic puts a spotlight on both sides of manual DDR5 optimization: the measurable uplift available in CPU-sensitive workloads, and the grueling cycle of failed boots, desktop crashes, and CMOS resets waiting for anyone who rushes the process.

Before you dive into your UEFI menu with dreams of smoother frame times, it pays to examine what manual subtiming tuning actually changes, what the testing on an AMD Ryzen 7 7700 demonstrated, why the benchmark methodology carries an important caveat, and how to approach memory tuning on a Windows PC without corrupting your operating system.

Beyond XMP and EXPO: Why Subtimings Matter​

When you install a new DDR5 kit, it defaults to standardized JEDEC specifications designed to guarantee that the system boots on virtually any compatible motherboard and processor combination. In Hrnjicevic’s test system—built around an overclocked AMD Ryzen 7 7700 and a 32GB Crucial DDR5 Overclocking kit supporting both EXPO and XMP—that JEDEC baseline ran at 5600 MT/s with primary timings of CL46-45-45-45-90.

Enabling the kit’s factory EXPO 1 profile raised the transfer rate to 6000 MT/s and tightened the primary timings to CL36-38-38-38-80. As Intel’s official RAM overclocking guidance explains, memory performance is a constant tug-of-war between frequency and latency: higher clock speeds move more data per second, while memory timings dictate how many clock cycles the system must wait between specific memory operations. Whenever a processor needs data that is not already sitting inside its high-speed onboard cache, it has to wait on system RAM. Reducing those cycle delays cuts down the time the CPU spends stalling.

Factory EXPO and XMP profiles are engineered for broad compatibility across thousands of motherboards and varying processor memory controllers. Because memory vendors want their one-click profiles to boot reliably without support tickets, they primarily adjust frequency and the first handful of primary timings printed on the sticker. Dozens of secondary and tertiary timings are often left untouched or governed by conservative motherboard auto-settings.

Key Timings Adjusted in the 16-Hour Test​

In his manual tuning session, Hrnjicevic kept his Crucial kit at its 6000 MT/s EXPO 1 transfer rate and left almost all primary timings alone, tightening only tRAS among the primaries while focusing on secondary timings. According to his report and corroborated by AMD’s Ryzen Master documentation for DRAM timing controls, the targeted parameters included:

• tRAS (Row Active Time): The minimum number of clock cycles a memory row must remain open for data transfers before it can be closed.
• tRC (Row Cycle Time): The total time required for a complete row cycle, encompassing both activating and precharging a row.
• tRRDS and tRRDL (Activate to Activate Delay): The delay required between activating two different rows in different bank groups (tRRDS, short) or within the same bank group (tRRDL, long).
• tFAW (Four Activate Window): A rolling window that limits when more than four row activations can occur on a memory rank.
• tRFC (Refresh Cycle Time): The duration a memory rank is unavailable while capacitors are refreshed to retain data.
• tREFI (Refresh Interval): Unlike the timings above—which tuners reduce to lower latency—tREFI controls how long the memory waits between refresh cycles. Increasing tREFI keeps the RAM available for actual work longer, though pushing it too high makes the modules vulnerable to heat-induced bit errors.

Section Summary: While XMP and EXPO profiles boost transfer rates and primary timings over default JEDEC speeds, they frequently leave secondary and tertiary timings conservative to ensure broad stability. Manually tuning parameters such as tRC, tRFC, tFAW, and tREFI aims to shrink the idle wait states between CPU requests and memory responses.

Inside the Ryzen 7 7700 Benchmarks—and the Hidden Confounder​

To evaluate whether 16 hours of BIOS tweaking yielded real-world dividends, Hrnjicevic ran four tests across three configurations: the 5600 MT/s JEDEC baseline, the stock 6000 MT/s EXPO 1 profile, and the manually tuned 6000 MT/s EXPO 1 profile. His test suite covered:

1. Intel Memory Latency Checker (MLC): To measure raw memory latency and bandwidth.
2. 7-Zip Built-in Benchmark: To gauge CPU compression and decompression throughput, which responds strongly to memory subsystem speed.
3. Shadow of the Tomb Raider: Using the built-in benchmark frequently favored for CPU and memory scaling tests.
4. Forza Horizon 6: Tested using the author’s standard everyday graphics settings, evaluating both actual gameplay framerates and the game’s built-in CPU simulation benchmark.

In the published text of the How-To Geek report, the raw numerical tables for Intel Memory Latency Checker, 7-Zip, and Shadow of the Tomb Raider were omitted where chart placeholders appeared, leaving only the author’s narrative takeaways and two specific Forza Horizon 6 figures. Hrnjicevic reported an 11-FPS average difference in Forza Horizon 6 alongside an even wider spread in CPU simulation when illustrating the impact of faster RAM and lower latencies, and specifically highlighted an approximate 4% increase in both average FPS and 1% lows in Forza Horizon 6's CPU simulation when moving from standard EXPO 1 to his custom-tuned configuration.

The Analytical Catch: Infinity Fabric Wasn't Held Constant​

From a strict hardware-testing standpoint, there is a critical variable in that third test run that every reader should note. When benchmarking the manually tuned subtimings against the standard EXPO 1 profile, Hrnjicevic also raised his Ryzen 7 7700’s Infinity Fabric clock (FCLK) from its stock 2000 MHz to 2067 MHz.

On AMD Zen 4 processors like the Ryzen 7 7700, the Infinity Fabric is the internal interconnect that links the Core Complex Die (CCD) to the I/O Die (IOD), where the memory controller lives. Boosting FCLK from 2000 MHz to 2067 MHz directly increases interconnect bandwidth and lowers communication latency between the CPU cores and the memory controller. Because the subtiming tweaks and the 2067 MHz fabric overclock were tested together, the 4% uplift in Forza Horizon 6 CPU simulation reflects the combined gain of tighter DRAM subtimings plus a faster Infinity Fabric—not the memory timings in isolation.

How Does This Compare to AMD’s EXPO Ultra Low Latency (ULL)?​

Hrnjicevic noted that his ~4% gain over standard EXPO mirrored the exact percentage AMD touted for its newer EXPO Ultra Low Latency (EXPO ULL) memory specification.

As reported by Tom’s Hardware’s Jake Roach following Computex 2026, AMD vice president and general manager of Ryzen and Radeon David McAfee explained that EXPO ULL works by adding extended subtimings directly into the memory module’s SPD profile to squeeze lower latency out of compatible DDR5 kits. In AMD’s internal testing across 30 games using a Ryzen 7 9700X and DDR5-6000 memory, the company claimed EXPO ULL delivered an average 4% gaming uplift (and 4% higher 1% lows) over standard EXPO, and a 13% average uplift (15% in 1% lows) over JEDEC DDR5-5600.

While the parallel is instructive—both approaches target the exact same secondary and tertiary timing slack left behind by standard EXPO—AMD’s figures come from a 30-game average on a newer Zen 5 Ryzen 7 9700X using factory-binned DIMMs from partners such as G.Skill, Kingston, Lexar, XPG, and TeamGroup. Manual tuning on an existing Crucial kit with a Zen 4 Ryzen 7 7700 is a DIY path toward the same engineering concept, though without the convenience of a pre-validated SPD profile.

Section Summary: Manual subtiming adjustments combined with a 67 MHz Infinity Fabric bump yielded roughly a 4% gain in Forza Horizon 6 CPU simulation averages and 1% lows over standard EXPO. While that matches AMD’s stated 4% average gain for factory EXPO ULL kits, the fabric clock change means the manual test did not isolate RAM timings alone.

Does Your PC Actually Need Manual RAM Tuning?​

Before you block off an entire weekend to stare at BIOS screens, ask yourself a simple question: is your system actually waiting on memory?

The performance impact of RAM tuning varies dramatically depending on your hardware architecture and display resolution:

• AMD X3D Processors (Ryzen 7 7800X3D, 9800X3D, etc.): As Hrnjicevic rightly points out, AMD processors equipped with 3D V-Cache are far less sensitive to memory latency. Because their massive stacked L3 cache holds a much larger pool of game data right on the processor package, the CPU suffers fewer cache misses and accesses system RAM less frequently.
• Non-X3D AMD Ryzen and Intel Core Processors: Standard desktop chips without expanded L3 cache stacks rely much more heavily on main memory whenever the local cache fills up. In CPU-bound games—such as competitive shooters, strategy simulations, or racing games at high frame rates—tighter subtimings and faster interconnect speeds show up most clearly in 1% lows and frame-time consistency.
• 4K and Heavy GPU Bottlenecks: If you play demanding single-player titles at 4K with maxed-out ray tracing, your graphics card is almost certainly the bottleneck. Shaving a few nanoseconds off memory latency will not rescue a GPU that is fighting for every frame.

Lessons from a 16-Hour Ordeal: How to Tune Safely on Windows​

Why did tuning one memory kit take 16 hours? Because Hrnjicevic initially broke the cardinal rule of hardware overclocking: he changed multiple variables at the same time.

At the start of his experiment, he pushed his Infinity Fabric clock all the way to 2167 MHz while simultaneously tightening multiple memory timings. Because Windows booted to the desktop, he assumed the 2167 MHz fabric clock was rock-solid and blamed every subsequent crash, boot failure, and non-POST event on whichever RAM timing he had just touched.

He ran into a second delayed trap when raising tREFI to 50,000. Because tREFI extends the window between capacitor refreshes, a setting that passes a brief cold test can start throwing silent memory errors an hour or two later once the DIMMs heat up inside a closed PC case—especially when a graphics card is dumping hot air directly across the motherboard.

On a Windows 11 system, unstable RAM is not merely an annoyance that causes a blue screen or game crash. Memory errors during active read/write operations can corrupt Windows system files, damage the registry hive, or scramble application data written back to your NVMe drive.

A Disciplined Step-by-Step Methodology for Memory Tuning​

If you decide to explore manual subtimings on your own rig, follow a methodical workflow to protect both your sanity and your Windows installation:

1. Document Your Known-Good Baseline: Before changing a single value, save your working BIOS profile to an internal slot or a USB flash drive. Write down your baseline timings and voltages.
2. Locate Your CMOS Reset Mechanism: Know exactly where your motherboard’s Clear CMOS jumper or rear I/O reset button is located before you start. When a memory timing is too tight to pass power-on self-test (POST), clearing the CMOS—or relying on features like AMD’s DDR5 Robust Training Mode—will be required to get back into the firmware.
3. Isolate Interconnect Clocks First: If you plan to adjust Infinity Fabric (FCLK) on AMD Ryzen, test and validate that clock speed completely on its own using stock or standard EXPO memory timings before touching a single secondary timing.
4. Adjust One Timing Group at a Time: Never tighten tRFC, tFAW, and tREFI in the same reboot. Adjust one parameter incrementally, reboot, and test.
5. Use Layered Stress Testing: Hrnjicevic ultimately stabilized his rig by running a 15-minute pass of TestMem5 (using the community Absolut @ anta777 configuration) after each individual timing adjustment, followed by multi-hour marathon sessions in both TestMem5 and y-cruncher (VT3 test) once all timings were dialed in. Keep in mind that a 15-minute pass is only a quick sanity filter to catch immediate instability—it is never proof that a system is fully stable.
6. Test Under Thermal Load: Especially when increasing tREFI or lowering tRFC, monitor your DIMM temperatures if your modules have onboard thermal sensors, and run long duration stress tests that allow the interior of your PC case to reach real-world gaming temperatures.
7. Verify Windows Integrity afterward: After any session that involved crashes or blue screens during boot, open an elevated Command Prompt or PowerShell window in Windows and run DISM /Online /Cleanup-Image /RestoreHealth followed by sfc /scannow to ensure no operating system files were corrupted during failed test runs.

Section Summary: Changing multiple timings alongside fabric clocks caused dozens of failed boots and misdiagnosed crashes during the 16-hour test. Safe RAM tuning requires isolating one variable at a time, respecting temperature-sensitive timings like tREFI, and validating final settings with multi-hour stress tests in tools like TestMem5 and y-cruncher.

The Bottom Line for Windows Enthusiasts​

There is an undeniable satisfaction in extracting extra performance from hardware you already own—particularly when memory pricing and supply fluctuations make buying a faster factory-binned kit unappealing. Pushing a DDR5 kit from JEDEC defaults to EXPO or XMP remains the single biggest, easiest win any PC builder can claim in thirty seconds.

Going beyond EXPO into the weeds of manual secondary and tertiary timings can deliver real, measurable ~4% gains in CPU-bound gaming scenarios and 1% lows, much like AMD’s newer EXPO ULL specification promises out of the box. Just go in with clear eyes: those last few percentage points are not free. You pay for them in hours of methodical testing, thermal vigilance, and the patience required to tune one clock cycle at a time.