DDR5 Memory Subtimings Tuning: tRFC, tREFI and Read-to-Write Turnaround: Advanced Implementation and Real-World Architecture

[Luke Taylor]

I’ve been running tREFI overclocking to 65535 on my 2x24GB DDR5-6000 kit for the past few weeks, chasing those sub-millisecond refresh stalls that are apparently eating my 1% low FPS in competitive shooters. The thing is, I’m seeing wildly different results depending on whether I’m testing with a static benchmark or actual in-game frames—my 1% lows seem to stabilize around 280-290 FPS in a 320+ FPS environment, but I can’t tell if that’s the refresh cycle or something in the engine’s frame pacing. For those of you pushing high refresh rates, how are you isolating memory-induced stalls from GPU bottlenecks? Do you rely on CapFrameX or your own custom logging to catch the specific frame-time spikes tied to refresh intervals?

On the read-to-write turnaround side, I’m noticing that tightening RAS-to-RAS and the R/W turnaround timings helps in single-threaded titles but occasionally destabilizes multi-core workloads during streaming. Has anyone found a sweet spot where the subtimings stay stable without having to push VDD/VDDQ voltages into risky territory? And more importantly, how do you benchmark the real-world impact—do raw synthetic memory tests actually correlate with the 1% low improvements you see in-game, or are you measuring frame-time variance directly from gameplay captures?

🎮 Battlestation & Rig: i5-13600K | RTX 4070 Super | 32GB DDR5 6000 CL32 | DeepCool LT720 | LG UltraGear 27GR95QE-B 27″ OLED 240Hz | Moza R9 V2 Direct Drive + Moza FSR Formula Wheel + CRP Load Cell Pedals | Sennheiser HD 560S Open-Back + Creative Sound BlasterX G6 DAC

📍 San Jose, CA (USA)

"144Hz is baseline, 360Hz is reflex, 540Hz is pure instinct."

[Sam Gray]

Great topic to open the thread with. Let me lay out how I approach these three subtimings on my DDR5-6000 kit, because they’re often misunderstood as interchangeable when they actually serve very different purposes.

tRFC (tRFC4) is your refresh command time — the time the controller waits during a refresh cycle. Lowering it from the JEDEC default (often ~360+ nC) can yield real bandwidth and latency gains since you’re spending less idle time refreshing. I tune using the SBE (Single Bit Error) mode via TestMem5 patterns, dropping tRFC4 in ~16-32 nC steps until I hit an error, then backing off ~16 nC for a stable daily. On modern 24Gb chips you can often get tRFC4 down to ~250-280 range.

🎮 Battlestation & Rig: Ryzen 7 9800X3D | RX 7900 XTX Taichi | 32GB DDR5 6000 CL28 | Corsair 5000D | ASUS ROG Swift OLED PG27AQDM 27″ 240Hz 0.03ms | Logitech G Cloud + Razer DeathAdder V3 Pro + Endgame Gear EM-C Pad | Sennheiser HD 560S Open-Back + Creative Sound BlasterX G6 DAC

📍 Sydney (Australia)

"0.03ms OLED response renders ghosting completely obsolete."

[James Hall]

On tRFC/tREFI, the key thing most people miss is that you’re trading refresh headroom for raw bandwidth?

🎮 Battlestation & Rig: Ryzen 7 9800X3D | RX 7900 XTX Taichi | 32GB DDR5 6000 CL28 | Corsair 5000D | ASUS ROG Swift Pro PG248QP 24.1″ 540Hz Esport-TN | Custom Tofu60 Lubed Gateron Oil Kings + Lamzu Maya 4K + Artisan Hien Mid | Beyerdynamic DT 770 Pro (80 Ohm) + Focusrite Scarlett Solo Gen 4

📍 Copenhagen (Denmark)

"Rapid trigger active, zero deadzone; never miss a counter-strafe."

[Luke Taylor]

Alright guys, it actually worked. After digging through your threads I went into the BIOS and started messing with the memory subtimings — I bumped tREFI up to 65535, tightened tRFC down to a more aggressive value, and fixed that read-to-write turnaround (RWT/WTS) that half of you kept pointing at. First boot was a little dicey and I had to clear CMOS once, but now it’s stable after 30 minutes of OCCT. The real payoff is what you all said — the random micro-stutters and framedrops in competitive shooters are completely gone, and my 1% lows are way more consistent than before. Big thanks to everyone who laid out the tuning order; this board really does respond well once you stop trusting the auto XMP defaults. I’ll drop a full timing breakdown once I dial in more stability testing, but for now it’s locked and running clean.

🎮 Battlestation & Rig: i5-13600K | RTX 4070 Super | 32GB DDR5 6000 CL32 | DeepCool LT720 | LG UltraGear 27GR95QE-B 27″ OLED 240Hz | Moza R9 V2 Direct Drive + Moza FSR Formula Wheel + CRP Load Cell Pedals | Sennheiser HD 560S Open-Back + Creative Sound BlasterX G6 DAC

📍 San Jose, CA (USA)

"144Hz is baseline, 360Hz is reflex, 540Hz is pure instinct."

[Zack Peterson]

Great topic. On tRFC for DDR5, the key thing everyone forgets is that refresh is

[Author]

Posts