Memory Sub-Timings and Minimum FPS in Hardware Benchmarks

This article explores how tuning specific memory sub-timings—specifically tRFC, tFAW, and tWR—directly improves minimum frames per second (FPS) in computer hardware benchmarks. While primary RAM timings and clock speeds get the most attention, secondary and tertiary sub-timings dictate how efficiently your memory handles rapid data requests. Optimizing these values reduces system latency, eliminates micro-stutters, and significantly boosts 1% and 0.1% low FPS metrics during CPU-intensive workloads.

The Connection Between RAM and Minimum FPS

In hardware benchmarking, average FPS primarily measures GPU capability, while minimum FPS (measured as 1% and 0.1% lows) reveals CPU and memory bottlenecks. A sudden drop in FPS usually occurs when the CPU experiences a "cache miss" and must fetch data directly from the system RAM.

Primary timings like CAS Latency (CL) only control the initial delay in a data request. Sub-timings control the internal operations and housekeeping of the RAM modules. Tightening (lowering) these sub-timings reduces the idle time of the memory controller, feeding data to the CPU faster and smoothing out severe frame drops.

tRFC (Row Refresh Cycle Time)

RAM stores data in microscopic capacitors that slowly lose their electrical charge. To prevent data loss, the memory must periodically pause operations and refresh these cells. tRFC dictates the exact number of clock cycles required to complete this refresh process.

  • The Bottleneck: While a refresh is occurring, the CPU cannot read from or write to that specific memory bank. If the CPU requests game data during a refresh cycle, it has to wait, causing a frame time spike.
  • The Tuning Impact: Stock XMP/EXPO profiles set tRFC very high to ensure stability across low-quality silicon. Manually lowering tRFC drastically reduces the "dead time" the CPU spends waiting for the RAM to become available. Among all sub-timings, tuning tRFC yields the most substantial improvements to 1% low FPS.

tFAW (Four Activate Window)

To prevent power spikes on the memory module, RAM is restricted to a maximum of four row activations within a specific rolling timeframe. tFAW defines the length of this timeframe in clock cycles.

  • The Bottleneck: Even if your RAM is fast enough to open rows quickly, a high tFAW value will force the memory controller to sit idle after the fourth activation until the window expires.
  • The Tuning Impact: Lowering tFAW removes this artificial throttle and maximizes command throughput. The optimal value for tFAW is typically four times the value of tRRD_S (Row to Row Delay), often bottoming out at 16 cycles. In benchmarks, a tight tFAW prevents data queues from backing up during chaotic, asset-heavy gaming scenes, directly raising the minimum FPS floor.

tWR (Write Recovery Time)

tWR specifies the number of clock cycles that must elapse between a valid write operation and the precharging (closing) of that memory row. It acts as a safety buffer to guarantee data is accurately written before the row is closed and another is opened.

  • The Bottleneck: Benchmark scenarios constantly alternate between reading and writing data. A loose (high) tWR delays the closing of a row, which consequently delays the opening of the next row needed by the CPU.
  • The Tuning Impact: Tightening tWR accelerates the transition from writing data to reading new data. This reduction in turnaround latency minimizes micro-stutters in games that constantly stream new data into memory, directly lifting 0.1% low FPS metrics.

Summary of Benchmark Results

When benchmarking computer hardware, leaving tRFC, tFAW, and tWR on automatic settings leaves significant performance on the table. While tuning these three sub-timings may only increase average FPS by 2% to 5%, it frequently improves 1% and 0.1% low FPS by 10% to 20% in CPU-bound scenarios. By minimizing refresh delays, removing activation bottlenecks, and speeding up write recovery, tuned sub-timings ensure the CPU is never starved for data, resulting in flat, consistent frame time graphs.