Top evaluation standard for Amd Infinity fabric
Top evaluation standard for Amd Infinity fabric
You're looking for a quick benchmark to detect Infinity fabric instability as performance issues.
Prime95 comes to mind, but focus on benchmark mode, not stress testing. Use configurations that simulate memory limits. Data size equals FFT size times 8, ensuring it exceeds CPU cache capacity so data loads into RAM. Not all FFT values are acceptable; try adjusting a small range to identify valid ones. You can accelerate the test by running just one worker (full-core task) and turning off HT. Set the run time to the shortest feasible duration—around 5 seconds. For many cores, matching workers to the number of cores may help. This approach works because RAM access is constrained by I/O, not just CPU speed. I should have considered this earlier. A 1B data size reaches up to 4.8GB, which means 3D cache won't improve performance much. I’m not aware of AMD FCLK issues; they don’t affect software stability. Neither causes serious slowdowns nor data corruption.
It’s a bit tricky to understand. There are some data issues present, but the main issue usually shows up as a big drop in performance when the clock speed fluctuates a lot. Y-Cruncher performs poorly in this case; on my 5900X with a stable 1900MHz, it finishes Y-Cruncher 1B in under half a minute, but at 1933MHz it takes over five minutes. Other tests behave differently—Time Spy keeps going until 1966MHz before slowing noticeably, and memory stress tests become much slower and eventually fail, unless it’s close to the maximum stable frequency. The FCLK has some error-correction features, so it tries to fix mistakes instead of just reporting them, but this makes it much slower and still leads to errors over time. Ryzen 7000 seems to handle it similarly, though the slowdowns are less extreme—Y-Cruncher 1B might only add a few seconds, not minutes. Unless the system is extremely unstable, it won’t throw WHEA warnings anymore, so you can’t rely on those for checking stability.
In premium modern consumer tiers Y-cruncher 1B experiences bandwidth constraints due to high-end systems. This likely impacts its FCLK performance. Based on the observed behavior, it may employ a retry strategy instead of repeatedly sending data for fixes, which can significantly reduce effective throughput when errors occur often.
Check the overall computation duration. A more stable IF typically results in a faster outcome compared to a less stable one.
Higher stable FCLK performs better than lower stability, though the gap is often small. Under similar conditions I wouldn't anticipate more than a couple of seconds difference between 1600MHz and 1900MHz for the clock speed. For real-time checks, tools like HWiNFO or ZenTimings can show actual performance.
Restore Ryzen Master settings back to the standard configuration, then attempt to operate solely through the BIOS interface without engaging Ryzen Master at all.
I attempted that, but it didn't alter the state I left in Ryzen Master.