Make sure your UCLK is correctly set!
Make sure your UCLK is correctly set!
I only recently realized details about my RAM setup after spending many hours testing. Now it might need some adjustments in parts. System: 7800X3D "stock" with two 32GB 6000C30 RAM modules. My motherboard is B650E-F, and I installed the latest but not the newest BIOS since they updated a few versions after Zen 5. When I assembled the system, I installed the RAM, enabled EXPO, and everything functioned properly! Later I discovered MCR and activated it too. Everything seemed fine. Because I bought from Asus, I used EXPO II which adjusts settings according to what the RAM expects, not just what the manufacturer suggests. I planned to test my configuration with RAM at 6000C30 while EXPO is off, expecting a drop to 4800C40. The only change was turning EXPO off. I wondered if it would help record other clock speeds. Then I noticed something. 6000C30: FCLK 2000, UCLK 1500 at 4800C40: FCLK 1800, UCLK 2400 I hadn’t focused much on FCLK or UCLK before. With EXPO on, it changed to a 2:1 ratio? Is this typical for a motherboard? I thought I’d try 1:1 and compare performance differences. What are the usual safe limits for UCLK (without manual tweaks) on Zen 4? I want something straightforward and stable, avoiding anything that needs precise voltage control. Are there other factors I should keep in mind?
Yes, 6000MHz with a 1:1 clock is commonly recommended for optimal performance.
I initially assumed the setup was secure, only to realize it wasn’t running as expected. Was the game choosing its own path by skipping the automatic adjustment? Or did I overlook a manual step I needed to complete? Later, I discovered the option buried deep within multiple submenus under the overclocking menu in BIOS. After restarting, the system performed well with no crashes so far—stable enough for now. I proceeded with my Prime95 test, which showed a 4% boost over 1500 UCLK.
I'm testing performance metrics on my system right now. It looks like everything is functioning well so far, though formal stability checks are still pending. Prime95 and Y-cruncher reported no issues, which is a positive beginning.
Initial trial outcomes show consistent performance across various tools. Testing on 6000C30 UCLK 1500 and partial runs on UCLK 3000 yielded stable results. Blender 4.2.0 recorded a modest improvement of +0.6% across three trials. Junkshop subtest achieved the highest score at +1.1%. Prime95 8192k 8c8w improved by +4.1%, a solid gain (three runs each). Pov-ray 3.7 added +0.2%, while Y-cruncher Pi 5B +0.7% and Pi 1B +0.4% showed incremental progress in single runs versus best of three. I'm taking a break from heavy benchmarking today—just a few small wins so far. Let me know if it's worth the effort.
You recently adjusted the speed to 5800MT/s with the ratio enabled, but it's showing a 1:1 output. It seems 6000 reaches a 2:1 ratio.
It's interesting about how you handle testing. Are you sticking to uniform timing settings across all speeds, focusing mainly on MT/s variations, or are you aiming to optimize each speed individually?
I plan to expand my report later because my goal is to evaluate rank effects with DDR5. This comes from mixed feedback and a lack of direct measurements. The initial test involved powering RAM on, enabling EXPO, and stopping it. For another measurement point, I considered disabling EXPO, which then runs at 4800C40. I don’t care about exact timings; it’s just for comparison. With this new insight, I’ve been running UCLK at 1500 for a while, wanting to see what performance loss would occur if everything ran in parallel. I’m also curious about the impact of full 1:1 operation, which led to partial UCLK 3000 testing so far. I’ll finish it later as I need a pause from repeatedly adjusting benchmark settings. It seems there are minor improvements possible, though they likely won’t be noticeable outside Prime95. Even with Prime95, a 4% gain appears questionable. The 5800 boot was simply to observe the chipset’s ratio selection. I haven’t tested that part at all—I’m just checking if it matches what I expect. I’ll receive a new “same” kit of 6000C30 RAM (2x16GB) and will continue testing, possibly trimming the workload selectively.