The new OC performs well overall but struggles with synthetic tasks and doesn’t meet expectations. Why?
The new OC performs well overall but struggles with synthetic tasks and doesn’t meet expectations. Why?
I understand this might interrupt your focus, but I’m here to give you quick, sharp advice if you have a spare moment. @WereCat @RONOTHAN## and others are great, so I’ll keep it concise. I switched my RAM overclock because the BIOS update affected settings and I wanted to fix the motherboard limits. I’m using 2x8GB Team T-Create Expert 3200C16 chips (SK Hynix 8Gbit, 1R A2), originally from AGESA 1.2.0.7, now running AGESA 1.2.0.A. The issue is clear: the new overclock is underperforming across all metrics. At 3333C14 it reaches about 90% of its theoretical speed at 48GBps, while 3533C16 sits at 82%. Even with similar speeds, latency jumps from 80.5 to 83.3. I’m hoping for around 50-52GBps and compensating the 3ns penalty. Testing was solid—no failures in stress cycles at 23°C with strong airflow. Should I aim for 3200C12 instead? What steps missed during this process? CPU voltages are set, PBO is active at a high scalar, and cooling is solid. GPU remains unchanged. No overclocking yet; I’ll revisit later if needed.
Based on what you shared, it seems the Power Delivery wasn’t quite suitable for your system, making it hard to assess without more details. You mentioned missing information about your hardware setup, which makes it difficult to pinpoint the exact cause. Could you provide a HWINFO64 log that outlines all the system states? That way, we can better understand what’s different from expected results.
That's correct, AM4 doesn't monitor RAM voltage. You can check PD settings in drive strength and procODT. Sharing HWiNFO64 logs isn't very useful since RAM temperature isn't tracked on either system. The ambient temperature around the motherboard stays steady at about 35 degrees.
It appears we're dealing with a TeamGroup RAM setup. Switching the RAM could provide insights from the experts. However, be cautious—real-world performance might differ.
Aim for the peak frequency you can reach; current CPUs like 5000+ DDR4 support this, and VCore settings can be adjusted further than usual. Regular Ryzen models often allow higher voltages, so even a 1.45V VCore works well. Focus on performance gains rather than just low clocks, as timing also affects latency. It’s better to chase stable high frequencies first before tweaking components. Frequency improvements matter more than just clock speed when it comes to responsiveness.
Most GPU drivers push VCore voltages higher because they’re tied to GPU settings. An IGUI guide can help you understand this. Frequency above timings gives the biggest performance boost, but chasing ultra-low clocks isn’t enough—frequency also influences latency. I usually target frequencies until they stop changing, then fine-tune only if needed.
For GPU tuning, fully relax primary voltages and push as high as possible, using whatever voltage is safe. Don’t make tiny adjustments until you’re near the limit; otherwise subs may underperform. At very high frequencies, you’ll hit scaling caps—gradually reduce frequency a bit to stabilize performance.
If you want solid results without overcomplicating, aim for around 2800 MHz with tighter CPUs. This is near the edge of what your components can handle. Once you reach that point, you’ll need to relax frequencies slightly and tighten the hardware for better stability.
This approach maximizes your RAM’s potential, delivering the best performance without unnecessary risk. The effort is worth it for noticeable gains.
Intel MLC appears faulty with AMD processors. Consider using MorePowerTool to adjust power limits as needed. Initially, I’d like to see better, more consistent benchmarks from AMD, Y-Cruncher, and PYPrime. If those drop in performance, it would indicate an issue.
Another obstacle during adjustments could be that certain parameters might return to their default values if they fall outside acceptable limits or become unstable. This issue might not be obvious while using software (like HWINFO64 or AIDA64) but is likely present at the hardware level. Additionally, some configurations are made in the BIOS in a specific manner, yet the system may revert to a manufacturer-set preset instead of your manual choices. I’ll also note that to truly access the maximum performance settings, using a thermal camera (which can be costly) might help—since it’s likely the RAM chips and memory power delivery units become excessively hot, reducing efficiency. For optimal overclocking, ensure the power supply provides at least 60% of the required wattage under load, without introducing voltage fluctuations.
I wouldn't prefer being assisted by chatGPT, respectfully. However, let's discuss why this approach is problematic for forum tech support—especially when it comes to issues like RAM speed and limitations. The main concerns revolve around accuracy and practicality.
Regarding RAM, I've found that voltage settings and timings matter a lot. My target was 1.4V, but pushing it higher (like 1.45V) could lock the system down. The IMC in my Ryzen 5 5500 is quite capable, handling speeds above 4000V, though it might need more tuning. If timing isn't precise enough, performance will suffer due to slower data movement.
It's also important to note that the board I'm using isn't optimized for XMP/DOCP profiles, which can affect stability. The QVL (Quality of Voltage) is less critical here since I'm not running memory tests yet. My current RAM model isn't meant for high-end gaming or heavy multitasking, so this isn't the best fit.
In short, while I understand the theory, the real-world results might fall short without proper calibration and testing. It's better to rely on verified data and proven setups rather than guesswork.
Seriously, I understand it’s tough. Have you gone back to the BIOS you used before and tried to recreate your old overclock? It might be worth a few hours just to keep things going.
Both outcomes came from BIOS MJ and they’re reliable. If I go back to MH, my PBO could fail again.