5900x and 5950x clocks
5900x and 5950x clocks
Only this way can it remain stable for FAH at 4.4G—raising the voltage into the 1.375-1.4 range is necessary... that’s the SVI2 TFN core voltage. I don’t want to investigate early degradation effects, so thank you for that. It really doesn’t matter much for gaming benches and has a negligible impact on FAH TPFs, especially with thermals even under a 240mm AIO.
everything looks good. currently i'm running a 2700x at 4.225ghz across all cores, which is just 125mhz under boost—very little impact on single core performance. however, upgrading to the R9 5000 series might increase single-core losses. i'm trying to decide exactly which CPU to purchase and what steps to take once I buy it. as you mentioned, we'll learn more in the coming weeks.
thermals don't bother me much, and as someone who's been using them for 20 years, I understand they can be a bit too cautious about temperature and voltage settings. I have a dual chamber duel open loop system with 1 240rad and 1 120rad just for the CPU, so I'm comfortable pushing it to its limits, but I won't accept dropping 300 to 400 MHz on single-core performance.
You don't need to evaluate every core for stability simultaneously. It's possible to temporarily turn off the cores you're not testing.
I attempted that but didn't notice any benefit. Not turning on SMT really matters a lot. Ultimately, I prefer overall efficiency over just fast clocks. Considering the issues and risks of early performance drops from overclocking, the results fell short compared to a more aggressive PBO approach. The top performers still suffered significant loss in single-core speed while only gaining slightly in multithreaded tasks, especially under heavy loads such as video encoding with Handbrake.
Here is a helpful video about AIO coolers. Watch it: https://www.youtube.com/watch?v=23vjWtUpItk&t=7s
I understand that three of my cores are exceeding 25MHz above the maximum, while three more are performing very close to the boost frequency and the other two rarely reach 25 under load. This matches the core ranking shown in HWInfo. HWInfo not only displays the silicon quality ranking AMD assigns to cores during binning but also the order in which the scheduler prioritizes threads. That priority is based not just on ranking, but also on shared resources like caches.
I can't anticipate much improvement...my processor is generally rated as 'bronze' by CTR. I received it last August, so it's definitely not one of the high-grade samples that are now common from a mature manufacturing process.
The issue isn't with the reservoir or radiator; the liquid doesn’t reach thermal saturation. I can run FAH for extended periods without the temperature ever exceeding 75°C. Prime95 and small FFTs stay within the 82°C range for over four hours before I shut it down. The main constraint is the water block, lacking sufficient fin surface area to effectively transfer heat to the liquid.
I’m well aware of the limits of liquid cooling—both AIO and CCL options. AIO units can cause the liquid to become saturated (as noted by Linus on 120mm rad), and performance may degrade as microfins in the water block become blocked (which I expect will happen, though it’s only been three and a half years). CCL is costly and requires maintenance (similar to the fin issues in WB units).
Given the effort and cost of installing a CCL, I might gain a few dozen more points in multi-thread performance. The trade-off would be watching ST performance drop and experiencing micro-stuttering during gaming—something I can’t justify.
I think I got lucky with a silver for my Ryzen 2. I managed to get one of the 1600 AFs early this year. Most were solid overclockers, but one could run at 4350MHz with decent voltage and had a good IMC as well.
I also think I'll need a fresh motherboard. I'm using a B450 Mortar with a four-phase VRM. It handles a lot of phases, which keeps things cool even when running Prime95 with an 8-core processor. Four isn't ideal for 10 or 12 cores, but it prevents voltage drops and keeps power steady without noticeable undershoot that could cause the CPU to freeze. I suspect others are experiencing the same thing.
If I had to spend on a new 12-phase board—like the B550 ROG Strix-A—and a compatible case, I’d only get a few extra points in multithread performance and it would still be similar overall. It wouldn’t really save me much money.