I'm glad to find someone already started a thread for this, as I was about to
.
Here is a more detailed description of the test.
Also I'd like to point this out:
Across six motherboards, countless bios'/agesa, my 3800x has never hit its rated boost clocks.
That sucks EarthDog. I assume this is on liquid as well? My feeling on all of this is two fold. On one hand with Zen+ people complained that their CPUs had a maximum boost clock, wishing that XFR2/Precision Boost/PBO would take the CPU higher if able. On the other hand, I think AMD is trying to meet up or at least get closer to rumors of higher clocks, and to some extent false advertising. They've been bashed hard for a presentation that mentioned a clock of 4.8GHz, something no stock CPU can do, and even with "AutoOC" enabled (this increases the max boost by 200MHz) while theoretically possible it hasn't been seen real world. I feel like they could have gotten away with missing the mark by 50-75MHz with some samples, but that guy writing 4.8GHz on the chalk board really got under people's skin.
The sad part is they don't need to. They're already matching or beating Intel at most price points regardless of if they're hitting a specific clock. However we all know how many consumers still at the end of the day look at clock and not benchmarks. I do think this is their true motivation, and it is disingenuous at best. I find this especially true for a SKU like the 3800X that could have just been sold as a 3700X and met it's advertised boost clocks. I think having a more powerful "AutoOC" or feature to extend the boost clock limitation to satisfy the enthusiasts like myself that always wonder, "how high could it really boost?" meanwhile having a more attainable stock setting would have been a more honest approach.
I've also posted before about the thermal scaling of the CPU, but I have a hard time imagining this having a substantial impact on
single core workloads. Maybe EarthDog could comment on that, what kind of temps he is getting running the R20 single core test.
For my part I just wanted to see how significant of a departure this is from Zen+ behavior, since I am honestly pretty guilty of just plugging it in and going, and there wasn't a lot of attention paid to it. I did the single core CinebenchR20 and every single core reached my 2600X's advertised max boost clock of 4250MHz (well 4249.2MHz to be precise, but that just comes down to clock gen, the multiplier was x42.5). Never did I see a thread fully loaded, and it seems that the CPU was sharing the tiles between one or two cores at a time, and then alternating the core doing the work, and at least one core was maintained at 4250MHz for nearly the entire test (some of the runs of easy wall only tiles on the right side saw clocks decrease, as well as some of the difficult couch tiles seemed to cause cores to throttle and switch unexpectedly a couple times). Of course my speculation on what the CPU is actually doing is just that, since I can't see specifically which core(s) is running the render, just thread load percentages. Also this was not with the CPU running stock, memory was at 3200Mhz, PBO enabled, and a negative vcore offset of -0.0635v. I could go back and check stock but I don't think it will change anything.
