I have X570 Aorus Pro, 64GB of 3000Mhz Sniper X memory and 3950X and 360mm Thermaltake AIO.
I am seeing max 3950x boost clocks around 4,5Ghz in one of the cores during a single core Cinebench 20 test. I get nowhere near 4,7. This is with the XFR and PBO enabled.
Is it simply that my 3000Mhz memory will hold me back no matter what settings I choose?
Now realistically speaking I don’t know if none of this make any practical difference for me as I work mostly in 3D modeling and rendering and certainly upgrading 64GB of memory to higher speed (even if it is possible?) is not going to be worthwhile investment for me.
I just wonder if I am leaving performance on table here.
Nope. PBO max frequency is not related to memory. Zen 2 systems have a separated memory, CPU and Infinity fabric clock.
Max PBO is related to power, temperature and voltage stability. It is also influenced by firmware (BIOS) configuration and by the limits imposed on Windows by the chipset/CPU drivers.
There is a know issue with max PBO related to chipset driver / AGESA version on the Motherboard.
Try to update to the latest available BIOS and chipset drivers.
Also take into account that you might not be watching the maxed out core on Windows task manager. Try to run your test and monitor the system with the HWinfo program. On that software you will be able to watch each core individual speed on real time.
I have also noted that maximum PBO freq is conditioned by the Load Line calibration policy that you set on the BIOS and maximum allowed current, also set in BIOS.
Not sure on zen2, but I did notice better boost behaviour on my 2700x by DROPPING core voltage.
So far I’ve dropped ~0.05v on vCore and have observed 100mhz or so better boost and cooler temps. Purely through 100% stock settings other than voltage. It seems Ryzen, like most AMD products is over-bolted a bit out of the box to ensure that even potato spec cores perform stable at spec.
I’m currently running with that for a while and will try dropping another 0.05v. Plenty of people are running -0.1V or so stable on the 2700x.
It may be worth experimenting with reduced core voltage to see if you get better boost via lower temps.
That said, if you’re mostly modelling and 3d rendering I would wager you’d get more benefit from cpu core clock speed than memory speeds (as AFAIK, the workload is dominated by the complex math, not so much bound by lots of high speed IO - the io that does happen is pretty cache friendly as well I suspect due to repeated accesses as it performs calculations to render each area of the scene).
Running the memory controller less hard/less hot may also help get better core clock due to lower temps. I.e., trade memory controller speed/heat for cpu core speed/thermal headroom.
But this is one of those workload specific cases where you’ll probably want to benchmark performance for yourself.
Running 3200 mhz ram overclocked to 3296, using an NHD-15. Highest single core speed I could muster was using ryzen master PBO on the gaming preset with the +200 mhz turbo offset in the bios. It is important to note gaming mode disables half the cores so I wouldn’t use this daily. I got it to make the max boost frequency 4.925 GHz, however the highest single core speed I could measure in ryzen master or cpu-z when prime95ing or cinebenching on a single core was 4.75. This clock speed was not stable at all, and would crash after one or two cinebench tests. Since then I wound up dialing it back to the auto overclock w/ pbo in ryzen master, with a second manual all core overclock setting for sustained compile/render workloads which I have set to 4.25 GHz all core. I think the chip will do 4.35 with some tweaking but I couldn’t make it stable under 90 degrees for over an hour.
Also, I should mention this was done on a pro ws X570-ACE board with a slight vcore underclock and a 1.375 max core voltage.