x570 Aorus Master with 3900x boosting behavior
Anyone with master f5 bios have terrible time recovering from dram overclock failure?
Took me 15 mins of power cycling to get back to the bios reset screen.
In that time, it's either black screen of nothing and then stuck aoruos splash screen to windows recovery screen and then finally bios reset.
15 mins of frickin powering on off and mashing reset in between, just to get into bios.:specool:
I've found that the easiest and quickest way to escape that situation is to always save your settings to a profile, and also save a copy of your profile to a file on a flash drive.
When faced with a boot failure situation like that, just immediately go for the Clear CMOS button so you can quickly get back to default settings. Then you can load your profile back up either from the BIOS, or from a file, and edit them as needed.
I am using a 3900x with an x570 Aorus Master, currently on version F5. I've found that the factory F3 version works the best as far as single-core boost, that is to say, only on version F3 (AGESA 1.0.0.3) do I see 4.6-4.625ghz (momentary) boost, and ~4.5ghz sustained under CineBench single-core with fans at max. I've tried the various F5 versions, and version N11 (AGESA 1.0.0.2), and every version other than F3 is restricting my fast CCD to 4.3ghz max under single-core load. I'm certain that my single-core boost testing is performed correctly, as I'm carefully setting thread affinity on CineBench after the bench run has started (as CineBench resets the affinity mask to default at the start of every bench run, thanks devs..), and verifying which core is at 100% load, and that the other cores are idling.
I've noticed that on version F3, during CB R15 single-core, the idle cores sit at about 2200mhz on average, while the single-core bench thread rides at about 4.5ghz sustained (with fans at max, sitting next to an A/C vent, with case side-panels removed). But on any other BIOS version, I just cannot get anything better than 4.325ghz single-core boost, if I'm lucky, and it generally just will not go past 4.3ghz at all. But, at least on BIOS F5 I'm seeing 4.3ghz max on 2 out of 6 cores on my slower CCD, which is slightly higher than I get on version F3.
On version F5 and N11, under CineBench single-core, even though only one thread out of the 24 is fully loaded (on one of my 4.6ghz capable cores), all of the other cores boost up with it to nearly 4ghz. So it would seem that since the idle cores aren't allowed to settle down to a low clock speed near to 2ghz, that the boosting algorithm thinks there isn't enough headroom to allow the 4.6ghz capable cores to boost up high as a result.
It seems clear that AMD has a long way to go in order to fully optimize the core boosting algorithms, and in my opinion, they should add options to manually control the boosting behavior. For example, there should be a setting to enable a boosting profile which guarantees that headroom is always maintained to allow the faster cores to boost to maximum clock speed when requested, at the expense of lowering the overall multi-core boost, which should be desirable for most games and light tasks to run optimally.
When it comes to the operating system integration, it also seems clear that a lot of work is left to do. Such as ensuring that low priority / background service threads *never* get scheduled onto the high speed CCD / cores, and that heavily loaded threads are always scheduled onto the high speed cores first.