As for temps, thermal limits & such - when my 1700X was sitting @ 3.9GHz/1.4v+ (under yet to be replaced Supremacy EVO waterblock if i'm correct) i got to 70+ degrees C at times when testing with Prime95 Blend test & it crash not even half an hour when i ran Small FFTs test on same frequency, chips are binned (if that's the correct word) differently, it's a lottery that's what i'm saying.
Few YT big guns/tech gurus if you will (JayzTwoCents, etc...), ran into same issues, albeit in case of JayzTwoCents he tried 4.4GHz on 2700X, but i digress.
Ahh, see I was thinking maybe there was something to the fact that most people tend to hit a temperature ceiling at the same time as a frequency ceiling. I kind of wonder if core clock limitations have to do with power... ...how rather drastic increases in current draw may affect temperature coefficient, among other things, which in turn affects stability. Most people overclocking these on air find that there is a specific temperature range that will reliably make any Ryzen wobble. It seems to be around 75-80C. Whatever your chip does south of that temperature range is all it will do. And it seems that increasing the cooling to take that same unstable overclock below that temperature range can sometimes make it stable. I'm not talking huge gains, here... ...maybe 50-100mhz. Still, it makes me think that Ryzen is more temperature-sensitive than we may realize, and that what we think is just silicon lottery and hard architectural walls may actually be a little more than that.
I'll put it this way. If I were to induce a modest overclock to run in that temperature window, it would still crash, even though the chip can technically do it. This is different from how overclocking as I've known it usually works. Temperature ceilings are a thing, but typically people will hit a clock speed wall first and temperature ceilings only come into play with inferior cooling. Not saying it can't happen. But I agree, it's only half of the problem. I can kinda see how having a really good cooling system might not be enough to prevent a crash, even at lower temps.
It's kind of a double whammy, you know? As it gets hotter, it becomes less efficient. Same effect as going further under load in itself. IF on its own soaks up current like crazy at higher clocks. What I'm suggesting is that if we had a similar core architecture with an interconnect architecture that doesn't comprise up to half of the whole current load at peak operation, that chip might actually be able clock higher without major changes to the rest of the architecture, simply because the whole package is easier to power (meaning better stability) and thus generates less heat (also better stability.)
But that is to say... clearly it's not just about heat. Perhaps it's equally the current generating that heat. Perhaps this split current load between cores and IF asks too much of conventional VRM's. It's no secret that Zen and Zen+ have significant droop. As you go further and further up, current demand for IF increases exponentially, eventually surpassing the rate at which the cores themselves ask for more current, in theory. When you get to full utilization of heavily overclocked cores linked via IF, it may simply not be possible to deliver power to everything properly. So even if you can keep the temperature under control, you still have to contend with that.
Makes me wonder what would happen if Infinity Fabric had its own dedicated power section, much like SoC does. Might help curtail problems with balancing voltage to current when the IF floodgates open and load ramps up. I sometimes wonder if that ramp up isn't a bigger part of what's throwing a wrench in the thing - both asking for a lot of power at different rates simultaneously. Don't know enough to know how possible that actually is to try and deal with, but I think it would be interesting to see a few things change regarding how power is handled later on. You could see attempts to boost individual cores intelligently as a "soft" way to get around this and push performance up without hitting the temperature or power draw walls.
Don't mind my nonsense. I have no clue what I'm talking about. Just kinda interesting to muse over from my uninformed point of view... while I wait for 7nm.
I'm more into finding out what each options in new "AMD CBS" thingy does to stabilize my OC, than touching RAM frequency, only thing i set manually was RAM voltage, was 1.2v, now 1.35v (kits standard/XMP 2.0 voltage). Works just fine. Also, Flare X 3200MHz/CL14 is the best kit for all Ryzen series, 1000s & 2000s. And yeah, i wanna test Zen 2 as well once it announced & hits PC stores shelves.
Cheers.
Overclocking these things is definitely getting interesting. Little by little they are taking systems that have always been automatic and allowing us to tweak more and more aspects of them. Lots of new ways to balance power/stability in the future. It's like a whole new world is opening up and everything that used to make sense... doesn't XD Options like that kinda tie into what I was getting at though. I think in the future we will be looking more at tweaking power delivery more in order to increase stability at higher clocks, rather than the old ways of just throwing more volts at it
It's going to take more nuance and finesse. Maybe in leiu of the gains we used to see going forward in the traditional way, we may still find a few more tricks lurking beneath the power side of things.
And yeah, generally 3200/14 b-die is among the best there is. G.Skill has the good stuff. I would've like to nab some, but I know now it wouldn't have made a difference, heh. This mobo won't even do XMP for these 3600/17 sticks at rated speed, so I don't feel so bad about having middle-road b-die.
All I really know is it is a good time to buy AMD - super glad I got back into this stuff when I did. It's like the universe knew I wanted cool new AMD stuff, so it lead me back to PC building, just in time for me to see AMD make a comeback with really compelling APU's, a whole new architecture with great value and potential, and the first 7nm consumer CPU's enroute. They've come up with some interesting stuff this time around. Even if Intel made more sense for me and my needs right now, I'd still want to have a Ryzen machine just to mess around with. Here's hoping we get more to explore soon!
jk
