Hmm kinda interesting article:
The new Ryzen 5800X3D is not going to be manually overclockable.
The reason seems to be because of the used 3D V-cache not being,
able to run with higher voltages then 1.35V
The IF and MEM bus can be overclocked but that’s about it.
AMD asked motherboard manufacturers to disable the functionality.
So what this actually going to mean raises some question marks.
And is food for discussion.
AMD ramps up voltage past that by default on idle, guess they’re locking that down
Yeah 4.5Ghz boost might be where it´s at due to the lower voltage.
But AMD still claims that the cpu is going 15% ish faster on average in games,
then a 5900X due to the 3D stacked cache.
Of course i personally take that claim with a big shovel of salt,
untill actually benchmarks are out in the wild but still.
What this is going to mean for the future of 3D V-cache,
on zen4 and overclockability is of course yet to be seen.
It’s interesting having that much cache, it evokes memories of swapping onto HDD I have from early windows 95 days.
Basically, as a programmer, how and when you prefetch and various performance counters just got a whole lot more important - if you prefetch too little you’re causing future cache misses and hurting IPC or CPI. If you prefetch too much, you’re causing evictions and hurting IPC or CPI again. It’s as if you need to have a separate version of your code for different cache amounts or need tons of inline-ing to happen in order to let the CPU prefetch things for you efficiently (security implications)… or you need to beg amd for more cache control instructions and/or tooling to either simulate chip behavior or trace which IP values cause cache misses.
Also, is it easier to deal with this, or just throw more cache onto the chip as next generation comes around and yell bigger numbers on stage.
Sounds rad, wonder how much ram will matter.
Yeah i´m really curious how much of an impact on performance,
all the additional cache is going to have on all sorts of workloads.
This is going to be an interesting situation.
Well since Milan-X is going to have a similar setup for the server market, AMD must have been testing it on a whole heap of workloads to find the ideal amount of cache to be used.
I’m sure there are going to be quite a few workloads that do see a very nice bump in performance and others not so much and probably some that may regress.
I’m assuming that means higher minimum frame rates, and you could be a politician lol.
Nah, I value some nice discourse, that being said, I’m not rich enough to play in that arena nor would I want to 
Now that I read about milan I can only assume the voltage lock is to keep temps sane, and latency low. Am I right in thinking of whatever a Vertical L3 cache is as a turbocharger?
I’m going to have to do some more reading before crawling back under my rock.
Same . Id love a deep dive into how cache effects IPC.
Threadripper and these new chips all cap out at 4.5GHz with one exception. Power / thermals allowing. Will be some interesting reviews.
Intels performance is coming from raws watts a little still at the moment but they are back in the game warming us in winter when WW III ends 
More like better air intake and fuel mixing for bigger bang. Not really forced induction where the performance can double, to beat the metaphor to death.
Depending on the serverity of the issues if cache runs at higher voltages than 1.35V I can see a motherboard manufacturer keeping OC enabled for those chips and calling it an oopsie.
That’s where true performance lies mostly. So I don’t see core overclocking missing as that big of a deal. But this might also be the CPU that’s gonna benefit less from memory overclocking due to the gigantic cache at it’s disposal.
It must’ve been a microcode nightmare to make it work properly!
Yeah I must say with things like the onboard boost management and in applicable cases (not sure where 5800X3D lies) PBO have made manual OC pretty much not a thing.
I had bulldozer and loved manual OC, nothing extreme but 900mhz was decent. But since I got my 3700X I have just let it do its thing and its been good. In would have to spend huge amounts of money and time on “the right” motherboard, CPU and RAM to get an actually worth it manual OC on current Ryzen.
I don’t really see the problem.
As a quick example:
Mine’s on stock. It doesnt really have much overhead in the first place…
I wonder if they’ve decided locking this chips ceiling , to maintain internal thermals
That additional sub-die component, overlapping the CCX, is an added layer to clear
It would be a major test, on L3s engagement with various software / gaming brags
My take is eh who cares the auto oc all chips do with a toggle is plenty good. Good to know won’t auto oc super hard but eh is my take on manual ok support.
Not sure anyone should care; my 5950x still manages to hit 90c on a custom looo with default PBO cause it’s willing to pump 1.4v into the damn thing
At this point, been wondering if I need to tear down the loop for cleaning and reseating, but those voltages are also suuuuuper high
I haven’t manually OC’ed anything since the days of my Phenom II x4 965. …I miss that era.
Since then, I’ve only ever made anything I’ve messed with draw three times more power for a negligible amount of extra performance in some specific workloads. I’ve made it run worse on occasion. Especially now with how everything auto-boosts, I just set XMP, make sure the motherboard isn’t doing anything stupid with power management (looking at you, Asus), and call it a day.
I have a manually tuned each of the power states and precision boost settings to get 4.85GHz to stick on the 5600X. The generic clock offset doesn’t keep it above 4.8, so extra fine tuning was required. I think the definition of “manual OC” has shifted since the days of keying in a BCLK and balancing each voltage. Now it’s selecting an offset and tuning power/frequency curves point by point to get your desired result.