Kids 
The Gigabyte SP3 board (IIRC MZ31-AR0) does not have that impressive of an VRM for 180W TDP CPUs.
I think Buildzoid overestimated everyday powerdraw by a bit (it is only 70W TDP more).
Buildzoid doesnât care about everyday powerdraw. For him to make sense you have to add âwhen you overclock the crap out of itâ to every sentence he says. 
2 Likes
His channel is called âActually Hardcore Overclockingâ not âactually running at stockâ 
3 Likes
they should have called it core ripper in reference to intel. 
I think it should have been obvious that any MB that claims to support 32core OC would need roughly 2x the VRM as the current gen.
This is why I take âcompatibility through 2020â with a grain of salt. Yes, it would boot, but things like this make MBs obsolete.
Now, did MB manufacturers have enough lead time and are they willing to build out 16-24phase VRMs? Theyâve done it in the past, but its expensive. They may cut corners in all but the marquee models designed for LN2.
I donât think thereâs any debating what heâs said here - a TR1 MB is good for 16 cores⌠MAYBE 24 without OC, but you should not expect to drop a 32 core model in there and crank it.
Does that mean that there will be no OC capable MBs? I find the odds of that pretty low given the LN2 world. There clearly is âenoughâ of a market and/or its worth it to drive sales for 1 or 2 âoverkillâ models. The only question, I think, is whether they knew far enough in advance to build them. Intel did not give them (or frankly even know themselves) enough warning that an 18 core part was coming. So, it took rev2 of various models and re-work of VRM heat-sinks.
They did eventually arrive and we are routinely moving 500W+ peak through certain MBs on intel. There are also configs capable of moving 1000W, but clearly these are even more exotic in their cooling configs.
Too many tranches of product and usage in one conversation. You have to break out what you are targeting to have 2 people actually communicating about this stuff. The range of product and use is HUGE now⌠50W to 500W.
Well this primal need to OC out of spec so my hardware breaks is cute. I imagine TR2 will have AMD zen+ per core overclocking so it can hit 4.0GHz but will never all core @ 4GHz because laws of physics.
Now the chips will have power and thermal limits. AMD new ryzen chips actively monitor it. I suspect TR4 motherboards are over delivering for 1950X chips. There is head room powerwise for a new chip and if cooling is good then performance.
Still I am navel gazing and need a month or two to actually see. I dont think we will need a 33 phase power delivery board like intel. EPYC allready does 32 cores with max 180W TDP that 3GHz boost.
So letâs see if we can keep the 500W and 1KW dreams for intel.
1 Like
Why?
Why not let the user and application decide whether 500W makes sense?
This is what p-states/c-states were designed to do. Theyâve already said they are targeting a 250W TDP not 180. They have to draw a line for warranty support and yield well inside what any given chip can likely do. I get that.
Let MB maker and user decide whether efficiency, cooling and warranty are primary or whether performance is. Better yet, let me decide that day to day minute to minute with the same hardware as I do now with intel.
Itâs not a primal need to overclock things. It is a recognition that cooling assumptions, yield and warranty are drastically impeding the potential of any given chip. I get 30-50% more out of a given chip than offered by factory settings and that means real improvement in my day to day productivity.
If XFR2 works on TR2 even close to how it performs on Ryzen 2000, overclocking is gonna be pretty pointless anyway.
Thatâs what worries me.
It all down to cost. A 500W MB is expensive and cooling already needs build in heat sinks and fans.
Its at the point we need smarter clocks not hard clock all numbers. You CB 7337 cause power and size. But at 7nm or smaller perhaps. Still a PC is not meant to be a space heater so < 500W.
⌠what?
How is it a bad thing to get pretty much full performance out of the box?
It not a consumer part if it is in a 1U rack on your desk whining at the top of its lungs
I would consume it⌠in my basement rack where it can whine all it likes.
- It would mean the cores were being limited from their potential by over all package power provision. Not âfull performanceâ.
- It would mean intel would still have a lead in many common use cases.
1 Like
The high core count Threadripper 2 CPUs could have been a crippled EPYCs except for one crucial detail: The Zen+ dies in Threadripper 2 are on GlobalFoudriesâ 12 nm process, but there is no existing EPYC on that process (and probably never be). 1st gen. EPYC is 14 nm, 2nd gen. EPYC will be 7 nm.
It is Buildzoid, everything he says is concerning hardcore overclocking.
The Epyc 7601 is a 180W TDP part (32c/64T, 8mem channels). Now you are telling me, a 250W TDP part with less memory channels and smaller process (12nm FinFET) is going to use 500W?
Something does not make sense.
Also: RTFM before stuffing some CPU into some socket!
The 500W number is a ballpark figure taken from 4 binned zen dies hitting 4Ghz (aka all core 4Ghz power consumption). I too am talking about OC because the market we find ourselves is one where top end performance is significantly limited by the lowest common denominator of cooling and VRM provision assumptions that TDP values set. The chips are generally capable of much more if more cooling and VRM are provided.
I think you are missing the crux of my question regarding socket/wiring. We have a socket used for 8 channel and a package outline used for 8 channel (aka epyc) being repurposed for 4 channel (aka TR1). In the first gen TR1âs interposer/package was wired differently than epyc (see der8aur).
Now that all 4 dies are hot in TR2 did they still use a distinct interposer from epyc2 or are they the same now and only fuses or MB wiring differ?
To put it another way - are the dark memory channels functional? Could Asus and the like offer a MB wired for 8 channel instead of dual quad for TR2 or are the pins permanently dark? There is no manual yet available to us. So I am wondering aloud.
I know I am a few day behind, I have not caught IP.with the thread so it may have been brought up, I will edit if it has.
The bigger problem I think is not whether the CPU can do it at various clocks and cooling solutions but that mother board. Four 8pin CPU power connectors and all the tracesand rounting that needs alone is a problem. Then there is the size of the thing, not any standard that will fit a ânormalâ desktop case. And finally the still rather dubious PSU situation, with only Super flower making a 2000w PSU and at that only for Europe because we jam at 240v. So likely running two PSUs both rated above 1000w alone just for the connectors needed leaving out GPUs and what ever else you want to put in the PC if it ever becomes a thing.
Edits as I catch up:
I think that they meant it as the i7-95.90°C in reference to its idle temps.
Its.rumoured.if you.listen closely you can hear The Steve Miller Band echoing from the Intel headquarters.
Time keeps on slippinââŚ