Will the Ryzen 3000 chiplet layout cause heat sink headaches?

Fuzzy_3D

Limp Gawd
Joined
May 27, 2007
Messages
195
I was looking at the chiplet layout on the ryzen 3950x:

Ryzen9_3800X_Hand_575px_678x452.jpg



The off-center chiplets got me thinking about heat sink/cold plate orientation and design. Past single chip dies meant concentrated cooling effects in the middle, but with this layout, it looks like some formerly reliable coolers might not be very efficient when mounted on a ryzen 3000 CPU. :depressed:
 

Fuzzy_3D

Limp Gawd
Joined
May 27, 2007
Messages
195
I imagine it won't be any worse than threadripper with its offset from the middle dies.

That's kinda the worst case scenario I imagined, since few standard heat sinks worked properly on the threadripper.

Looks like neither my NH-D15 or my H115i will be much good here. Not the end of the world, just one more item on the shopping list. (y)
 

Randall Stephens

[H]ard|Gawd
Joined
Mar 3, 2017
Messages
1,591
That's kinda the worst case scenario I imagined, since few standard heat sinks worked properly on the threadripper.

Looks like neither my NH-D15 or my H115i will be much good here. Not the end of the world, just one more item on the shopping list. (y)

Everything I remember seeing about TR was that if a die was covered it got sufficient cooling. Did you see something else?
 

dragonstongue

2[H]4U
Joined
Nov 18, 2008
Messages
3,162
you'll know more once released etc.
it could be by them doing said chiplet it might likely actually help as I can only imagine the heat from the IO die is as much or more than the main "chips" seeing as they have the majority of the high speed connections inside that new chiplet, so it might be the 14nm IO chiplet is where the "hotspot" is, vs standard designs where the heat is a more generalized "everywhere"

IMO....^.^
 

N4CR

Supreme [H]ardness
Joined
Oct 17, 2011
Messages
4,948
That's kinda the worst case scenario I imagined, since few standard heat sinks worked properly on the threadripper.

Looks like neither my NH-D15 or my H115i will be much good here. Not the end of the world, just one more item on the shopping list. (y)

Could always run a copper heatspreader but it might hurt temps a little. Just hope it's not the end of the line for my OG 939 Scythe Ninja that's been through three or four builds now, hopefully it has enough footprint.
 

N4CR

Supreme [H]ardness
Joined
Oct 17, 2011
Messages
4,948
you'll know more once released etc.
it could be by them doing said chiplet it might likely actually help as I can only imagine the heat from the IO die is as much or more than the main "chips" seeing as they have the majority of the high speed connections inside that new chiplet, so it might be the 14nm IO chiplet is where the "hotspot" is, vs standard designs where the heat is a more generalized "everywhere"

IMO....^.^
The same IO chip is what is used in the x570 and they are only 15W (all the RGB POS boards using active cooling for 15W... ffs) so it will be less than the chiplets.

They should do boards like pic related again, I'd pay extra for it to save hassle of modding it myself..

p35p_circu2.jpg
msi8.jpg
msi6.jpg
 

mnewxcv

[H]F Junkie
Joined
Mar 4, 2007
Messages
8,875
the amd IHS does a great job transferring heat as it is. I'm really not worried about temps with this new CPU, though water cooling with 16 cores seems like a good idea if you want to attempt overclocking.
 

IdiotInCharge

NVIDIA SHILL
Joined
Jun 13, 2003
Messages
14,675
The only issues would be with assembly, i.e., if the IHS isn't making full contact with all three dies there could be uneven cooling performance.

And that's not very likely. AMD has been doing this for a while, as have many in the industry.

Now, if you want to delid...
 

N4CR

Supreme [H]ardness
Joined
Oct 17, 2011
Messages
4,948
^^At first glance, I thought the first pic was Roller Coaster Tycoon with RTX enabled.
Hahaha great observation it is quite a lollercoaster. Or one of those crazy water park rides :D

The only issues would be with assembly, i.e., if the IHS isn't making full contact with all three dies there could be uneven cooling performance.

And that's not very likely. AMD has been doing this for a while, as have many in the industry.

Now, if you want to delid...
Yeah no thanks on the delidding. There were already some issues with Vega not having same die heights depending on where the molding was done and I can't imagine it'll always be perfect on Zen2.
Intel and AMD using indium solder regularly these days (an increasingly rare material - not the best move IMO) should prevent any of these die height issues for most normal use. Other part is I get a feeling bit like the 9900 that we will begin to increasingly run into the thermal dissipation limits of Silicon. So the heatspreader won't be the weak link.. getting the heat out of the silicon will be especially with such high densities as 10-5 marketing nm and beyond.
I know one solution that would help but it costs more than the CPU itself in many cases and I have not even seen LN2 guys using it yet ( really want to try it in future).. it's all about thermal gradients at this point.
 
Last edited:

TheRookie

Limp Gawd
Joined
May 21, 2019
Messages
213
Since the heatsink should cover the entire heat-spreader on the processor, there shouldn't be an issue
 
  • Like
Reactions: N4CR
like this

N4CR

Supreme [H]ardness
Joined
Oct 17, 2011
Messages
4,948
Noctuas come in 40mm...
Oh gosh I didn't know that. If anyone could get the bearings to last ( other than delta) it would be them. My main issue is the diameter and speed required to do more than make noise is the elephant in the room. 40mm and quiet is basically useless. Even 60mm is leagues ahead. Just like subwoofers and engines, there is no replacement for displacement.
 

Randall Stephens

[H]ard|Gawd
Joined
Mar 3, 2017
Messages
1,591
Oh gosh I didn't know that. If anyone could get the bearings to last ( other than delta) it would be them. My main issue is the diameter and speed required to do more than make noise is the elephant in the room. 40mm and quiet is basically useless. Even 60mm is leagues ahead. Just like subwoofers and engines, there is no replacement for displacement.

I use a noctua 40mm on the heat break for my 3d printer. it's quiet. it cools well. If they were to slap noctuas on these boards, I'd have zero hesitation. I don't think they'll be noctuas though.

Not my printer, but here's one.
54578611656798245210940077c32ced_preview_featured.jpg
 

Nathan_P

[H]ard DCOTM x3
Joined
Mar 2, 2010
Messages
3,479
Well, its roughly 65w for the 1 chiplet cpu's and 105w for the 2 chiplet cpu's. In simple maths that 40w for each chiplet and 25w for the I/O die.

But its a valid question, lets see what happens when the reviewers get their chips.
 

N4CR

Supreme [H]ardness
Joined
Oct 17, 2011
Messages
4,948
I use a noctua 40mm on the heat break for my 3d printer. it's quiet. it cools well. If they were to slap noctuas on these boards, I'd have zero hesitation. I don't think they'll be noctuas though.

Not my printer, but here's one.
View attachment 167805
Wow ok that's significantly better fin to motor ratio than most 40mm I see... Might actually have a chance to move air. Looks great!! Thanks for posting that.
 

NKD

[H]F Junkie
Joined
Aug 26, 2007
Messages
9,393
The only issues would be with assembly, i.e., if the IHS isn't making full contact with all three dies there could be uneven cooling performance.

And that's not very likely. AMD has been doing this for a while, as have many in the industry.

Now, if you want to delid...


I think AMD probably thought of this ahead of time before going more chiplets and io with zen 2. THere is a thread on reddit about them using material to gain common height.

 

Nobu

[H]F Junkie
Joined
Jun 7, 2007
Messages
8,846
Wow ok that's significantly better fin to motor ratio than most 40mm I see... Might actually have a chance to move air. Looks great!! Thanks for posting that.
Yeah, they have 10 and 20mm versions, the 20mm ones have a much lower noise spec, probably don't have to spin as fast.
 
  • Like
Reactions: N4CR
like this
Top