How is the GPU able to dissipate so many W at the same temp vs the cpu ?

Associate
Joined
3 May 2006
Posts
1,448
So I have a Ryzen 3950x on a dark rock pro 4 cooler & an MSI rx 6800 gaming X on the cooling that comes with.

So i was playing Shadow of the Tomb Raider at stupid 4k settings & very nice it is indeed. Meanwhile, I had AMD adrenaline or whatever it's called open out of curiosity. The rx 6800 would generally be at 65-100% usage, depending on the scene, often 90% + - this gave 70 deg GPU temp, 80 deg hotspot when under 95+% load. Power consumption was in the 150w - 200w range.

Meanwhile, the CPU was always around 6% (16 core CPU, Tomb Raider just doesn't require heavy threading). Adrenaline said it was consuming 50-60 watt but the temp was 70 deg ? It wasn't even hitting top single-core clocks, it was around 3.2 ghz.
OK, so I run a CPU benchmark, Cinebench R23 & the all core temp is 75 deg, consuming 103 watt, so it basically has almost the same temp regardless of the utilisation & watts being consumed.

So I have a question:

- Given that the CPU & GPU are both TMSC 7nm (I think ? I could be wrong) - how come the GPU using 4* as much power in W is able to maintain similar temps? Especially given that I have one of the best mainstream air coolers on the cpu ?

- Or to put it another way : Why is the CPU at 70 deg when lightly threaded & not even hitting max clocks ?

& then more of a statement:

It's crazy how we've got to the point where how many cores are being loaded makes basically no difference to CPU temps, where heavy single-core loads at much lower wattages result in basically the same temps, give or take. This is due to how modern CPUs work, but it takes a lot of mental adjustment to get used to.

I suppose what's happening is that the CPU is load balancing the fan to maintain lowest possible noise at steady temps. I'm just amazed how much more Watts the GPU can deal with, without becoming noisy as hell.
 
Last edited:
Soldato
Joined
28 Jun 2013
Posts
3,657
die area is smaller on cpu i think, heat is harder to get from its die into the heatsink to get rid of it. im probably wrong but that is why i thought it was
 
Soldato
Joined
15 Oct 2019
Posts
11,694
Location
Uk
Yeah the GPU die is directly contacting the heatsync while with the CPU it has to go through the IHS.

The GPU die itself won't be using all the power as you have other things like the GDDR6 drawing power also.
 
Associate
Joined
24 Mar 2021
Posts
48
Yeah as others have mentioned the CPU die has to go through the IHS, which also means going through two layers of thermal paste (CPU die-TP-IHS-TP-heatsink .V.S. GPU die-TP-heatsink). That's also why you get people delidding their CPU's and replacing the stock paste with e.g. liquid metal to get better cooling (for e.g. overclocking).
 
Associate
Joined
12 Jun 2021
Posts
1,663
Location
Leeds
The heatsink on that RX6800 is massive, that is how it manages to dissipate so much energy and remain quiet. I was shocked how much bigger my Asus Tuf 6800 was than my GTX 1080, would say over 2x the fin surface area. Also , for this generation of graphics cards the 6800 is very frugal in its power requirements. 6800 uses less Watts than 3070ti that has similar performance

You answer your own question about the cpu temps. Fan curves, fans run low rpm when cpu is cool and as the cpu heats up they speed up till equilibriam point is reached and temps remain almost steady. Don't you have fan expert 4 installed to control all your fans via software ? can see the curves very easily that way.
 
Soldato
Joined
18 Oct 2002
Posts
3,256
Location
Andromeda
As said die size and what's on the die, the components that make up a ryzen chiplet is high power CPU cores and cache, the IO die has low power components like memory controller, USB, PCI-e, various hubs, etc. A GPU is a single die with all the high and low power components.
 
Soldato
Joined
6 Jun 2008
Posts
11,618
Location
Finland
3950X Die Size 2 x 72 mm + 125 mm (IO-Die) mm2

6800 dies size 520 mm² also in direct contact with a massive, dense cooler
And not only that, but GPU produces heat evenly along most of its area.
Again in CPU only few cores are under full load in CPU and responsible for most of the power draw.
Meaning heat production is focused into very small spots.
 
Associate
Joined
22 Jun 2018
Posts
1,582
Location
Doon the watah ... Scotland
Again in CPU only few cores are under full load in CPU and responsible for most of the power draw.
Meaning heat production is focused into very small spots.

this was discussed somewhere in relation to heat pipe positions on cpu coolers as well iirc.

because of the spot locality of Ryzen heat generation, it’s possible that heat pipes don’t pass over the top of those heat spots in the most efficient layout.

Ie, if the pipes are orientated one way, a single pipe may pass over multiple areas of cores so have to deal with dissipating a lot of heat (leaving other pipes to pass over little else) where as if the pipes were rotated 90deg, each pipe may only pass of a single area of cores and the heat is dissipated across a wider amount of heat pipes.

edit:
Having gone and read up articles again about it, this notion appears to have been debunked and that orientation doesn’t really matter.
 
Soldato
Joined
6 Jun 2008
Posts
11,618
Location
Finland
Ie, if the pipes are orientated one way, a single pipe may pass over multiple areas of cores so have to deal with dissipating a lot of heat (leaving other pipes to pass over little else) where as if the pipes were rotated 90deg, each pipe may only pass of a single area of cores and the heat is dissipated across a wider amount of heat pipes.
In usual front-rear airflow positioning single towers have different heatpipes above different chiplets.
But for direct touch heatpipes you want one with heatpipes tightly together for two heatpipes per chiplet.

Again in case of solid base heatsinks that bottom spreads heat more evenly to all heatpipes.
Though instead of Intel optimized convex shape, base should be flat to get best contact to heatspreader at wide area.
 
Associate
OP
Joined
3 May 2006
Posts
1,448
Wow, thanks everyone, some very informative answers. I had an inkling as to what it might be about but these replies really clarified things. I should mention, I was playing at a v-synced 60fps which is why the GPu wasn't always at 100% load - it just depended on how demanding the scen was whether it needed to push that hard.
 
Associate
OP
Joined
3 May 2006
Posts
1,448
The heatsink on that RX6800 is massive, that is how it manages to dissipate so much energy and remain quiet. I was shocked how much bigger my Asus Tuf 6800 was than my GTX 1080, would say over 2x the fin surface area. Also , for this generation of graphics cards the 6800 is very frugal in its power requirements. 6800 uses less Watts than 3070ti that has similar performance

Yeah, I came from a 1080ti FE & the new card is absolutely massively bigger, even though on paper it consumes less power than the 1080ti (people say the new cards are more 'spiky' than older gens). It's also noticably a lot quieter than the 1080ti, although that was never a problem, it wasn't too loud & it was a nice, whooshy sort of noise that didn't really bother me. I think it's partly because most companies are using the same cooler for their 6800 as their 6800xt, so they are kind of over-engineered for the 6800.
 
Associate
OP
Joined
3 May 2006
Posts
1,448
The GPU die itself won't be using all the power as you have other things like the GDDR6 drawing power also.

Good point - I think that's especially true of the NVidia cards with their GDDR6x - although I guess they have less of it, so it might balance out.
 
Soldato
Joined
15 Oct 2019
Posts
11,694
Location
Uk
Good point - I think that's especially true of the NVidia cards with their GDDR6x - although I guess they have less of it, so it might balance out.
A good example of this is the 3070 vs 3070ti with the later having a power draw of around 70w more despite only having a few extra cuda cores.
 
Back
Top Bottom