Playstation 5 [PS5] [Release November 12 2020]

[iroboto]

Austin Evans and co. took apart the newer, lighter PS5 and found a much smaller heatsink assembly. Unsurprisingly the new unit runs hotter.

The look of it gives me PS4 Pro PTSD...

tldr; without measuring the chip itself it's impossible to know if the increase in exhaust temperature is an efficiency improvement or not. Everyone saying it's hotter means the cooling is worse is no more wrong than people saying that more exhaust means cooling is more efficient. Both are wrong because to measure efficiency you have to measure the chip temperature vs the air temperature, or in this case, the exhaust. That being said, reducing the conductivity of materials for cooling and reducing surface area will likely result in a loss of cooling efficiency. I don't see a design change here with the fins/setup that will warrant a massive improvement over the original with such reductions in place. The chip may not have needed it to begin with so cutting back materials now that they see the maturity of their system may just be what this is. Ultimately this is a cost cutting setup. it's very likely running hotter at the chip by some amount, but it probably doesn't matter.


**
It is difficult to know whether or not the cooling is improved but we can make some very conservative assumptions about its cooling with respect to what we can see.

Using the equation Q = (Tchip - Tair) R you can determine the heat resistance level of a heat sink configuration. By in large a lower Q means a better transmission of heat from the chip to the air.

Ie: when the claim is made exhaust is hotter means it’s cooling better: by keeping all variables the same except adding say 5 Kelvin to Tair your new Q is lower than the previous one.

however we know they aren’t the same and the components that make up R in this case we know are worse. What makes up R is the combination of resistance from TIM, it’s Spreading capability, and heat sink capacity which the latter 2 are changed. In particular the sp value is reduced in area and the conductivity has moved away from copper.

In the best case scenario here, Sony would have aimed to have a heat sink configuration to be the same as the original PS5 but doing so by cutting costs. If that is the same and we assume that Q1 = Q2; then a 5 degree Kelvin exhaust increase also means a 5 degree Kelvin increase at the chip.

If they managed a better Q (lower) then likely the chip is the same temperature but the exhaust is more. Frankly, it’s not determinable without measuring the heat sink/base contact point.

but reasonably one has to assume that the reduction in surface area combined with a reduction in conductivity is leading to a higher temperature at chip. Unfortunately there is no real way to know without measuring the heat sink bases directly to find out. We jus have to make some assumptions around what Sony feels is tolerable heat for their chip.

 

[Globalisateur]

From the PS5 new 1100 model hands-on from Digital Foundry we learn interesting power consumption measures.

- Control regular gameplay (30fps RT mode): 170W
- Most demanding scene (corridor of death, 30fps RT mode), but still gameplay: 200W
- Photo mode unlocked (non gameplay, taxes mainly the GPU according to a reliable source who has actually profiled the game during photo-mode): 214W.

So what does it tell us if we trust Cerny and the dynamic clock system that is based on instructions load which are themselves based on actual power draw? It means the PS5 should run at max clocks in the vast majority of time as it seems the actual max power consumption (so logically the max number of instructions before a downclock) could be higher than 210W (and not 200W like previously thought).

During normal and demanding gameplay the system should run at max CPU / GPU clocks and it could lower the clocks only in very specific conditions: photo modes which are similar as cutscenes / when game is put in the background. All those scenes share the fact that the GPU is not limited by any CPU logic and is stupidly (and uselessly) trying to work as much as it can. It can also happens in plenty of games when you enter the dashboard while a game is running in the background. Also in some games menu and maps.

 

[iroboto]

From the PS5 new 1100 model hands-on from Digital Foundry we learn interesting power consumption measures.

- Control regular gameplay (30fps RT mode): 170W
- Most demanding scene (corridor of death, 30fps RT mode), but still gameplay: 200W
- Photo mode unlocked (non gameplay, taxes mainly the GPU according to a reliable source who has actually profiled the game during photo-mode): 214W.

So what does it tell us if we trust Cerny and the dynamic clock system that is based on instructions load which are themselves based on actual power draw? It means the PS5 should run at max clocks in the vast majority of time as it seems the actual max power consumption (so logically the max number of instructions before a downclock) could be higher than 210W (and not 200W like previously thought).

During normal and demanding gameplay the system should run at max CPU / GPU clocks and it could lower the clocks only in very specific conditions: photo modes which are similar as cutscenes / when game is put in the background. All those scenes share the fact that the GPU is not limited by any CPU logic and is stupidly (and uselessly) trying to work as much as it can. It can also happens in plenty of games when you enter the dashboard while a game is running in the background. Also in some games menu and maps.

low frame rate doesn't necessarily imply high workload on the GPU. it may be the most demanding scene workload wise, but that doesn't mean it's the most demanding on the GPU. When a GPU is stalled due to a bottleneck the rest of the GPU rests. Higher FPS will usually translate to higher GPU saturation leading to increased wattages and temperatures.

During normal and demanding gameplay the system should run at max CPU / GPU clocks and it could lower the clocks only in very specific conditions:

I would disagree with this, as I have in the past, but there's no indication here that PS5 variable clocking behaves anything like how PC does. Power is capped on PS5 and there is the reality that yields are a function of aggressively they can push a chip while still being low binned enough to be sold in a console for a fraction that we pay on PC GPUs. Wattage may be capped at a certain limit for the SoC as a whole, but that provides no insight into how variable clocking will work with respect to activity levels, the assumption is that it is trying maximize its 200W which is may or may not be true, or at least how aggressive it is in doing that.

 

[novcze]

From the PS5 new 1100 model hands-on from Digital Foundry we learn interesting power consumption measures.

- Control regular gameplay (30fps RT mode): 170W
- Most demanding scene (corridor of death, 30fps RT mode), but still gameplay: 200W
- Photo mode unlocked (non gameplay, taxes mainly the GPU according to a reliable source who has actually profiled the game during photo-mode): 214W.

So what does it tell us if we trust Cerny and the dynamic clock system that is based on instructions load which are themselves based on actual power draw? It means the PS5 should run at max clocks in the vast majority of time as it seems the actual max power consumption (so logically the max number of instructions before a downclock) could be higher than 210W (and not 200W like previously thought).

Power draw where? ...imo power limit is power draw of the APU and not power draw from the grid. So even if APU power limit is 200W, whole console can draw more because PSUs are not 100% efficient and there are other components (wifi,bt,etc.).

 

[3dilettante]

3-4 degrees centigrade = 9 degrees fahrenheit (well done mate, not)
1.4 decibels quieter though seems a larger improvement

There is some variability in noise among PS5 units, since there are multiple fan suppliers and variation in assembly. Perhaps Sony has refined this further, but at least anecdotally there is something of a lottery in terms of what fan a given unit has, and whether it's mounted in a way that worsens noise. This is not taking into account coil whine.
The new revision's fan might be found among older revision consoles as well.

It's not really about the temperature of the exhaust. Temperature is a great measurement of the heat energy in a moment of time, but when you are looking at cooling systems, the volume of air being moved through your heat exchanger is equally important. They are missing the volume component when looking at just the exhaust like that. It's supposed to be hot. That means the cooling system is doing it's job.

Somewhat higher temps might also mean a tradeoff is possible with lowered air volume due to lowered fan speed, which may give lower noise. Although if that were the case I'm curious why it wouldn't potentially carry over to the other PS5 revision, since its larger setup may tolerate lower airflow as well.
Also less clear to me, and perhaps I missed a portion of the video that showed it, was whether there was much change in the ducting/airflow of the case around the heatsink, which could influence thermals and noise levels.

I think by the way how a thermal camera works, it’s mostly picking up temperature of the body of the machine than the exhaust air? More heat is trapped in the new unit.

The camera measures the amount of infrared light emitted that reaches the sensor. There's more complicated factors that go into determining this, based on how much given materials emit and things like line of sight with the sensor. The output air itself seems to be a minor contributor to the images. If materials are consistent between the consoles, there can be variation in how much material that is in view of the sensor is at a given temperature.
The older cooler had more metal that was further out from the heatpipes or was potentially further down.
That could put more slightly cooler metal in the way of the camera on the older model, versus the newer one's exposing more metal that is next to a heat pipe to the back vent.
One of the most notable reductions are a significant cutback in the heatsink nearest the SOC, which would reduce heat transported to the more externally visible vent-side fins, and would itself be hidden by multiple layers of material. The original heatsink had a chunk of fins below the vent-side fins that was removed, with some of the length moved up to the vent-side fins. Those could be partially blocked by other material in the case as well.

I assume we all agree that the logic is fine, ie more heath is bad.
But there must be some point where the diminishing returns just kicks in and the difference between the temperatures are negligible.

Much of the metal removed tends to be metal that is further from a heat pipe or along a curve that winds up obstructed by ports at the back. So it does seem like what was removed would have been less effective. What metal that was added looks to be along extended lengths of heatpipe more in line with the main vent, which would seem to be more effective. The curve would seem to be tighter than the original, which could be more turbulent, but at the same time the outer curve leads to a less effective outlet.

I'm not sure if the dust catcher feature would be affected, or if it was particularly effective to start with.
There are subtle differences in the metal shield that may indicate some other adjustments that could affect thermal performance for specific components or other changes on the PCB.

As far as over-engineering goes, it would seem to be safer to start out with excess before profiling down to a more optimal layout. Another point I think DF brought up is that the initial heatsink was likely designed in parallel with other system decisions, which means it could have been a risk-management decision to have a layout that was sub-optimal for the eventual case if there were alternate designs in the running. Another factor is the amount of manufacturing changes needed to support a given design. Perhaps the old design was one that was easier to spin up for volume manufacturing early in the learning curve.

 

[see colon]

Somewhat higher temps might also mean a tradeoff is possible with lowered air volume due to lowered fan speed, which may give lower noise. Although if that were the case I'm curious why it wouldn't potentially carry over to the other PS5 revision, since its larger setup may tolerate lower airflow as well.
Also less clear to me, and perhaps I missed a portion of the video that showed it, was whether there was much change in the ducting/airflow of the case around the heatsink, which could influence thermals and noise levels.

To clarify what I was saying, if you look at the older design, there are 2 fin stacks that air is forced through to cool the components. The newer design appears to have only one. But if both designs are outputting the same amount of heat (and they must be, right? It's the same SOC running the same workloads), the older design is going to have the heat spread out between the two fin stacks, while the newer one is going to be using just the one. The newer design is going to have a concentration of heat coming out of that stack, so it must be a higher temperature at that point than the 2 stack design. But that doesn't matter if the same amount of heat energy is being exhausted from the system.

What we need is temperature reading from the SOC to know how efficient the new cooling is.

 

[iroboto]

But that doesn't matter if the same amount of heat energy is being exhausted from the system.

correct. Efficiency of cooling is a fixed amount of thermal transfer given an amount of air volume, materials, and spreading capability. If the heatsink is unable to keep up with the heat output of the soc, then the soc and heatsink temperature will both increase until they reach an equilibrium, this would also mean air volume temperature would have to increase since it is part of the equation for equilibrium of the system. * by equilibrium I don't mean the soc and heatsink are the same temperature. The heatsink is not 0 resistance material. It would still take time for the soc to heat up the heatsink material itself.

Definitely need that measurement at SoC to determine efficiency.

 

[3dilettante]

To clarify what I was saying, if you look at the older design, there are 2 fin stacks that air is forced through to cool the components. The newer design appears to have only one. But if both designs are outputting the same amount of heat (and they must be, right? It's the same SOC running the same workloads), the older design is going to have the heat spread out between the two fin stacks, while the newer one is going to be using just the one. The newer design is going to have a concentration of heat coming out of that stack, so it must be a higher temperature at that point than the 2 stack design. But that doesn't matter if the same amount of heat energy is being exhausted from the system.

What we need is temperature reading from the SOC to know how efficient the new cooling is.

Having fewer stacks would tend to produce to higher temperatures all else being equal. I think it's possible that some of the extra present on the older cooler was not as effective on average, and the individual stacks vary in dimension.
Deciding to lower fan speed to reduce noise would reduce air volume and increase temps further, which could be part of why things are hotter.
It's possible there are other reasons for the reduced noise besides lowered airflow.

 

[Globalisateur]

Having fewer stacks would tend to produce to higher temperatures all else being equal. I think it's possible that some of the extra present on the older cooler was not as effective on average, and the individual stacks vary in dimension.
Deciding to lower fan speed to reduce noise would reduce air volume and increase temps further, which could be part of why things are hotter.
It's possible there are other reasons for the reduced noise besides lowered airflow.

There is. This is a brand new model of fan for the PS5. The fourth.

 

[see colon]

Having fewer stacks would tend to produce to higher temperatures all else being equal. I think it's possible that some of the extra present on the older cooler was not as effective on average, and the individual stacks vary in dimension.
Deciding to lower fan speed to reduce noise would reduce air volume and increase temps further, which could be part of why things are hotter.
It's possible there are other reasons for the reduced noise besides lowered airflow.

Higher local temperature, sure. But not less heat in totality. What I'm saying is that the SOC is going to produce the same amount of heat in both models of the PS5 because the SOC is the same. It's just going to be dissipated in a more localized area because there is only one fin stack. The video in question looks at a small area of the exhaust, sees higher temperatures, and concludes that the higher temperatures mean it's worse than the previous model. All I'm saying is that if an equal amount of heat energy is being exhausted from the system then both solutions are equal at what they were designed to do. And that it's possible that the newer design with the single fin stack is exhausting the same amount of heat, it's just in a more localized area. And that area would be expected to be of a higher temperature.

 

[3dilettante]

There is. This is a brand new model of fan for the PS5. The fourth.

Sony has been willing to mix its fan supply, so I wasn't sure if it was introduced with this revision.

Higher local temperature, sure. But not less heat in totality. What I'm saying is that the SOC is going to produce the same amount of heat in both models of the PS5 because the SOC is the same. It's just going to be dissipated in a more localized area because there is only one fin stack. The video in question looks at a small area of the exhaust, sees higher temperatures, and concludes that the higher temperatures mean it's worse than the previous model.

The infrared footage looks at a small volume of the case and metal that is in line of sight of the sensor. The exhausted air doesn't seem to be particularly emissive compared the the solid surfaces, so what I'm going by is limited to that superficial information.

The old heatsink had fins that stretched further from their heatpipe in the stack nearest the vent, and that extended metal was angled in a way that could have put more of that metal in the way of seeing the portion nearest the heatpipe.
So one portion of the increased temperature reading could be more visibility of hotter metal.
Another way to get hotter readings is if the stack is genuinely hotter, which can be because there's less area to dissipate heat, but this can scale independently from airflow.
Airflow, be it from fan speed or perhaps changing the cooling setup's flow pattern, could lead to different temperatures as well for the metal that it goes over. A quieter cooler could result from less air due tp a slower fan curve.

All I'm saying is that if an equal amount of heat energy is being exhausted from the system then both solutions are equal at what they were designed to do. And that it's possible that the newer design with the single fin stack is exhausting the same amount of heat, it's just in a more localized area. And that area would be expected to be of a higher temperature.

The cooler needs to reach a steady state at some point, otherwise heat would build up without bound. At steady state, energy in equals energy out, regardless of cooler efficiency. Dissipating the same amount of energy short of damaging the silicon wouldn't mean the coolers are equal.If Sony profiled that its thermal guard-band was too broad, the new revision could be accepting a hotter steady-state temperature for the cooler and components.

 

[see colon]

The cooler needs to reach a steady state at some point, otherwise heat would build up without bound. At steady state, energy in equals energy out, regardless of cooler efficiency. Dissipating the same amount of energy short of damaging the silicon wouldn't mean the coolers are equal.If Sony profiled that its thermal guard-band was too broad, the new revision could be accepting a hotter steady-state temperature for the cooler and components.

This is true, but we really don't know how the new cooler performs. We need temperatures from the SOC and the rest of the board to know how well it's actually cooling, or we would need to know how much heat is being moved outside the system. All we know right now is that part of the exhaust port is a few degrees hotter. Honestly, if I had real concerns for it, it would be in long term use after there is dust build up on it preventing good airflow.

 

[iroboto]

or we would need to know how much heat is being moved outside the system.

unfortunately not sufficient to determine SoC temperature. 3dilettante is correct, heatsink must reach steady state, and the heatsink absorbs the energy from the chip, the slower it is at dissipating the heat, the heatsink gradually goes up trying to match the heat of the SoC. If you run air over a hotter heatsink, the exhaust is hotter. In the opposite direction, if the thermal resistance is very low, the heatsink will transfer the heat off the SoC very quickly into the air, and the exhaust will be hotter and the SoC cooler.

 

[Deleted member 11852]

unfortunately not sufficient to determine SoC temperature. 3dilettante is correct, heatsink must reach steady state, and the heatsink absorbs the energy from the chip, the slower it is at dissipating the heat, the heatsink gradually goes up trying to match the heat of the SoC. If you run air over a hotter heatsink, the exhaust is hotter. In the opposite direction, if the thermal resistance is very low, the heatsink will transfer the heat off the SoC very quickly into the air, and the exhaust will be hotter and the SoC cooler.

I didn't quite follow the "heatsink gradually goes up trying to match the heat of the SoC", heat sinks typically aren't actively 'trying' to do anything, what you're seeing is just materials thermal properties at work. Materials with high thermal conductivity like copper are typically used because although the heat aperture (like the core of a semiconductor) is very small, the properties of the material mean the heat transfers through the whole of the heat sink block very efficiently.

For example if you expose the tip of one corner of a 2cm x 2cm x 2cm block of copper to 80 degrees celsius, the heat will transfer through the whole block very quickly and it will get very hot. If you use a much larger block of copper, such 20cm2, the heat will be mild because it's being distributed and passively dissipated from all surfaces area over a much larger mass. Increasing the surface area (exposed to air) is why some heat sinks have a prong-like shape because each prong increases the surface area.

In some aerospace applications we would sometimes use massive silver heat sinks as a passive heat solution because it was impossible for any amount of work generated by a component to completely saturate the mass of silver with enough heat that it would ever cause a problem. Some kind of airflow is always helpful though, if nothing else it brings down the size of the heat sinks overly a truly passive solution.

I have often wondered how much silver I'd need to completely passively cool my PC.

 

[wco81]

How far are they from a die shrink anyways?

Hard to believe this generation has only been less than a year. Seems like much longer, with all the news about how impossible thees consoles are to get.

With the chip shortages, maybe the timeline for die shrink is pushed further out.

 

[iroboto]

I didn't quite follow the "heatsink gradually goes up trying to match the heat of the SoC", heat sinks typically aren't actively 'trying' to do anything, what you're seeing is just materials thermal properties at work. Materials with high thermal conductivity like copper are typically used because although the heat aperture (like the core of a semiconductor) is very small, the properties of the material mean the heat transfers through the whole of the heat sink block very efficiently.

yea perhaps a poor way of just saying that the temperature of the heat sink will rise (given a long enough period of time) until the cooling amount reaches equilibrium with the heating amount. The heatsink could never surpass the temperature of the chip so the absolute hottest the heatsink could get is the temperature of the SoC itself, given there was a significantly more power than the heatsink could cool off and given enough time to sufficiently heat the materials: ie it would take concrete and stone very long to pass heat through, but would be much faster with conductive materials.

 

[Silent_Buddha]

I have often wondered how much silver I'd need to completely passively cool my PC.

A silver PC case would be beautiful. The only hard part with completely passively cooling a PC is the GPU. I've got an almost completely passively cooled PC (only the GPU has a fan which only kicks in when gaming). I do have an emergency safeguard fan on the CPU cooler just in case something (like a "power virus") were to trigger runaway temps on the CPU, but it's never had to kick in so far.

It would be interesting to convert everything to silver.

Regards,
SB

 

[Deleted member 11852]

How far are they from a die shrink anyways?

Probably years. Zen4 is 5nm and due to launch next year (2022) which is also when TSMC are starting major lines for 3nm production but only the likes of Apple will be using 3nm.

The barrier to die-shrinking a console APUs isn't so much technical as cost-related. The biggest reason to go smaller is heat savings and energy efficiency, neither are which are massive considerations for a home console, and the savings to power/thermal systems from having a more efficient chip will be less than the cost of the more expensive process.

And yeah, this generations feels long already. But the nice thing about using my PS5 and Series X is that I have no major wants from them. Had you asked you about PS4 or PS3 I'd have said I'd want a quieter console, or higher resolutions or better frame rates. This generation of consoles is just fantastic.

 

[orangpelupa]

Btw does all PS4 games now automatically got boosted to PS5 (like Xbox Series). Or it's still only boosted to ps4 pro?

My pc just died and now playing mainly on PS4 pro again and... Yeesh the frame rate is atrocious. Looking to grab yet another PS5 raffle

 

[Globalisateur]

Btw does all PS4 games now automatically got boosted to PS5 (like Xbox Series). Or it's still only boosted to ps4 pro?

My pc just died and now playing mainly on PS4 pro again and... Yeesh the frame rate is atrocious. Looking to grab yet another PS5 raffle

The vast majority yes. Some old (non cross-gen) Ubisoft game (like AC Unity, Ghost recon breakpoint or Division 2), supposedly don't.