Playstation 5 [PS5] [Release November 12 2020]

The impact was described in terms of bandwidth, rather than latency. I think Sony should be profiling this, since there can be large disparities in latency between the best/average case and 99th percentile depending on the quality and fullness of the drive. Can the PS5 assert itself over the expansion drive's potentially inconsistent firmware and controller, or would this mean that Sony's latency promises are not as iron-clad as the bandwidth ones?
(edit: corrected wrong word for bandwidth)
Agree there’s a lot unclear. This is why I’ll wait after launch for benchmarks on various drives and just live with the ~700GB for a while.
 
I'm actually curious to see how game sizes compare as we move to next-gen. There has been a lot of talk on better compression and less asset duplication with SSD's. But with customer expectations for next-gen fidelity in terms of resolution and geometric detail, I can only see them getting significantly bigger. Historically physical media set those caps, but I'm not sure those are going to hold through the generation. Maybe Internet data caps will limit customer willingness to purchase games requiring additional downloads of enhancement packs and that won't become a thing. But I can see a future where games require additional downloads to get the best experience.
 
Can the PS5 assert itself over the expansion drive's potentially inconsistent firmware and controller, or would this mean that Sony's latency promises are not as iron-clad as the bandwidth ones?
at first I took the values at face value until I started looking into NVMe throttling. I don’t know what the tech is Sony built so we don’t know if they will throttle their SSD if heat is too high. We have no information on how it behaves. I just know that the more bandwidth a NVMe can provide the hotter the controller gets. And PS5 is way beyond what we have on PC by magnitudes order in transfer speed it only makes sense heat and power will go with it. They didn’t provide a guaranteed bandwidth number, just optimal performance, so I’m left to ask if there are safeties here.

And if they don’t throttle it; do they have data on when the silicon will fail if the heat is excessive? another possible RROD type situation. The hardware is bespoke; it can’t be replaced; so it must last the life of the console (7 years without dying). These are larger challenges to solve. Throttling is the only thing I can think of and a really good cooler.
 
at first I took the values at face value until I started looking into NVMe throttling. I don’t know what the tech is Sony built so we don’t know if they will throttle their SSD if heat is too high. We have no information on how it behaves. I just know that the more bandwidth a NVMe can provide the hotter the controller gets. And PS5 is way beyond what we have on PC by magnitudes order in transfer speed it only makes sense heat and power will go with it. They didn’t provide a guaranteed bandwidth number, just optimal performance, so I’m left to ask if there are safeties here.

And if they don’t throttle it; do they have data on when the silicon will fail if the heat is excessive? another possible RROD type situation. The hardware is bespoke; it can’t be replaced; so it must last the life of the console (7 years without dying). These are larger challenges to solve. Throttling is the only thing I can think of and a really good cooler.


But with a dedicated design, they can put a decent cooler on the controller, vs the inexistant space in most nvme drive for PC.
 
  • Like
Reactions: jgp
This a case the rasterizer is at maximum efficiency. Simple geometry equals to maximum rasterizer efficiency. This is the reason Mark Cerny said simple geometry push further the GPU than complex geometry. This is as simple as that but many people seems to forget how rasterizing work.

optimizing-the-graphics-pipeline-with-compute-gdc-2016-51-638.jpg
This reminded me of something.

When using VRS to get maximum coarse size shading of 4x4 pixels, the 2x2 quad of rasterizer becomes 8x8 pixels on screen.
Pretty sure this would mean that small polygon problem becomes worse.
 
But with a dedicated design, they can put a decent cooler on the controller, vs the inexistant space in most nvme drive for PC.
I’m more concerned about whether it will throttle or not under heat. If it throttles then the throttling speed is the most that can be used for games in terms of design. Like if you have throttling and you build your game expecting 5.5GB/s but some consoles are
Running hot and only getting 1.5-2.0 Gb/a due to throttling that’s an issue.

combined with their SoC; heat is a big challenge for PS5. A smaller GPU with less
CUs means higher occupancy; running a higher clock rate. That just means significantly more heat. Then you have SSD speeds that are up to 2x better than what is available on the market with higher speeds translating to more heat. Nvme drives can hit over 100C.

While I believe all of this can be handled; I don’t believe cooling will be cheap as a result
 
Last edited:
I'm kind of concerned about the soc, but not the ssd controller. This is 2 differents chips, right ? If so, it will have its own radiator and it will be fine, if the case allow decent air flow. MS could have the same problem btw. it's pretty packed in here.

I'm really curious to know more about PS5 cooling anyway. And if it's not revealed yet, my guess is they are still tweaking it.
 
A smaller GPU with less CUs means higher occupancy;
Only for the same workloads. If a dev is getting 90% utilisation out of PS5, they should also be getting 90% utilisation out of XBSX and doing more on screen with that. The heat difference will be based on how much more heat is created at the higher clocks over the wider design.

52 CUs at 1.8 GHz will set our XBSX baseline heat output. 36/52 is 70% heat from PS5 based on width. Now if temp increases linearly with clocks, 2.2/1.8 would be 22% more heat, so 0.7 * 1.22 = 85% heat output versus XBSX. So it's only the exponential ramp of heat that risks pushing PS5 over the same heat output of XBSX. We'd need that 22% more clock speed to result in over 40% more heat to overtake MS on heat output.
 
I’m more concerned about whether it will throttle or not under heat. If it throttles then the throttling speed is the most that can be used for games in terms of design. Like if you have throttling and you build your game expecting 5.5GB/s but some consoles are
Running hot and only getting 1.5-2.0 Gb/a due to throttling that’s an issue.

Not saying this won't happen but it would be a massive engineering failure.
 
Don't think cooling the SSD will be much of a problem, MS seems to be able to cool their tiny SSD memory cards with a very small heatsink embedded on them, while they are somewhat slower, they are also smaller. Sony could put a bigger heatsink on their SSD.

I hope Sony has made the expansion bay large enough to accommodate large heat sinks that may be on 7 GB/s versions. :rolleyes:. He said not everyone will fit.
 
I hope Sony has made the expansion bay large enough to accommodate large heat sinks that may be on 7 GB/s versions. :rolleyes:. He said not everyone will fit.

Not unknown territory for Sony to include large expension bays. The fat PS2's drive bay takes up a rather large amount of the total PS2 casing.
 
Only for the same workloads. If a dev is getting 90% utilisation out of PS5, they should also be getting 90% utilisation out of XBSX and doing more on screen with that. The heat difference will be based on how much more heat is created at the higher clocks over the wider design.

52 CUs at 1.8 GHz will set our XBSX baseline heat output. 36/52 is 70% heat from PS5 based on width. Now if temp increases linearly with clocks, 2.2/1.8 would be 22% more heat, so 0.7 * 1.22 = 85% heat output versus XBSX. So it's only the exponential ramp of heat that risks pushing PS5 over the same heat output of XBSX. We'd need that 22% more clock speed to result in over 40% more heat to overtake MS on heat output.
Thinking out loud:
  • I think temps are directly correlated with power usage over the area of the surface, in which this case a smaller chip has more more heat per mm^2 compared to a larger chip.
  • Regarding clock speeds and voltage, that relationship is nearly exponential/logarithmic in shape.
  • Regarding load, it should be #pixels * frame rate
When you look at these topics individually, I don't think the argument is _only_ at the same workload. Because of the diminishing returns of clockspeed/voltage, or exponential depending on how you want to look at the graph, once you hit the optimum you are increasingly getting less return on frequency for each mV you are putting in. So even on lesser loads, you could be working through more power.

From wikipedia:
While performance per watt is useful, absolute power requirements are also important. Claims of improved performance per watt may be used to mask increasing power demands. For instance, though newer generation GPU architectures may provide better performance per watt, continued performance increases can negate the gains in efficiency, and the GPUs continue to consume large amounts of power.[32]

The efficiency of some electrical components, such as voltage regulators, decreases with increasing temperature, so the power used may increase with temperature. Power supplies, motherboards, and some video cards are some of the subsystems affected by this. So their power draw may depend on temperature, and the temperature or temperature dependence should be noted when measuring.[34][35]

So Cerny said the system was deterministic so that heat would not play a factor. But clearly that can't be true as per physics. There are thermal limits. We run 2 PS5s side by side and alter the heatsink to be woefully inadequate on one; I expect it to shut down.

If I make a PS5 operate in a small enclosed space (air tight box), I expect it to run into issues with voltage regulators eventually. So I'm curious to see what will happen in this scenario, will it down clock or will it shut down. If the latter that is deterministic, if the former then it is not. It is an easy test when the full retail units are released.

Not saying this won't happen but it would be a massive engineering failure.
Not necessarily. It could just mean more costs for cooling.

Don't think cooling the SSD will be much of a problem, MS seems to be able to cool their tiny SSD memory cards with a very small heatsink embedded on them, while they are somewhat slower, they are also smaller. Sony could put a bigger heatsink on their SSD.
Then they should guarantee the bandwidth if it's so trivial. MS worked with Seagate to develop a custom SSD that would guarantee 2.4GB/s as well as their external drives which are not 3rd party NVMes. That means they've already done their testing and it's proven for developers that they can rely on the 2.4 GB/s for whatever they want it for and for whatever features it support. 3P hardware nvme devices all throttle when in trouble with heat this is out of control for Sony. In reference to an older lower performing one:
thermal_write.jpg

So Sony must approve only drives that will be able to maintain their 5.5GB/s under heavy load while not throttling in the PS5 bay. There will be additional design considerations to support cooling this bay without necessarily knowing if users will have something in there.

I do not think it is easy to guarantee something as high as 2.4GB/s bandwidth in a small form factor, as nothing I can see (a laptop with a m.2 nvme slot) will support a nvme drive this fast. As written in this article by Techspot (https://www.techspot.com/review/1893-pcie-4-vs-pcie-3-ssd/)
"Even with the heatsinks on, the drives may still throttle if they are under sustained load for more than about 15 minutes." I would see this as a warning flag for trying to use the SSD as a RAM, or if to be used as RAM the customizations and cooling need to be significant to support this level of operation for play times as long as 3 hours.

These SSDs have to last the lifespan of the console, so how it will behave will be critical and I don't think it's a topic I would just say, hey it should be easy if MS can do it. MS did do it, and they did it at a cost of bringing the speeds to nearly 1/2 of what is offered out there, but they guaranteed it. And likely I suspect the plan is for it to last the lifespan of the console.

Or to put plainly; if it's so simple to guarantee performance consistency by slapping on more cooling, then why go with a reduce speed of nearly 1/2 of what is on the market today? Or are there factors we are not yet aware of. (though this could be linked with the fact that they wanted to do an external expansion port)
2019-09-16-image.jpg

I hope Sony has made the expansion bay large enough to accommodate large heat sinks that may be on 7 GB/s versions. :rolleyes:. He said not everyone will fit.
Yea, I didn't think about this until after you mentioned it.


These are not real issues though. It can be waved away, but it will be done so with higher costs.
 
Last edited:
Yes, those 3TB drives are usually really old tech. I saw my NAS speeds doubled when I replaced some 7200 RPM 3TB drives with newer 4TB drives. Even speeds from my 5900 RPM Shingled 8TB drives are more than double what I got from the 7200 RPM 2TB to 3TB drives.
 
at first I took the values at face value until I started looking into NVMe throttling. I don’t know what the tech is Sony built so we don’t know if they will throttle their SSD if heat is too high. We have no information on how it behaves. I just know that the more bandwidth a NVMe can provide the hotter the controller gets. And PS5 is way beyond what we have on PC by magnitudes order in transfer speed it only makes sense heat and power will go with it. They didn’t provide a guaranteed bandwidth number, just optimal performance, so I’m left to ask if there are safeties here.

And if they don’t throttle it; do they have data on when the silicon will fail if the heat is excessive? another possible RROD type situation. The hardware is bespoke; it can’t be replaced; so it must last the life of the console (7 years without dying). These are larger challenges to solve. Throttling is the only thing I can think of and a really good cooler.

In absolute terms, the SSDs are likely small consumers, with existing drives consuming 3-5W. A doubling of that shouldn't be a challenge for existing cooling tech, if Sony allotted the necessary airflow and heatsink capacity.
Long-latency events can come from poor garbage collection or wear-leveling algorithms, or issues with flash management, which would generally be the responsibility of the controller. If Sony's vetting process doesn't catch poor long-term performance consistency, then games are going to have unexpectedly choppy performance as users fill up the drives or time passes. If Sony's controller can override more of the controller's usual duties, that might be something they could address (more likely than getting SSD vendor firmware patches onto a console).
This assumes the Sony is more adept at handling the issues that stymie experienced SSD vendors, which is where there could be an additional risk factor with Sony's native drive, depending on how much of the engineering and design was done by them or a major controller vendor.
 
Thinking out loud:
  • I think temps are directly correlated with power usage over the area of the surface, in which this case a smaller chip has more more heat per mm^2 compared to a larger chip.
  • Regarding clock speeds and voltage, that relationship is nearly exponential/logarithmic in shape.
  • Regarding load, it should be #pixels * frame rate

Also reading from flash needs 50% of voltage compared to writing, AFAIR.
And even more and longer for erase.
All of the current PC SSD drives are heavily optimized for a considerable write pressure.
I suspect that console write pressure is much lower.
 
Also reading from flash needs 50% of voltage compared to writing, AFAIR.
And even more and longer for erase.
All of the current PC SSD drives are heavily optimized for a considerable write pressure.
I suspect that console write pressure is much lower.
yea I have confidence as a typical streaming type functions all should be good. Using it as 'RAM' might be more involved.
 
Back
Top