Support NeoGAF

[Md Ray]

He was talking entirely about a patent when referring to the SRAM, I was talking about how it was depicted in Cerny's slides. That is where most of the confusion was.

I am now reading the patent and it so far is exactly as I said.

The entire first half of it so far is describing exactly the kind of system I was talking about. The mapping between logical addresses that the game and PS5 CPU uses, and the actual physical addresses of the flash memory that the controller is responsible for.

In the patent this is referred to exactly as logical and physical addresses, with the SSD controller consisting precisely of a flash controller that deals in physical addresses, a host controller that deals in logical addresses from the host (CPU/Game), and a fast pool of RAM used by both to store the address conversion table that connects the two.

It is so far nothing like as described by @psorcerer and is working exactly as it does on PC at the level of the SSD controller, which is all I was ever talking about. The host does not directly address the physical flash storage address range.

Usually there is an amount of DRAM built into high end SSD controllers for not just caching writes to improve write performance, but for keeping this mapping table very close to the SSD controller. It's not on chip because that's not how DRAM works. DRAM is also expensive, but traditionally you need 1 GB of it for every 1 TB of storage, just to keep that mapping table resident.

You need to keep that mapping table resident and close to reduce latency, which directly affects random read performance. Something that will be common in consoles now they're using SSD and don't want to duplicate data.

DRAMless SSDs in PCs were designed for things like laptops where you want to use less energy (DRAM costs energy), and also save costs in manufacturing. Not only do they suffer in write performance, they also suffer in random read performance, as demonstrated in the bench-marking article I linked.

Now what's interesting, and what I didn't realise until I started reading this patent, is that Sony has been experimenting with coarser mapping tables, and having them completely or partially resident not just in SSD controller DRAM, but on-chip itself in SRAM, for even better latency improvements.

The patent is generic enough in scope in that it may or may not relate to PS5. The SRAM as shown in Cerny's slide is clearly outside of the SSD controller, so that isn't it, but there's still a chance PS5's SSD controller is using this technology, which is a step above even a DRAM equipped SSD in latency.

The idea that an SSD would keep its flash mapping table resident in system RAM is nonsense.

Still going through the patent and it's extremely interesting so far. Considering who provided the link to it, it's unsual they didn't seem to understand the point I was making, which is made exactly in the opening summary of the patent.

tl;dr "DRAMless SSDs" don't just suffer from poor write speeds, but also random read performance. They are budget mobile devices. Random read will be leveraged by SSD equipped consoles to get rid of data duplication. From the patent provided and the concerns it lays out in the opening, it's unlikely Sony would have opted for a "DRAMless SSD" solution, and they may have even gone a step further and be using a pool of SRAM in the SSD controller itself to further reduce latency over DRAM equipped SSDs.

EDIT: The idea seems to be that the controller can load in different complete or partial mapping tables saved in flash depending on what it's doing, hence it can get away with a small but ultra fast SRAM pool, rather than a large and slightly more distant (but still before the main SoC) DRAM pool.
As game packages are basically copied from Blu-ray into flash and aren't rewritten to, it could be that a mapping table related to the game package gets loaded into that pool. Maybe it gets sent an address and an offset to get something more specific. Still reading. The point is it doesn't need the entire 825 GB mapping table to fit into a smaller SRAM pool if the only random reads likely to be happening relate to an area that isn't really ever changing.

Thanks, what an interesting read. I really appreciate posts like this.

Now, I wonder what the size of this SRAM would be? 16MB, 8MB, or even less?

 

[MasterCornholio]

What image?

This good guy Phil image that he has. Alot of people still see him this way. Supported these individuals could change that for some of them.

 

[Md Ray]

5 hours who is excited ?

 

[FeiRR]

Doesn't it look like it won't be able to stand on its side (aka vertically)?

It doesn't look so but you'll probably be able to park your bike in it.

 

[BadBreathOfTheWild]

if couple of percent is worst case then it would dip around 2186Mhz for GPU giving 10.07 Tflops i would not say 44Mhz is a few but i guess this is worst case scenario reducing powerdraw by 10%.

The point is it's not like you hit the power-draw ceiling and then decide to run at 90% of it for an arbitrary amount of time. You scale the clocks down to stay at 100% power usage.
The "couple of percent" drop in clocks uses 10% less power is just a metric of how little you need to actually vary the clocks to stay within the power threshold. There's no need to drop down 44Mhz to then also be dropping your power consumption down to 90% of maximum.
The system targets a fixed peak power draw. That means a fixed peak amount of calculation, regardless of the clocks fluctuating.
How much the clocks need to drop for a high CU occupancy task to stay at 100% isn't known, but it doesn't need to be anywhere near 44Mhz from the paradigm Cerny is explaining.
It really does make sense that the GPU will be spending most of its time at or close to 2.23Ghz if they are doing exactly what they say they are doing.
Cerny mentioned a game that was using a full frame of GPU drawing that stayed at peak clocks in the Eurogamer article.

Anyone that has overclocked a modern GPU knows that you aim for enough cooling to be limited by peak power draw. They'll also know that dropping your core clock even by 2% isn't even really measurable in game, and barely detectable in a stress test.

PS5's variable clock system--if implemented as described in Road to PS5--will target a fixed and unvarying amount of calculation done per unit of time. Clocks varying doesn't mean work done varying, as it's a combination of clock-rate AND the instructions being run that draws power, creates heat and actually does useful work.

 

[Kusarigama]

5 hours who is excited ?

Are we expecting a controlled leak or something?

 

[KURDISTAN YPG]

So fellas, i have 2 year 7.2 channel denon aV, Do you guys think i should upgrade n be ready for ps5?
Im a audio freak but denon i have atm does not support hdr i dont think.

 

[SSfox]

5 hours who is excited ?

 

[geordiemp]

if couple of percent is worst case then it would dip around 2186Mhz for GPU giving 10.07 Tflops i would not say 44Mhz is a few but i guess this is worst case scenario reducing powerdraw by 10%.

And how many milliseconds does it take to switch and for how many milliseconds will the lower clock be active.

What percentage of a frame ? All we know is AMD smart shift is 2 ms, so 8 times a frame, but we dont know what sony has done YET for those predicted high heat workload sets.

 

[BadBreathOfTheWild]

Thanks. It was an interesting read and I really appreciate posts like this.

Now, I wonder what the size of this SRAM would be? 16MB, 8MB, or even less?

It will depend on PS5's particular implementation. The patent is talking about a method of getting data quickly from flash storage into an application, rather that specifics. The patent for example uses a 4-channel controller to keep things simple.

The SRAM size isn't that critical with how they're managing the translation table data, and how they're storing a file on the flash chips. A file request at the application side eventually gets turned into a load of read requests at the flash controller with each chunk being wholly loaded into the SRAM before undergoing an ECC check and then being shipped off to system RAM waiting to be decompressed etc.

This patent is aggressively targeting latency and random read performance just as much as it is overall read speed. That 5.5 GB/s figure isn't telling half of the story. Neither really is the 8-9 GB/s. Think satellite broadband at 10mbit versus DSL broadband at 10mbit. Both will download a single 100MB file in roughly the same time. One will struggle with lots of small files being requested one after the other. One will also struggle with latency critical applications like gaming or VoIP.

The latency gains Sony have made if they're using what's in that patent are just as much dependent on hardware as they are new API.

The patent discusses ways of improving overall speed, minimising latency, maximising random read and minimising CPU overhead so the CPU can just hand off a task and be alerted when the data is ready for it. Super efficient and extremely optimised. The File Archive API that bypasses traditional virtual file-systems and talks directly to the host controller of the SSD controller is extremely low level file IO API.

It is demonstrably a complete myth that Sony only focused on hardware solutions to fast IO in lieu of software optimisations if that 2017 patent is what PS5 is based on.

 

[xacto]

This good guy Phil image that he has. Alot of people still see him this way. Supported these individuals could change that for some of them.

I don't think Spencer has any image to save, not since he's patting Tim&Co on the back.

 

[Corndog]

WIPO - Search International and National Patent Collections

Doesn’t seem like they show the ram in this patent. Just shows cooling of apu and power supply as far as I can tell.

 

[FeiRR]

That patent doesn't look like "a cutting edge and very expensive cooling solution", like the rumours were saying. Maybe it's just for the first devkits that were reported to be loud?

 

[BadBreathOfTheWild]

That patent doesn't look like "a cutting edge and very expensive cooling solution", like the rumours were saying. Maybe it's just for the first devkits that were reported to be loud?

Multiple fans and two large heatsinks(?) is extremely expensive compared to typical console solutions. This isn't in the context of gaming PCs, but budget boxes like consoles.

 

[geordiemp]

Multiple fans and two large heatsinks(?) is extremely expensive compared to typical console solutions. This isn't in the context of gaming PCs, but budget boxes like consoles.

I ony see 1 heat sink and 2 sets of fans.

Looks like they are pulling air over the API heatsink to the left, and pulling another stream of air over other power componentsto teh right, splitting it up. But there is no heatsink on the right we cna see.

Maybe the consumer unit will do something different.

 

[BadBreathOfTheWild]

I ony see 1 heat sink and 2 sets of fans.

Looks like they are pulling air over the API heatsink to the left, and pulling another stream of air over other power componentsto teh right, splitting it up. But there is no heatsink on the right we cna see.

Maybe the consumer unit will do something different.

I'll take a closer look! Either way, 6 fans is expensive compared to one large one! If a large one at console manufacturing scales is $1 or less, then 6 is comparatively splashing out! Cerny only said we'd be happy with what the engineering team came up with. Happy to me is quiet and low maintenance. I don't care how it does it particularly.

 

[3liteDragon]

What image?

 

[BadBreathOfTheWild]

That patent doesn't look like "a cutting edge and very expensive cooling solution", like the rumours were saying. Maybe it's just for the first devkits that were reported to be loud?

Got any sources on the "cutting edge" part of that? I've heard it's comparatively more costly, but haven't heard any rumours saying it's cutting edge myself yet.

 

[MasterCornholio]

5 hours who is excited ?

 

[Corndog]

Thanks, what an interesting read. I really appreciate posts like this.

Now, I wonder what the size of this SRAM would be? 16MB, 8MB, or even less?

SRAM is very expensive I assume it is very small.

 

[Reficul]

Yet again I wish Sony had some marketing phrase similar to "Velocity Architecture" to refer to their storage architecture, just so the whole thing stops getting referred to merely as "PS5's SSD".

Maybe Sony has a changed the meaning of the acronym. Maybe SSD is Sonys Supernatural Datatransfer?

Probably not...

 

[THE:MILKMAN]

That patent doesn't look like "a cutting edge and very expensive cooling solution", like the rumours were saying. Maybe it's just for the first devkits that were reported to be loud?

The other weird thing with these patent releases today is why they would time the release of it on the day of the original reveal date but be of the first dev kit version. Apparently there is a new slightly different revision out now.

I just hope the retail console doesn't look like this or have 6 fans! That would be loud........

 

[Bladed Thesis]

COD: Modern Warfare is now at 214 GB, and with the update coming up, it will be closer to 240 GB. And that's just for 1 game. It's egregious!

Holy shit, I did not realize the game is that big. Is that due to the standalone BR mode? Dear god.

 

[FeiRR]

Got any sources on the "cutting edge" part of that? I've heard it's comparatively more costly, but haven't heard any rumours saying it's cutting edge myself yet.

Unfortunately it's just Bloomberg. I find it hard to believe that a retail console can have 6 fans.

Holy shit, I did not realize the game is that big. Is that due to the standalone BR mode? Dear god.

Cancer grows fast.

 

[BadBreathOfTheWild]

Unfortunately it's just Bloomberg. I find it hard to believe that a retail console can have 6 fans.

The entire Bloomberg quote seems to be:

Most of the components for the console have been locked down, the people said, including the cooling system, which is unusually expensive at a few dollars per unit. Typically, companies would spend less than a dollar, but Sony opted to lavish more on making sure heat dissipation from the powerful chips housed inside the console isn’t an issue.

This just suggests it costs a few dollars per unit instead of one. Nothing at all about being cutting-edge. I’d seen the “lavish” word thrown about before but it’s being used in this context to describe spending more money; not a description of the hardware.

Multiple fans is a lot more expensive than a single fan solution.

Multiple fans is also the only way to ensure a high enough volume of air passing through a heatsink if you want to keep the console form factor more constrained in some way.

Xbox Series X’s absolute chonk comes from the fact its form factor is build around a fan. It’s cheaper and more efficient to do that, at the cost of console size in pretty much all axis.

I don’t know where this “cutting edge” cooling is coming from. It’s not what Sony or this analyst has said. It’s probably evolved from information being passed around and embellished by fans and others enough until it becomes something different and somehow also canon.

 

[Metroiddarks]

WIPO - Search International and National Patent Collections

Good finding ... 6 ventilators wtf

I trust Dolby over Cerny. Dolby state that they support more than 32 sources, but they dont recommend it.

I trust in both in the end Cerny is working for a product for Sony one main companies in the world related to the sound/music business. I this is more probably
Sony don't want to use Dolby for license and similar things, also the are proud of use its technology.

The technology of Sony doesn't require your headphones be certified by Dolby.

 

[LordOfChaos]

Newly published Sony patents on a cooling solution involving 6 fans to manage the balance between CFM and noise... Would be odd to patent something that's only used for dev kits, eh?

WIPO - Search International and National Patent Collections

 

[JareBear: Remastered]

Keep hoping the new PS5 event date pops up on my timeline somewhere.....

I need it

 

[geordiemp]

I'll take a closer look! Either way, 6 fans is expensive compared to one large one! If a large one at console manufacturing scales is $1 or less, then 6 is comparatively splashing out! Cerny only said we'd be happy with what the engineering team came up with. Happy to me is quiet and low maintenance. I don't care how it does it particularly.

6 small fans at total 3 bucks maybe or even less, dont think its a big deal. As long as it performs and is not a jet.

 

[Neo Blaster]

That's another reason why the wanted today presentation. Again next week APU's are starting being delivered and whole production will start for good. At this etap of production leaks are inevitable, just ask Apple CDP escaping from 11, and other event changes are making me fell comfortable about 11-12 as new event date. And im pretty sure we are getting sneak peek at box itself.

If Sony wait beyond next week the form factor will definitely leak, they will have to take action and fast.

 

[JareBear: Remastered]

It's a bummer that the Future Games Show got delayed from Saturday. Was looking forward to more Quantum Error footage. Understandable though.

Tough time right now. Hopefully things get better

 

[Andodalf]

Newly published Sony patents on a cooling solution involving 6 fans to manage the balance between CFM and noise... Would be odd to patent something that's only used for dev kits, eh?

WIPO - Search International and National Patent Collections

Plenty of patents never come to life, but this could happen for sure. The fans do look small, which makes sense given that their are so many, which is a bit concerning as smaller fans can be whiny. Should be cool AF if it pans out

 

[FeiRR]

6 small fans at total 3 bucks maybe or even less, dont think its a big deal. As long as it performs and is not a jet.

One fan like that in X360 was enough to piss me off totally. It was a crappy Sunnon thing and I hated it wholeheartedly. I think I prefer a jet engine than a high-pitched whine.

 

[LordOfChaos]

Plenty of patents never come to life, but this could happen for sure. The fans do look small, which makes sense given that their are so many, which is a bit concerning as smaller fans can be whiny. Should be cool AF if it pans out

Hopefully they avoid that. I would guess the solution means they can maintain a lot of static air pressure on the heatsink as the left side enclosure is literally all fans and heatsink, no shrouds needed, no other components in the way, so they can probably keep the RPMs low and the pressure high.

Also with everything else that's been talked about by Cerny too, i.e a fixed power target rather than a fixed clock speed, so probably less cycling up and down of fans that current consoles do and staying closer to constant.

 

[DaGwaphics]

6 small fans at total 3 bucks maybe or even less, dont think its a big deal. As long as it performs and is not a jet.

I think many look at retail CPU coolers and judge costs based on that. The cooling system in the 1X is probably $10 max, and that's likely an overshoot. LOL

You could have 8 fans and still be near silent, it's all about the RPMs, especially with smaller fans. If it's six fans similar dbs to the 360 fans... JK

 

[thebigmanjosh]

Wait until you read through the VR and controller patents:

 

[Bonsaibäumchen]

6 small fans at total 3 bucks maybe or even less, dont think its a big deal. As long as it performs and is not a jet.

As far as I know having more bigger fans at lower speeds is better for cooling and staying quite. Though you obviously have to take care of vibrations or it won't help at all.

 

[Dee_Dee]

Event will probably happen next week. They pretty much have to or everything will just leak. I’m fine with either option

 

[BadBreathOfTheWild]

Hopefully they avoid that. I would guess the solution means they can maintain a lot of static air pressure on the heatsink as the left side enclosure is literally all fans and heatsink, no shrouds needed, no other components in the way, so they can probably keep the RPMs low and the pressure high.

Also with everything else that's been talked about by Cerny too, i.e a fixed power target rather than a fixed clock speed, so probably less cycling up and down of fans that current consoles do and staying closer to constant.

I agree. If they know what power usage they’re dealing with and therefore where the fans will spend the majority of their time regardless of game, they can really fine tune the acoustics of it in that zone.
More smaller fans gives flexibility on form factor more than anything else I think.

 

[Metroiddarks]

WIPO - Search International and National Patent Collections

What's this?

Is not PS5 related.

 

[geordiemp]

I think many look at retail CPU coolers and judge costs based on that. The cooling system in the 1X is probably $10 max, and that's likely an overshoot. LOL

You could have 8 fans and still be near silent, it's all about the RPMs, especially with smaller fans. If it's six fans similar dbs to the 360 fans... JK

It depends, apple used lots of small fans and changed the blades to variable pitches to get a broader range of frequencies so less whine / noise of the same pitch. That would also be cool.

Maybe they have sound engineered the cavity and chosen fan blades to match for each fan (maybe going a bit far lol), but this is CERNY

 

[BadBreathOfTheWild]

If it's six fans similar dbs to the 360 fans...

...then the pictures of our ears we’ll be texting Cerny will have blood flowing from them

 

[Three Jackdaws]

Cerny seemed pretty confident and happy about the Ps5 cooling solution if you watch the tech talk again, also rumours are that the PS5 will operate "whisper quiet" (according to SpawnWave and Foxy, yes I know they are not reliable lol) which sounds plausible given Cerny's remarks.

 

[FranXico]

I take it moving on to the audio chips battle is an admission of IO speed defeat.

 

[geordiemp]

LOL Mister Xmedia talking about GFops, wonder what they are, maybe in the powerbrick.

It reads like his typing, I wonder if he also has insider there. God I loved me some mister x, comedy gold.

 

[Three Jackdaws]

I see that Mark Cerny's tech talk is ageing like fine wine!

 

[FeiRR]

I take it moving on to the audio chips battle is an admission of IO speed defeat.

But there's also some weeping in the corners. I guess we've just discovered the name of our hype train.

 

[LordOfChaos]

 

[MasterCornholio]

But there's also some weeping in the corners. I guess we've just discovered the name of our hype train.

Oh dear.....

 

[Leo_Barcellos]

"Mark Cerny's dick" hahahahaha