• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Next-Gen PS5 & XSX |OT| Console tEch threaD

Status
Not open for further replies.
It be here...



L0 cache 2 per dual cu each 16kb 32 way
L1 cache for each 5 dual cus 128kb 16 way
L2 4mB 16 way

This has 40 cus ( 4x10) ... most likely upgrade is to add one dual cu to each group = (4x12) 48 cus .. or add a whole extra group = (5x10) 50 cus ... or do both = (5x12) = 60

56 active seems likely

If you add a whole new group you will add more latency and wont be able to clock as high.


what you refer to as a "group" is called shader array in RDNA. i've drawn a navi 10 diagramm back in august to show what's what:

navi10dielayout28oj25.png



the problem with that

.. or add a whole extra group = (5x10) 50 cus ... or do both = (5x12) = 60

is, that each shader array shares parts of it's front end with another shader array to form a shader engine. so you had to add a whole new shader engine. while that would be nice, that die would become rather big at that point.


so in the gonzalo thread i speculated about adding four CUs per shader array, which would also make spacial sense. that would give 56CUs of wich you would likely have to dissable some for yield reasons:

consoledielayoutqojie.png


we don't know, how dCUs are disabled in RDNA. in GCN you had to disable one CU per shader array to maintain load symmetry. it not clear if this is still necessary with RDNA.

so there are lot of options for such a layout. and going by navi 14 there are other spacial layout options as well that could make sense.


If you add a whole new group you will add more latency and wont be able to clock as high.

i don't think that this is necessarily true. there's no clock speed disadvantage between navi 10 and 14. the challenge would be to keep the additional CUs fed. but thats more a front end / logic issue as long as you have sufficient bus bandwidth.

the point why larger dies often have lower clocks is, because the probability of less than stellar silicon is higher and even more than that heat density is becoming critical in such dies.
 

demigod

Member
Cant you read or understand?

That pic is from scenario with OPTIMIZED nvme, not from your bullshit PC.

Are you really so self-centered that you think that your PC = only thing on this world?

Point is that on "normal pc world" software must run on slow ass HDD, SSD and NVME.

So, it is not optimized to take all benefits from nvme and be as fast as it is possible.

On consoles they dont have to think of somebody with ancient HDD so software/games can be coded that in mind.

Same with PC, if there would be only super fast nvme drives = things would be faster than now with mixture of slow and fast drives, even when using that fastest one.

How it can that hard to understand?

On motorway there can be slow cheap cars and expensive super cars, all have the same max speed under the rules. Does that mean that fiat Punto is fast as ferrari on a race track? Aka "average conditions vs optimized"

PC OS/software just probably cant keep up with nvme, so when ssd already maximises speed that software can handle, adding more speed doesnt show up unless making software to take advance of that additional speed.

I have noticed real differences between ssd and nvme on my pc, so it can already happen in the right conditions

Lmao, I didn’t even realize he edited his post and talked about system boot for pc when that pic is talking about CONSOLES.
 

Lort

Banned
what you refer to as a "group" is called shader array in RDNA. i've drawn a navi 10 diagramm back in august to show what's what:

navi10dielayout28oj25.png



the problem with that



is, that each shader array shares parts of it's front end with another shader array to form a shader engine. so you had to add a whole new shader engine. while that would be nice, that die would become rather big at that point.


so in the gonzalo thread i speculated about adding four CUs per shader array, which would also make spacial sense. that would give 56CUs of wich you would likely have to dissable some for yield reasons:

consoledielayoutqojie.png


we don't know, how dCUs are disabled in RDNA. in GCN you had to disable one CU per shader array to maintain load symmetry. it not clear if this is still necessary with RDNA.

so there are lot of options for such a layout. and going by navi 14 there are other spacial layout options as well that could make sense.




i don't think that this is necessarily true. there's no clock speed disadvantage between navi 10 and 14. the challenge would be to keep the additional CUs fed. but thats more a front end / logic issue as long as you have sufficient bus bandwidth.

the point why larger dies often have lower clocks is, because the probability of less than stellar silicon is higher and even more than that heat density is becoming critical in such dies.

Thank makes sense i did cpu design at uni a long time ago so im not sure what elements are still important.
So you think maybe 52 or 54 available for use? ( from 56)
 
Last edited:
As much as anyone wishes for Reram, I think it's too early for it to be realistic. Maybe next-next-gen.

No man. I am almost convinced now that they will use ReRAM. That's the only way they can achieve that "no loading times" that has been said and reiterated many times over.
 

Ar¢tos

Member
No man. I am almost convinced now that they will use ReRAM. That's the only way they can achieve that "no loading times" that has been said and reiterated many times over.
To launch a decent spec console (12tf+,16gb+ ram) with Reram and 1tb ssd, Sony would take a big loss per unit (150+$, selling at 500). I think that even with Playstation being the 2nd most profitable division (insurance is the 1st, iirc), the whole Sony wouldn't approve that after the 350$ loss per unit fiasco of the ps3.
 
To launch a decent spec console (12tf+,16gb+ ram) with Reram and 1tb ssd, Sony would take a big loss per unit (150+$, selling at 500). I think that even with Playstation being the 2nd most profitable division (insurance is the 1st, iirc), the whole Sony wouldn't approve that after the 350$ loss per unit fiasco of the ps3.

Again, we cannot discount the fact that the professional analysis on the price of ReRAM (which is a third of Optane, and equal to SLC Nand) might be true.

Of course it's totally fair to think that it will be expensive because it's a new technology and we don't know yet. But there are data that points to it to be not that expensive as we think both near-term and more importantly long-term.
 
If the graph from the professional analysis we've seen on ReRAM is actually accurate, then we might be looking at $25-$30 for 128GB of ReRAM.

And then the fact that ReRAM will scale better than Nand. It it actually expected to replace Nand Flash in the future because it has the potential to become cheaper.
 
Last edited:

psorcerer

Banned
So you expect the games stored on our SSD next-gen are already decompressed? How big of space do you expect our next-gen uncompressed game occupy?

Do you not see a cache approach from where uncompressed data could stream?

Yep. I expect decompressed but encrypted data. I also expect some sort of tpm and hw decryption engine.
I would also expect loading in-place. I.e. data on disk and in memory being exactly the same data.
Overall I expect PS5 to be a data streaming machine. Like PS2.
 
Thank makes sense i did cpu design at uni a long time ago so im not sure what elements are still important.
So you think maybe 52 or 54 available for use? ( from 56)


i think that's likely. but there are also other options. something like a mirrored navi 14 could also make sense:

navi14mirroreds4jja.png


^ quick mockup

would be a bit bandwidth hampered in this form, though...
 
Last edited:

psorcerer

Banned
Besides that he's right.

Windows load time.

f2c1285ee8536e19441e85016e2fe889.jpg


Tomb raider:

f880efee742bfbdafe59e68f05feeed9.jpg


SSD M.2 - Samsung 970 EVO 500 GB (3400/2300 )
SSD SATA3 - Samsung 860 Evo-Series 500GB HDD ( 550/520 )
Western Digital Blue 7200rpm 1TB ( faster then PS4 )

Reducing 1 minute towards 1 second needs 60x the performance. Even if after that u get a SSD that's 3000 times faster. U will only gain 1 second of load speed advantage over that 60x. Which makes it completely pointless.

This makes 5400rpm > sata 3 ssd huge while sata 3 ssd > nvme not so much.

This is the issue that SSD's have and why i said earlier on even nvme ssd's are overkill for those consoles. Xbox series x kinda stole the thunder from sony by announcing a 2gb's ssd that no matter what sony does isn't going to move away from much noticeable as they are limited by pci-e 4.0 which is 5gb's.

Any exotic hardware design they are planning to push out to capitalize on that SSD performance in exotic ways will be ignored entirely if the rest of the market won't join it. Much like blu-ray in the PS3 area. So it will just be for there first party efforts which frankly don't even remotely need it at all. Just a waste of space.

Your PC SSD drive uses a block device emulation.
I.e. there is a 3core ARM cpu inside your Samsung drive that emulates HDD block device. While the flash drive itself is built as a k/v storage.
On top of that your OS uses a filesystem driver to supply a k/v storage known as "file system".
So. To finaly use a read() operation you pass 2 levels of abstraction and a lot of various checks.
More than that your drive and os and drivers use "exotic" API calls like TRIM just to make sure your SSD won't shit itself.
On console, where OS and drivers are fixed, you can use read() directly inside the flash controller, no abstractions. And nothing changes for the developer.
So, before you embarrass yourself by the next PC benchmark, like these DF idiots do each time, think.
 
Last edited:

onQ123

Member
So after 4 seconds ... and youve used all 100 gigabytes of game data... you going to then insert another blue ray disk ? Or just keep downloading 100 gig ever 4 seconds over your internet.

You are not half as smart as you think you are.

Not saying that they will but they could have standardized assets that is shared across many games & take up over 100GB
 

Ar¢tos

Member
The SSD's options then are (ignoring ssd size, I'm sure skus with bigger SSD's will come later) :
1 - 1TB SSD 5GB/s with custom controller and file system (or some other changes, but still just normal SSD chips)
2 - 1TB SSD 5GB/s + 128GB ReRam 25GB/s (expensive and 100GB or more games might not fit uncompressed in the ReRam)
3 - 1TB SSD 2-3GB/s + 128GB ReRam 25GB/s (with the fast ReRam the SSD doesn't need to be as fast and it will save costs)
4 - 1 TB SSD 5GB/s + 256GB ReRam 50GB/s (even more expensive but the ideal solution since big games would fit uncompressed in the ReRam and the ReRam is fast enough to keep feeding Ram. Bigger, more detailed game worlds with zero loadings and zero pop-in)
5 - 1TB SSD 2-3GB/s + 256GB ReRam 50GB/s

I'm more inclined to option 1, since it's the simplest and probably the cheapest.
 

psorcerer

Banned
I'm more inclined to option 1, since it's the simplest and probably the cheapest.

I would also bet on that.
You can extract quite lot of performance and they will need to build controller, drivers and OS support anyway. There's also option 0 - use the off the shelf ssd drive. Which everybody in the pcmr camp imply.
 

THE:MILKMAN

Member
He is . They r comparing it to Xbox series s .its a 6 tf gpu which For sure is stronger than Lockhart by a good margin

Lockhart doesn't even officially exist yet never mind have specs. Even if it will likely turn out to be true it is still a little cringe for NVIDIA to put that on a public graph.
 
Lockhart doesn't even officially exist yet never mind have specs. Even if it will likely turn out to be true it is still a little cringe for NVIDIA to put that on a public graph.
it for sure is cringe to try and trick the market but thats PR for you . if you watch the videos you know its a mobile variant of 2080 which is 6.5 TF.way below PS5 and XSX.
 
The SSD's options then are (ignoring ssd size, I'm sure skus with bigger SSD's will come later) :
1 - 1TB SSD 5GB/s with custom controller and file system (or some other changes, but still just normal SSD chips)
2 - 1TB SSD 5GB/s + 128GB ReRam 25GB/s (expensive and 100GB or more games might not fit uncompressed in the ReRam)
3 - 1TB SSD 2-3GB/s + 128GB ReRam 25GB/s (with the fast ReRam the SSD doesn't need to be as fast and it will save costs)
4 - 1 TB SSD 5GB/s + 256GB ReRam 50GB/s (even more expensive but the ideal solution since big games would fit uncompressed in the ReRam and the ReRam is fast enough to keep feeding Ram. Bigger, more detailed game worlds with zero loadings and zero pop-in)
5 - 1TB SSD 2-3GB/s + 256GB ReRam 50GB/s

I will add another option. First, if you remember the early "leaks" that says small SSD paired in HDD. So I think a relatively good HDD is still in the consideration.

But we're also hearing now of rumors of 1TB and also 2TB of SSD. So I would assume that is actually a cheap SATA SSD. The cold storage does not need to be top of the line in performance.

I'm more inclined to option 1, since it's the simplest and probably the cheapest.

Option number 1 will not support what the Control Developer has said. 1TB of SSD at 5GB/s will always, at all point, and at all scenerio, load the RAM (16-24GB?) in less about 3-6 seconds.

"If games would stay the same in terms of scope and visual quality it’d make loading times be almost unnoticeable and restarting a level could be almost instant [in PS5 games].

However, since more data can be now used there can also be cases where production
might be cheaper and faster when not optimising content, which will lead into having to load much more data, leading back into a situation where you have about the same loading times as today."
 
I would also bet on that.
You can extract quite lot of performance and they will need to build controller, drivers and OS support anyway. There's also option 0 - use the off the shelf ssd drive. Which everybody in the pcmr camp imply.

I'm sure you can extract quite a lot of performance on that set-up. But it does not support the statement made the Control developer. What he said sounds very much like a tiered storage. It sounds like there will be an ultra-fast cache paired with a slower cold storage. And that cache can only be an SLC Nand, Intel Optane, or Sony ReRAM. I'm betting on ReRAM.
 
Last edited:
I'm sure you can extract quite a lot of performance on that set-up. But it does not support the statement made the Control developer. What he said sounds very much like a tiered storage. It sounds like there will be an ultra-fast cache paired with a slower cold storage. And that cache can only be an SLC Nand, Intel Optane, or Sony ReRAM. I'm betting on ReRAM.
but reram is so expensive man. Reram will be there for PS6 i feel. lets see if you are right or not
 

Tsaki

Member
For me I'm really interested to see what secret sauce they both come up with like this, and what effect it has.
Yeah me too. Once the official specifications release it will interesting to see the "secret sauce" and customizations (however deep and intricate they are) that each company chose to incorporate, the reasons they chose to incorporate them and more importantly how much the videogame studios will leverage them. Having said that I don't expect big differences in the CPU and GPU department. In my opinion both next-gen consoles will have the same base architecture, meaning one will not have the full RDNA2 while the other will have RDNA "1.5" or one console has hardware VRS while the other does not. Some minor differences are expected, like different amount of cache memory on different levels of the APU and of course how the XBox and PS5 APIs communicate with the hardware. And this is one reason why I find the SSD part of next-gen more itriguing. The Sony patent can shed some light to the substantial differences the PS5 (and probably the XBox) SSDs will have when compared with regular market NVMe drives. There is a dedicated hardware decompressor, there is a small amount of SRAM in the place of DRAM, there is dedicated CPU that handles the file organization on the drive, and most importantly there is the software stack that is tailor-made for parallel I/O communication with the drive (and more customizations that I can't really remember right now).
As far as RAM is concerned there are some interesting questions to be asked. What kind of memory does the PS5 have and what is the amount and bandwidth it will provide? Are 16GB of GDDR6 enough to make the generation last until 2026/2027? How much leeway will the SSDs provide so that next-gen machines will not need a huge jump in terms of RAM when compared with previous gen transitions?
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
The SSD's options then are (ignoring ssd size, I'm sure skus with bigger SSD's will come later) :
1 - 1TB SSD 5GB/s with custom controller and file system (or some other changes, but still just normal SSD chips)
2 - 1TB SSD 5GB/s + 128GB ReRam 25GB/s (expensive and 100GB or more games might not fit uncompressed in the ReRam)
3 - 1TB SSD 2-3GB/s + 128GB ReRam 25GB/s (with the fast ReRam the SSD doesn't need to be as fast and it will save costs)
4 - 1 TB SSD 5GB/s + 256GB ReRam 50GB/s (even more expensive but the ideal solution since big games would fit uncompressed in the ReRam and the ReRam is fast enough to keep feeding Ram. Bigger, more detailed game worlds with zero loadings and zero pop-in)
5 - 1TB SSD 2-3GB/s + 256GB ReRam 50GB/s

I'm more inclined to option 1, since it's the simplest and probably the cheapest.

Option 2 looks so good though! But no way it's possible for $500.

I remember when I was the only one talking about ReRAM and people thought I was crazy now it seem like I'm the only one not talking about it lol

Because ReRAM is CRAZY talk! It's just not ready yet. Not for the consumer at least. I'd have to be smoking on that gas, before I think ReRAM is even possible in the PS5.
 

Gamernyc78

Banned
It's actually hard thinking of a product that can't be improved.
But when talking about ssd's there will be a point of diminishing returns.
I mean, you can't really improve apon "no loading times"
It's something I don't get about the ps5 having a better ssd discussion, what good is having a better ssd on paper when in use it's no better then the xsx's slower one.

Cerny will make you a believer. Sony isn't harping on about it for no reason, this goes beyond just loading times 😏
 
Last edited:

Norse

Member
But not vice versa🤔👌 bcus you know there's only one side of fanboys. You'd be surprised what sides would really implode given past history.

Boy you can't make up the shit you read on here 🤔😂
The first half of my post was for both sides. 2nd part was clearly bs but this thread is as well lol.
 
I remember when I was the only one talking about ReRAM and people thought I was crazy now it seem like I'm the only one not talking about it lol

You were one of the first ones I've read who talked about it though.

I think you also have a thread about downloading RAM, right? Haha. What if we can insert a "memory card" that can augment the ReRAM cache capacity. That will technically validate your theory. LOL
 

psorcerer

Banned
I'm sure you can extract quite a lot of performance on that set-up. But it does not support the statement made the Control developer. What he said sounds very much like a tiered storage. It sounds like there will be an ultra-fast cache paired with a slower cold storage. And that cache can only be an SLC Nand, Intel Optane, or Sony ReRAM. I'm betting on ReRAM.

Can you quote exactly what was said?
Maybe there's a different interpretation.
 

vpance

Member
The SSD's options then are (ignoring ssd size, I'm sure skus with bigger SSD's will come later) :
1 - 1TB SSD 5GB/s with custom controller and file system (or some other changes, but still just normal SSD chips)
2 - 1TB SSD 5GB/s + 128GB ReRam 25GB/s (expensive and 100GB or more games might not fit uncompressed in the ReRam)
3 - 1TB SSD 2-3GB/s + 128GB ReRam 25GB/s (with the fast ReRam the SSD doesn't need to be as fast and it will save costs)
4 - 1 TB SSD 5GB/s + 256GB ReRam 50GB/s (even more expensive but the ideal solution since big games would fit uncompressed in the ReRam and the ReRam is fast enough to keep feeding Ram. Bigger, more detailed game worlds with zero loadings and zero pop-in)
5 - 1TB SSD 2-3GB/s + 256GB ReRam 50GB/s

I'm more inclined to option 1, since it's the simplest and probably the cheapest.

64GB ReRAM will be more than enough, no need to get greedy. They don't need to fit the entire game in there, just the really important assets.
 
Can you quote exactly what was said?
Maybe there's a different interpretation.


If games would stay the same in terms of scope and visual quality it’d make loading times be almost unnoticeable and restarting a level could be almost instant [in PS5 games].

However, since more data can be now used there can also be cases where production
might be cheaper and faster when not optimising content, which will lead into having to load much more data, leading back into a situation where you have about the same loading times as today.


So it's either the SSD is not truly faster than anything on the PC right now. (which is possible)
But even at 4gb/s it won't result to a loading times similar as today.

Or this is an evidence of tiered storage.
 
Last edited:

psorcerer

Banned




So it's either the SSD is not truly faster than anything on the PC right now. (which is possible)
But even at 4gb/s it won't result to a loading times similar as today.

Or this is an evidence of tiered storage.

I wouldn't read too much into this. It's just a generic argument that if you increase the amount of data with the bandwidth the time will stay the same. I.e. thank you captain obvious, let's move along.
 
Can you quote exactly what was said?
Maybe there's a different interpretation.

There this too that could mean either: a. there's a dedicted asic decompression chip, b. decompressed assets streaming from a fast cache.

"When it comes to the PS5, faster hardware is always appreciated and will make life easier in the short term, but it's the new SSD that really stands out; essentially streaming will become something that we don't really have to worry so much about and it will free up some extra CPU bandwidth in the process," said Remedy's lead programmer Sean Donnelly while speaking to the Official PlayStation Magazine.
 
Last edited:

Ar¢tos

Member
64GB ReRAM will be more than enough, no need to get greedy. They don't need to fit the entire game in there, just the really important assets.
Won't 64GB be half speed of 128GB since it means 4 chips instead of 8?
That would be 12.5GB, just barely double of the SSD, and not enough to justify using the ReRam at all.
 
I wouldn't read too much into this. It's just a generic argument that if you increase the amount of data with the bandwidth the time will stay the same. I.e. thank you captain obvious, let's move along.

It's either that or he was onto something. I wouldn't dismiss it like that easily either.
 

Norse

Member
Too bad they can't figure out a way to load directly off the 100+ GB blue ray disc fast enough so we don't have to install games. And fast ssd or m2 drives won't speed up the current long installation of games off of media.
 

vpance

Member
Won't 64GB be half speed of 128GB since it means 4 chips instead of 8?
That would be 12.5GB, just barely double of the SSD, and not enough to justify using the ReRam at all.

Unless there's some reason they can't or don't want to make 8GB chips, they could do that to keep it at 25GB/s.

The total streaming speed may actually be SSD plus the ReRAM.
 

THE:MILKMAN

Member
Too bad they can't figure out a way to load directly off the 100+ GB blue ray disc fast enough so we don't have to install games. And fast ssd or m2 drives won't speed up the current long installation of games off of media.

This is actual a good point. RDR2 for example can take well over an hour to install off the discs. I can't believe they will push the 'no load times' line if these install times remain. Is there anything that can be done about it?
 

DeepEnigma

Gold Member
This is actual a good point. RDR2 for example can take well over an hour to install off the discs. I can't believe they will push the 'no load times' line if these install times remain. Is there anything that can be done about it?

Games are larger and larger because of data duplication on mechanical drives for bigger assets. You won't need to duplicate data for an SSD, so files sizes won't need to be as bloated on the consoles. There is also the piecemeal install options (single player/multiplayer/etc). Cerny talked about both in the Wired article.
 
Last edited:
Status
Not open for further replies.
Top Bottom