Next Generation Hardware Speculation with a Technical Spin [pre E3 2019]

Status
Not open for further replies.
A small SS "cache" + HDD does not mean you can't have insta boot up, at least, most of the time.
With a big enough SS partition, say 128-512GB the system can fit 3-5 full games in there (most recently played), the OS and recent apps, and still leave a big portion as runtime scratch-pad memory. With enough storage (and efficient use of it) they can even keep the game state (for insta boot-back) for more than just the last played game, and do that without any power when standby. You can un-plug your ps5 from the wall after turn-off and you still can restart from the exact same save-state as you left off once you turn it back on (impossible on ps4 because it relies on mantaining data on memory by keeping the memory active at a low-power mode)
And there are multiple levels of sophistication at which that could be done too.
For example, as new games are booted up, and older fully cached ones don't fit any more, they could be de-promoted into just partial cache, the main things needed for a quick boot up, giving the game some time to load the rest from HDD to SS as you are playing. With good tagging system, they can make it smart enough so that it prioritises the portions of the game that player is actually in at the moment. In a linear game, that would mean prioritizing current level, and current checkpoint, in an open world, the current area and it's immediate surroundings and only the necessary data for the world state the player finds himself in at the moment. In a game with many modes, that could mean prioritizing the mode that player has been playing at (ex: he's only playing MP now, we can clear SP campaign of the cache and keep fast boot up into the MP or He is only playing Zombie Horde on CoD, if we get to need more space, clear the data of the other MP modes)
If this gets done right, you can very well get a system that effectively has near-insta boot into the software you want 90% of the time, and it still only takes a dozen or so seconds of wait when it doesn't. And that while still having 1TB (or more) of total software storage and not burning $200+ on storage alone.
 
Yep, it's clear now insta-loading will be an exclusive feature of PS5 like Play-go was (and still is) for PS4.

Xbox One has had similar functionality as PlayGo since launch and from my experience they are both terrible at it. I always wait for my games to fully download, i don't want to play half a tutorial level and then have to wait for the rest of the game to download before i can continue.

Last year at E3 Xbox unveiled FastStart(which i think is supposed to be a replacement) which supposedly uses machine learning make you be able to play twice as fast after you start downloading without devs having to do anything.

I wonder if the same method or a variation of it be adopted to improve load times?
 
Again, saying that the solution is anything but an SSD drive seems intentionally obtuse, or devil's advocate to the point of being obtuse.
This. Especially considering that NAND is expected to reach $0.08/GB by the end of 2019, effectively putting it in the same ballpark as mechanical drives.

Lots of mental gymnastics here just because Cerny didn't specifically say "the PS5 has a SSD", despite him mentioning (1) the SSD in his laptop, (2) developers asking him to put an SSD in the PS4 and (3) the loading time difference from putting a SSD into the PS4.

Why wouldn't Sony (and Microsoft) use a SSD for their next consoles? During their lifetime, NAND will definitely become cheaper per GB than magnetic discs.
At this point, going with anything slower than QLC is little more than assuming a huge compromise.


As I said, the novelty here isn't even using fast storage. We've had that in the PC for years. The novelty is related to how fast they can decompress and decrypt the data that's coming from the mass storage, with the practical result being small loading times. That's something the PC crowd doesn't have.
 
I an one of those who believe Cerny didn’t state PS5 will have an SSD, simply because the SSD will not be the main storage device, but a huge scratchpad for faster loadings,
As such, the PS5 will remain having a HDD for mass storage, and cheaper expansion over the USB ports.
The SSD part will just act as a buffer, something that will keep the used game files present for faster access.
This will not speed a first time game loading, but it will speed all other access if required files continue to be placed on the SSD after the game loading.
This is a tradeoff. Not always as fast as an SSD solution for storage, but cheaper not only for console cost, but also for future expansions.
This would also remove any eventual USB speed restrictions on external HDD’s.
 
And that while still having 1TB (or more) of total software storage and not burning $200+ on storage alone.

Full stop. How much do you think a TB of QLC NAND Flash will cost in late 2020?

Right now, you can buy TLC on the spot for ~$95/TB. Large-scale purchases agreed well in advance pay less than spot. QLC is cheaper than TLC. The price of QLC is expected to fall by more than 50% in the next 12 months.

I would be very surprised if Sony had to pay more than $30 per TB of QLC.

With a big enough SS partition, say 128-512GB the system can fit 3-5 full games in there (most recently played), the OS and recent apps, and still leave a big portion as runtime scratch-pad memory.

My pet peeve here is that everyone is going "flash is expensive", while proposing systems that would likely be more expensive than just shipping a terabyte of QLC in 2020. Which you are doing here. Hard drives have a hard cost floor at ~$35. That alone already buys Sony more than a TB of flash in late 2020. If you feel that 1TB is too little for a console launched in 2020, shipping 2TB doesn't exactly break the bank either.

Now, if you add a fast scratchpad on top of that, you need to consider write endurance (because scratchpads see a lot more writes than the mass storage, and cheapest possible QLC lasts less than 500 writes), and speed, because individual cheap flash chips aren't very fast. To make a fast system, there needs to be many in parallel, so you have to use lower-capacity chips to make that small pool while still having enough to get those "better than any PC" speeds. And those smaller chips are more expensive per GB.

I still feel very strongly that for what they are proposing to do, a single dumb PCIe 4.0 NVMe controller attached to 1-2TB of cheapest possible flash is not just faster and all around better than any tiered storage solution, it will also be cheaper. Which is why that is what they will do.
 
This. Especially considering that NAND is expected to reach $0.08/GB by the end of 2019, effectively putting it in the same ballpark as mechanical drives.

[...]

But doen't this have to do with oversupply of NAND instead of better manufacturing techniques? At least that's what I've read over the last months.

Could this oversupply help MS and Sony in getting better deals? Something along the lines of:

"hey samsung/miron/whoever we know the NAND prices are currently crashing due to oversupply. listen, we want to include large quantities of NAND in our next generation projects and would be willing to do a frontloaded contract with you. when we start buying at the end of 2019 we pay a bit more than what you would get by selling to the market, but in return we get much better prices over the lifetime of the contract when you sell at better margins to the spot market again".
 
My pet peeve here is that everyone is going "flash is expensive", while proposing systems that would likely be more expensive than just shipping a terabyte of QLC in 2020. Which you are doing here. Hard drives have a hard cost floor at ~$35. That alone already buys Sony more than a TB of flash in late 2020. If you feel that 1TB is too little for a console launched in 2020, shipping 2TB doesn't exactly break the bank either.

Exactly how I feel.

How big must a scratch pad be? 120GB minimum I think. So how much does a 2TB mechanical HDD and 120GB flash cost combined? Don't think it will be much cheaper than a TB flash especially not in the long term.
 
So then what is their message for all those currently using their own external drives on PS4s and Xbox Ones? Are they now told they cant use them on nextgen unless they are SSDs?

Even an SSD would need to cache to internal storage if the game intended to benefit from Sony's "faster than anything on PC" storage solution. USB3 ain't gonna cut it.

If you want to use external storage, and benefit from the kind of performance Sony demonstrated, then that internal storage is going to be used as tiered storage anyway and having it's write lifetime worn down. You can't get away from it.

You might not even be able to run natively from external without caching to internal - even if you wanted to - it'd play havoc with QA and really fuck with consistency and the user experience.
 
Last edited:
So then what is their message for all those currently using their own external drives on PS4s and Xbox Ones? Are they now told they cant use them on nextgen unless they are SSDs?

For installing games, yes. In fact, they cannot use any of them on the next gen, because PS4 cannot accept NVMe. By the time PS5 launches, the most common mass storage sold will likely be NVMe, and that is what PS5 will support.

It probably wouldn't cost them too much to have 2 slots, so you can expand storage without having to replace the existing one.

And it might be worth it to still allow attaching USB mass storage for audio or video.
 
Now, if you add a fast scratchpad on top of that, you need to consider write endurance (because scratchpads see a lot more writes than the mass storage, and cheapest possible QLC lasts less than 500 writes), and speed, because individual cheap flash chips aren't very fast. To make a fast system, there needs to be many in parallel, so you have to use lower-capacity chips to make that small pool while still having enough to get those "better than any PC" speeds. And those smaller chips are more expensive per GB.

It would be a shame if there were no high endurance flash partition next gen. I think consoles will have to be able to cache games from external HDD or SDD anyway, particularly early on, so that QLC is going to get worked and used as tiered storage regardless. Additionally, part of the appeal of high bandwidth / low latency storage is that the game may be able to use it in ways to expand upon the gameplay we currently have. Perhaps that's a pipe dream, but it'd be nice if there was the option.

Finally, game saves and suspends can add up. A game with auto checkpoint saves into the megabytes every few minutes might not sound like a lot, but when you add it up, along with a nightly suspend of several GB, and put them on a drive that's almost certainly going to be maxed out with installs so you have to start using wear levelling that brings on even more writes .... I can't helping thinking that 500 writes on new, cheap QLC may be insufficient. At least throw in 64GB of SLC or MLC for that!

Whether it's ultimately flash or mechanical for mass storage, I would really like to see smaller portion of high speed, high endurance flash. QLC only, especially in a none user-replaceable drive (as with MS on the X1) would not be as exciting or fill me with confidence.

I still feel very strongly that for what they are proposing to do, a single dumb PCIe 4.0 NVMe controller attached to 1-2TB of cheapest possible flash is not just faster and all around better than any tiered storage solution, it will also be cheaper. Which is why that is what they will do.

I can see your case for it, but I definitely don't agree that it would be better all round. Endurance for running games from external storage would be much worse, and you lose the realistic possibility of a developer controlled portion that might be used to expand upon gameplay ideas.

So while I can definitely see what you mean about QLC replacing mechanical HDDs, I don't think it's an ideal solution on its own.
 
Seen it mentioned a couple times that SSD's will also get cheaper over time.
When that happens swap out production to use 2TB of SSD instead of HDD. I'm not sure why that's not considered an option.

If you only had SS then sure it will get used by 1P same way Stadia unique features will also be used by 1P.

Depending how tired solution is done and what size, could do things you can't with HDD also.

3P would use pc SSD as baseline.

For the record I expect Lockhart to use same storage strategy as Anaconda. Maybe smaller is all.
 
Exactly how I feel.

How big must a scratch pad be? 120GB minimum I think. So how much does a 2TB mechanical HDD and 120GB flash cost combined? Don't think it will be much cheaper than a TB flash especially not in the long term.

As a consumer a 2 TB HDD + 128 GB NVMe SSD (M.2 2280) combo can be had for around ~80 Euro. But that includes VAT + multiple retailer margins. For almost the same price (~92 Euro) you can already get a 2 TB HDD + 256 GB NVMe SSD (which has 3.5 GB/s peak read compared to 1.6 GB/s from the 128 GB SSD. 128 GB SSDs with ~3 GB/s go for more than 256 GB SSDs).

For comparison the same setup with half the HDD size would cost ~65 and ~77 Euro in comparison. 1 TB NVMe SSDs with 3+ GB/s read are ~140 Euro at the moment.
 
That's something the PC crowd doesn't have.
The irony of this statement isn't lost on me when Wired described a shootout of
(The devkit, an early “low-speed” version, is concealed in a big silver tower, with no visible componentry.)
vs
PS4Pro.
Or if that' isn't visual enough. PS5 kit as described is this except silver.
Screen_Shot_2012-08-14_at_4.41.38_PM.png


and it's being compared to this
ps4-pro-neo-devkit-testkit-ofw-3-70-pictures-and-screenshots-5-jpg.1685


To re-ground ourselves from this article, If you followed Durango; The kit is powered by both an Intel CPU and Nvidia GPU and packs more than 8GB of memory, although claims that it will have an 8-core CPU have been described as "difficult to believe". And we know how this story turned out. What's in that silver tower could be a full blown high end PC grade Zen 2 processor with a full on navi GPU.

When PS4 alpha kits came out, the place I was working at received nothing more than an envelope containing a GPU. It wasn't until much later did the real devkit arrive. Do we all remember 'downgrade' city from E3 to release for Watch Dogs? I mean, something told them that the console would look like X and not Y.

So I have a healthy dose of caution when we use the performance of an alpha kit compared to an official dev kit to guess the underlying hardware of consumer release. I much prefer the arguments that talk about cost of NAND flash as being the central reason for seeing SSD storage, because economically that makes sense. And those are the reasons I'm cautiously optimistic we'll have SSD.

Sony has not provided anything but a couple brands to spawn this level of positive emotions.
I mean can you tell me which console I'm talking about here:
  • AMD Polaris GPU
  • AMD 8 Core Custom CPU Jaguar
  • 1 TB HDD
  • DX12_0 hardware support

Is this console
a) Xbox 1S / Slim
b) 4Pro
c) Scorpio

the answer is D, it's all of the above. The devil is in the details where 1 does 720p and the other does native 4K for RDR2. Sony gave you exactly what you wanted to hear, but none of the details that actually matter. They gave you the performance of a devkit that is actually a PC and compared it to a completed official devkit.

This isn't mental gymnastics we're performing here, it's called reservation. I think we're allowed to have that for now until the signal is stronger.
 
I generally have issues when I read numbers like 19 times faster.
We have major revolutions in gaming just being to be 2x to 3x faster. That's just ray tracing.
Being 2x to 3x faster is already, massive. 19 seconds becomes 6. Or in the case of his PS4 example, 19 should become 12 by installing SSD.
When 19 seconds becomes 0.8 seconds and I see an argument that its performance is strictly just storage (not the massive bump in going from Jaguar to the unreleased Zen 2, not the change in the bus speeds 8 years later; and I can't don't have access to it or can't purchase it because it's faster than what's available on PC ssds, red flags are up for me on the argument.

The idea that they could come up with a I/O raw storage reading better than industry players who do data centers, is intriguing at the very least, how many records does google have to smash through to get your results in a fraction of a second; I mean they run really slow hard drives, they can't all be SSD.

We need that type of tech everywhere in the world, if this is actually a Sony designed solution, being told it's limited to a PS5 is non-sensical to me, Sony would profit more from selling it. There's certainly a lot more money there to be made.

15 seconds became 0.8 not 19.
Essentially you’re saying Cerny is purposefully being vague or misleading to make something sound better than it actually is. An example would be ‘my car can do 100mph when to achieve that I need wind behind me and driving down a steep hill on a straight road with no chance of traffic/obstacles.

Please can you show any evidence where Cerny has done something similar? At least if there is signs he’s done this type of thing before then it might sound more than just doubt for the sake of doubt.

yeah... if true Sony has again guessed a winning feature... maybe a bit les of Teraflops but warp speed loading in certain games are soooo good

Educated guessing, but also I agree...something as simple as ultra fast loading could be a game changer and major feature advantage vs (say) 20% more performance.
 
Educated guessing, but also I agree...something as simple as ultra fast loading could be a game changer and major feature advantage vs (say) 20% more performance.

My biggest reason for opting for the PC versions of some games games is slow load times on console. I care less about resolution and 60fps than I do about not staring at one of Bethesda's spinning assets for 30 seconds because this one tiny room has it's only load zone which takes 2 seconds to go in and 30 seconds when I go back out.

That was probably cut from the Wired interview.

FakeWired said:
Wired: Why did you invest so much effort reducing loadtimes?. Mark Cerny: Bethesda Game Works and Bioware RPGs. QFT.
 
15 seconds became 0.8 not 19.
Essentially you’re saying Cerny is purposefully being vague or misleading to make something sound better than it actually is. An example would be ‘my car can do 100mph when to achieve that I need wind behind me and driving down a steep hill on a straight road with no chance of traffic/obstacles.

Please can you show any evidence where Cerny has done something similar? At least if there is signs he’s done this type of thing before then it might sound more than just doubt for the sake of doubt.



Educated guessing, but also I agree...something as simple as ultra fast loading could be a game changer and major feature advantage vs (say) 20% more performance.
It's not Cerny that's misleading. We're the ones trying to reverse engineer an editorial non technical piece to figure out what PS5 is. It's our own arguments that are being presented that I'm casting doubt into.

I'm just conservative, because the release spec is unlikely to match some of these 'guess specs' within a reasonable price point. It's a lesson well learned with this generation and with the mid-gen refresh.
 
For comparison the same setup with half the HDD size would cost ~65 and ~77 Euro in comparison. 1 TB NVMe SSDs with 3+ GB/s read are ~140 Euro at the moment.

Retail prices don't give us the full story, way to many parties that have to make a profit.
 
I'm just happy this concept is finally getting the attention it deserves. So many people failed to see the utility in having some form of storage to memory acceleration. Fortunately, the system architects of the next-gen consoles weren't among them.
 
Status
Not open for further replies.
Back
Top