Next Generation Hardware Speculation with a Technical Spin [2018]

Status
Not open for further replies.
Multiplatform + cross-gen game devs, come 2020/2021, will have to do that for 2 known consoles from 2013, 2 different but known consoles from 2016/17 and two different and unknown consoles from 2020.
Adding yet another console would make an incredibly difficult task into an almost impossible one.
I dunno. I think developers can be significantly more flexible than that.
Examples: Destiny 1
X360, XBO, PS3 + PS4
new IP, from scratch, all released at the same time.

That's a significantly harder lift, than developing a new title for Scarlet and Scarlet+, which should essentially be the same architecture with the only difference being the available resources. It would be easier than coding 2 separate pathways for XBO and PS4, since they have underlining memory substructure differences.

We saw something similar with BF4 which launched on all 4 platforms as well. Overall I don't think having multiple SKUs is a big lift especially if they are of the same architecture and behave the same. APIs are there to do the heavy lifting. Optimizing the game to run smoothly should be trivial in comparison to having to optimize architecture differences.
 
No idea how practical that would actually be, but the concept seems sound.
The concept is sound, but there must be a reason why it isn't widely adopted.

I think it probably comes down to an artistic choice instead of a technical one. At some point someone chooses to have a more consistent image quality at all times instead of allowing something akin to Switch's Wolfenstein II that dips down to 360p resolution (obviously an extreme case here).


Examples: Destiny 1
X360, XBO, PS3 + PS4
That example has 4 different consoles. Your suggestion of a cross-platform launch would mean 7 to 8 different consoles:
- PS4
- PS4 Pro
- PS5
- PS5 Pro
- XBone
- XBoneX
- XBtwo
- XBtwoX
 
That example has 4 different consoles. Your suggestion of a cross-platform launch would mean 7 to 8 different consoles:
But only two different architectures versus four for Destiny 1, in theory. Destiny needed to work on XB360 tiled with limited eDRAM, PS3 with Cell, XB1 with limited ESRAM and PS4 with standard PC UMA, with three different CPUs and three different GPU architectures. Future platforms would consist of, presumably, a standard PC style UMA and XB1's limited ESRAM; same CPU architecture and same GPU architecture and same fundamental rendering models except for tiling/paging on XB1. So a future engine that can scale is one that can already scale on PC from integrated graphics of a laptop up to mega CPUs and monster GPUs. It won't have to work differently on different platforms. (All assuming future architectures avoid the exotic.)
 
On the subject of tiled resources for memory saving, and the possibility of only seeing 16gb in the Ps4, I think such a setup would justify bringing back eDRAM. It's not like Sony didn't consider it for Ps4, so it's possible. Perhaps 128mb? Where are eDRAM speeds at nowadays? I know that at least intel has made use of that size eDRAM.

I still think we'll see more than 16GB and likely with a 384 bit bus at that, but it's a thought.
 
I regularly see the 2TB Micron for $250. By 2020, we should be under 10 cents/GB.

That drive is typically priced around $300, and its a 15% off coupon for Rakuten that knocks the price down to $255 for brief sale periods. The drive is a 2016 model that uses early 3D NAND and warranty should be 3 years from the place of purchase.

So ... I still think that realistically theres still a long ways to go for SSD to hit low enough prices to be used as mass storage for consoles. However, as I said from the beginning, it could be used as part of SSHD hybrid drive.

Though I think we would see better performance with the console explicitly having a small area for SSD acceleration (maybe 64 GB) and letting the OS or developers control what gets cached. This is where MS's deep learning and telemetry could speed up game loading of already installed games just like they're speeding up game download to launch with FastStart.

For full reference, here is the Amazon listing for the drive: https://www.amazon.com/Micron-MTFDD...scsubtag=e88bcc1b-0e4e-4ec8-abcd-a45bd74c8459
 
On the subject of tiled resources for memory saving, and the possibility of only seeing 16gb in the Ps4, I think such a setup would justify bringing back eDRAM. It's not like Sony didn't consider it for Ps4, so it's possible. Perhaps 128mb? Where are eDRAM speeds at nowadays? I know that at least intel has made use of that size eDRAM.

I still think we'll see more than 16GB and likely with a 384 bit bus at that, but it's a thought.
As I understand it, eDRAM limits your fab choice dramatically. That's why XB1 went ESRAM instead.
 
I doubt the esram idea will come back, it was interesting on paper and failed in the real world. The most telling is that MS ditched it in the middle of the generation. Still we don't really know if it was a failure of the entire concept or just a failure of implementation. Or even just a bad prediction of the memory market.

Next gen will have a very high silicon cost at 7nm or beyond, so a large area for esram would be even more difficult to justify. Any fancy internal ram will later become another emulation problem for mid-gen if they change course (again).
 
On the idea of a hybrid drive...
Isn't most of the data loaded from the disk read in a sequential manner? (Large textures)
If so then the advantage of a hybrid drive would be pretty minimal. The data would mostly be read from the disk and then into the SSD cache part and since (I would guess) most of the game data isn't repeatedly accessed the data in the SSD cache wouldn't be reused much so your not taking advantage of the low latency part or high throughput that a dedicated SSD brings.
Also I don't believe the SSD part of the hybrid drive is a separately addressable entity so there isn't any way to put particular data in the SSD part of the drive for later use.
Maybe you could take time to load data you think you will need by reading them to null and pre-warming the cache part for later use, but that seems like a time waster.

The general consensus in the 3d xpoint thread was that even tho 3d xpoint was super low latency and about 6GB/s throughput it wouldn't be much use as a cache for vram.
 
I doubt the esram idea will come back, it was interesting on paper and failed in the real world. The most telling is that MS ditched it in the middle of the generation. Still we don't really know if it was a failure of the entire concept or just a failure of implementation. Or even just a bad prediction of the memory market.

Next gen will have a very high silicon cost at 7nm or beyond, so a large area for esram would be even more difficult to justify. Any fancy internal ram will later become another emulation problem for mid-gen if they change course (again).
They paired eSRAM with ddr3, though. It says nothing about how it would've benefited gddr5 memory. As it was, esram was a patch for low bandwidth. In contrast, ps2 had high main memory bandwidth (for 2000) with eDRAM to supplement, not as a necessary patch. Esram fulfilled MS's purpose. If anything failed it was MS's prediction in needing ddr3 to be able to have 8gb.

If eDRAM is out because of process nodes, then eSRAM is the next best thing if paired with gddr6. But again I think it should only be used if we're limited to 16gb or less than a 384 bit bus.

Edit : Can't link, but according to wikichips 7nm process page, they can make a 256mb chip on that process. It's eSRAM, can't seem to find specs.
 
Last edited:
The drive is a 2016 model that uses early 3D NAND and warranty should be 3 years from the place of purchase.

So I need to post what I found from the HardOCP thread on this drive, namely you have no warranty at all from Micron or Platinum Micro, so it comes down to what Rakuten or Amazon would do when sold from the 3rd Party.

"Buyers be aware.
I have contacted Micron and Amazon (since I have bought one from Platinum Micro on Amazon), Micron claimed there will be NO WARRANTY from manufacturer since Platinum Micro is not an authorized distributor.
When I turned to Platinum Micro, they simply forward me to Amazon for warranty answer.
So Neither Micron Nor Platinum Micro would provide warranty for this product.
If Amazon refused to provide warranty, I would return mine, as $320 is not a small amount."
 
On the idea of a hybrid drive...
Isn't most of the data loaded from the disk read in a sequential manner? (Large textures)
If so then the advantage of a hybrid drive would be pretty minimal. The data would mostly be read from the disk and then into the SSD cache part and since (I would guess) most of the game data isn't repeatedly accessed the data in the SSD cache wouldn't be reused much so your not taking advantage of the low latency part or high throughput that a dedicated SSD brings.
Also I don't believe the SSD part of the hybrid drive is a separately addressable entity so there isn't any way to put particular data in the SSD part of the drive for later use.
Maybe you could take time to load data you think you will need by reading them to null and pre-warming the cache part for later use, but that seems like a time waster.
Yes, which is why an SSHD makes no sense to me. It'd be more advantageous to have a separate flash drive that can be accessed either through OS calls to work as a cache for the HDD or directly as a flash drive to be used as the devs see best.

SSHD's only exist as an awkward cost compromise for PCs where you can't afford a dedicated SSD and HDD but drive manufacturers wanted a faster option to differentiate.
 
Yet SSHD drives do show performance increases through faster loading times in existing consoles, both on Xbox and Playstations. Every little bit helps, right? And if its under a $1 per unit, then it just might be included. I cant say what the actual difference in costs will be in bulk for the likes of MS/Sony, but there is an absolute floor price on normal HDDs, so the cost may be completely negligible.

However, as mentioned multiple times, having a cache SSD under OS or developer control would provide more potential. But at what costs?
 
On the subject of tiled resources for memory saving, and the possibility of only seeing 16gb in the Ps4, I think such a setup would justify bringing back eDRAM.
No way.
eDRAM or eSRAM in larger capacities are an area hog that is immensely best spent on more execution units and a wider memory controller, as the Xbone -> Scorpio transition showed.
Another example are the Intel Iris Plus with the eDRAM that are easily beaten by the integrated Vega 8/11 on Ryzen APUs with a fraction of the eDRAM's bandwidth using regular DDR4, compared at similar power levels.
 
That was my thought as well, hybrid drives are only really good for thing that have random access and are reading multiple times in a data set that is < cache size.
What scenarios would a (say) 16/32 GB SSD really be helpful for? (Thinking this size from a cost perspective)
I can't really think of anything in a game where a scratch area like that is helpful.... if you get say a 500us response time and say 1GB/s throughput what could this actually help in a game?

To Brit -
Do you have any references that show a console that normally has a HDD which is then replaced by a hybrid drive so I can see the replacement strategy and performance improvements?
I would like to see the performance differences and testing methodology.

My reference is from windows where replacing a HDD in a laptop with an similar hybrid drive certainly didn't increase windows boot time by hardly anything. (And there is some random read there)
 
Last edited:
I think game developers want some guarantees their data is in fast storage, so they can stream assets on demand instead of preloading gobs of data. The HDD cache, caching optical drive data, on the 360 worked wonders when it worked and completely broke when it didn't. We want the first, and not the latter.

Say we get a console with around 250GB SSD. Half of that could be reserved to cache games from slow storage. Console vendors could make a "precache" syscall, where the game is notified when all specified assets are in the SSD cache. With a 120GB SSD cache you could probably fit data for 2-3 games in there. Modern HDDs can stream sequential data at a fair clip (>100MB/s), so precaching shouldn't be too intrusive.

Cheers
 
I got a quick glance at a BOM for a small form factor 4K Android STB the box is smaller than a apple tv in size.
They said that currently the RAM prices was hurting them a lot.
They have 2GB DDR4 (no idea of speed etc) and the current price they had to pay was 16 USD for the RAM.
The volume here is not even close to a console manufacturer numbers and its long contracts, but still they do order in the tens of thousands.
 
I did find a video comparing SSD vs Hybrid vs HDD (hybrid drive was a larger drive with greater physical density) and there was a ~4% difference in boot time between hybrid and hdd. Also there was only a ~8% difference between SSD and HDD.

Would still like to see some game load comparisons. But if they were similar then the hybrid drive cost would have to be very small indeed to make this worth it.

As shifty mentioned - having consistently fast performance is what is desired, and a separately addressable SSD space would provide that vs a hybrid drive (unless there is some pre-cache function in the drive itself which I dont *think* is the case).
 
Last edited:
That was my thought as well, hybrid drives are only really good for thing that have random access and are reading multiple times in a data set that is < cache size.
What scenarios would a (say) 16/32 GB SSD really be helpful for? (Thinking this size from a cost perspective)
I can't really think of anything in a game where a scratch area like that is helpful.... if you get say a 500us response time and say 1GB/s throughput what could this actually help in a game?
Tiled resources are an ideal fit. You can also use it more effectively to procedural content creation, which can be created, saved, and streamed very efficiently, rather than constantly procedural creating in realtime or having to save and load as monolithic textures with super slow seek-times between each texture.

And that's off the top of my head. The moment the tech is there, it's going to be used. same as putting in 32 GBs RAM. If the RAM is there, devs will find a way to use it. If the storage is there, devs will find a way to use it. And if the fast storage is there, devs will find a way to use it. What flash provides is a good compromise between RAM and HDD, offering latency which cannot be address any other way beyond more RAM which is both cost prohibitive and volatile.
 
I can't really think of anything in a game where a scratch area like that is helpful.... if you get say a 500us response time and say 1GB/s throughput what could this actually help in a game?

Could it help with keeping more NPCs in memory meaning the 'active play area' is larger? Could it be used as a pre-emptive scratch-pad so larger areas are possible? Could it be used to hold onto data for longer soon, I'm thinking sometimes in a game you might get sidetracked and when you get back 'on track' the mission has reset or gone.

Do you have any references that show a console that normally has a HDD which is then replaced by a hybrid drive so I can see the replacement strategy and performance improvements?

My reference is from windows where replacing a HDD in a laptop with an similar hybrid drive certainly didn't increase windows boot time by hardly anything. (And there is some random read there)

I posted my PS4 stock/SSHD/SSD comparison times somewhere here but I confess it's not perfect (but does show clear SSHD advantages and better bang for buck)
 
Could it help with keeping more NPCs in memory meaning the 'active play area' is larger? Could it be used as a pre-emptive scratch-pad so larger areas are possible? Could it be used to hold onto data for longer soon, I'm thinking sometimes in a game you might get sidetracked and when you get back 'on track' the mission has reset or gone.



I posted my PS4 stock/SSHD/SSD comparison times somewhere here but I confess it's not perfect (but does show clear SSHD advantages and better bang for buck)

Do you remember what % of differences were really seen?
Was the hybrid replacement drive the same capacity/generation as the hybrid drive?
Did you document the testing methodology?

Im honestly curious to see what the improvement was.
 
Status
Not open for further replies.
Back
Top