Pup will always be there. New hardware will just enable to put more [and more detailed] stuff on screen, but LOD distances will not change. Raytraced shadows will at least remove distance shadow fuckups that we are so used to see in every game.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
PS5 - a patent dive into what might be the tech behind Sony's SSD customisations (technical!)
- Thread starter gofreak
- Start date
I guess it depends if the engine is coded to trigger the LOD by distance or proportionally to resource use. If it's the latter then it can be helped. That's how MS does backcompat on the X isn't it.
This is they key bit that's important. More than the numbers or anything else, it's that it can transform the way games are designed by making an experience that was otherwise infeasible (outside of cutscenes, very controlled settings) a possibility.
At the behest of many users, I came here to learn. Your technical specs did not disappoint! Thanks for posting this.
OK, sure, there are uses which benefit from small sectors. But the benefits seem to me like they would extend to most media consumption, at the very least. Even songs tend to weigh megabytes.
At any rate, if an SSD manufacturer makes a PC SSD with such characteristics (hybrid or big sectors, SRAM internals, driver with priority system), can Sony sue them? How encompassing this patent is?
Question - can any of this let the SSD function as high bandwith memory? For example could ps5 be 16 GB of Gddr6 and 8 GB partitioned to the SSD to use as a scratch pad for memory?
Ram reads/writes and latencies are still orders of magnitude better than this. Not to mention that for a CPU, latency is a real killer of performance.
This also depends on the Nand used. MLC Nand is a lot cheaper than SLC, but has higher access latencies. There are ways around this, but a lot of this depends on the raw Nand that they can source cheaply.
Last edited:
This is my big question. This patent could have effective uses by walling off the SSD to serve as RAM. It would be lit.
Yes, but the difference is nowhere near the difference between RAM and mechanical drive HDD. The point was that the SSD will be able to quickly fill out huge parts of the total RAM size with new data whereas that would take a long time to do with an HDD's slow read speed.
OP
OP
That's a bit complicated - but to say, although this application discusses lots of things in a lot of detail, it's only 'claiming' some of the stuff around address look up granularity. Not everything. Also not sure what the status of this application is. The other Japanese patents are probably 'claiming' more things - there's one that seems specifically to make claims around command buffer priorities and the like - but again, don't know if these applications were actually granted.
OP
OP
There are definitely people who dream of memory, and gpu memory, backed by memory-mapped files on fast ssds. A transparent memory address space handled by the OS. (Carmack is one big proponent).
But I'm not sure we're there. It's a complicated problem - I'm not sure how it would work with the security checks and decryption you usually want to have going on with data coming off a hard disk. On the application side you might also want to tune things anyway to avoid page faults. AMD released a workstation GPU a little while back that had SSDs on the GPU - but it wasn't memory mapped, required app support to really take advantage of it.
But games explicitly using RAM as a 'cache' or something 'cache-like' for assets on disc? Games are already doing that, for textures at least. Texture streaming and virtual texturing are variations on the theme.
Of course, a big boost to file IO performance will help those things. File IO is one factor in how much you have 'going on' for a given speed of world traversal, in a streaming system.
Depending on what we actually get, and performance characteristics, it might also be possible to start talking about less preloading and more on-demand asset fetching outside of textures also. Let's say 2GB/s coming off the SSD. That's 34MB of data fetchable within a single frame at 60fps. That's not bad at all really. You could also potentially leave all event-driven data fetches to on-demand loading (e.g. assets needed when a car crashes in your open world game, or assets needed when you trigger a NPC interaction) - not having to keep potential event data warm in memory, and only having to store what you need for events actually in progress = lower initial load times AND higher memory budgets per event AND more memory free for non-event data.
I'm not an expert on this stuff though. Game devs would know a lot better. I do think there is possibly really exciting territory to be explored here by devs for asset streaming and memory management in games if we can start relying on SSD-grade throughput and IOPs, though. We'll have to see what we actually get though, and how much they can remove in terms of performance barriers between ram and ssd.
Last edited:
gofreak getting recognized for this great thread :)
I can't wait to see how the faster drive on the next gen console translates to future and past games.
OP
OP
I mentioned some other Japanese SIE patents by the same inventor, around SSDs - I couldn't initially find their full text, but managed to track them down.
They have some interesting stuff.
AGAIN, A CAVEAT: Don't take it to mean this is what SIE is definitely using in PS5. It gives an idea of what Sony has been thinking about and researching, but maybe these things didn't work out. Or maybe they're now doing something different. Indeed maybe they're fairly bog-standard ideas, or variants on pretty normal things - I'm not a SSD expert! Perhaps they're using ideas like this, or parts of them, or maybe none of them at all :)
This first one talks about a NAND flash controller optimised to provide a maximum worst-case delay for high priority reads from a SSD. It suggests that with the high speed of SSDs, the potential is there for applications to save system memory by delaying some data loads until needed - urgent 'on the spot' data requests from SSD - instead of preloading a lot of data into memory. However using a SSD like this causes a corresponding increase in the frequency of management tasks that the SSD needs to perform in order to keep things working and maintain durability. These management commands can interrupt and delay requests from the application, leading to inconsistent read latency, making the SSD unreliable for frequent on-demand data access. As an example it talks about using data from the SSD during current frame rendering - but if there's a delay, you'll have stuttering or frame freezing.
So it describes in lots of low level detail a flash controller that manages queues of commands of varying priority, predicts when management commands are likely to be made based on access patterns, and juggles these queues to ensure management tasks are performed, while high priority reads get through with a known worst-case delay.
The controller can also accept signals from the application (the game) that hint to the flash controller when it is a good time to perform management tasks, or to pre-emptively do so. It illustrates this example of a game rendering frames using data requested 'on the spot' from the ssd, and signalling to the flash controller in between frames that it's a good time to prioritise any management tasks.
(The shaded '5' is periods of management command processing...the other periods are fulfillment of urgent read requests for each frame)
That's one example, but the host/application could send these signals whenever it knows it's a good time for the SSD to do management stuff.
I'm not sure if something like this would require custom hardware, but I guess if they were using something like that, it would at least require custom firmware in the flash controller on the SSD - for those weighing the odds around whether the SSD will be a proprietary/custom unit.
The second patent is an interesting idea - using the SSD and the SRAM talked about it the other patent as the working memory for the OS in stand by mode, in order to reduce power consumption. Normally in a standby mode the system keeps some DRAM memory 'alive', but doing that with DRAM requires sending power to the memory periodically to refresh it. This patent talks about using the SSD and some SRAM cache as the working memory instead - I guess SRAM doesn't need this power cycling? End result being lower power consumption.
They have some interesting stuff.
AGAIN, A CAVEAT: Don't take it to mean this is what SIE is definitely using in PS5. It gives an idea of what Sony has been thinking about and researching, but maybe these things didn't work out. Or maybe they're now doing something different. Indeed maybe they're fairly bog-standard ideas, or variants on pretty normal things - I'm not a SSD expert! Perhaps they're using ideas like this, or parts of them, or maybe none of them at all :)
This first one talks about a NAND flash controller optimised to provide a maximum worst-case delay for high priority reads from a SSD. It suggests that with the high speed of SSDs, the potential is there for applications to save system memory by delaying some data loads until needed - urgent 'on the spot' data requests from SSD - instead of preloading a lot of data into memory. However using a SSD like this causes a corresponding increase in the frequency of management tasks that the SSD needs to perform in order to keep things working and maintain durability. These management commands can interrupt and delay requests from the application, leading to inconsistent read latency, making the SSD unreliable for frequent on-demand data access. As an example it talks about using data from the SSD during current frame rendering - but if there's a delay, you'll have stuttering or frame freezing.
So it describes in lots of low level detail a flash controller that manages queues of commands of varying priority, predicts when management commands are likely to be made based on access patterns, and juggles these queues to ensure management tasks are performed, while high priority reads get through with a known worst-case delay.
The controller can also accept signals from the application (the game) that hint to the flash controller when it is a good time to perform management tasks, or to pre-emptively do so. It illustrates this example of a game rendering frames using data requested 'on the spot' from the ssd, and signalling to the flash controller in between frames that it's a good time to prioritise any management tasks.
(The shaded '5' is periods of management command processing...the other periods are fulfillment of urgent read requests for each frame)
That's one example, but the host/application could send these signals whenever it knows it's a good time for the SSD to do management stuff.
I'm not sure if something like this would require custom hardware, but I guess if they were using something like that, it would at least require custom firmware in the flash controller on the SSD - for those weighing the odds around whether the SSD will be a proprietary/custom unit.
The second patent is an interesting idea - using the SSD and the SRAM talked about it the other patent as the working memory for the OS in stand by mode, in order to reduce power consumption. Normally in a standby mode the system keeps some DRAM memory 'alive', but doing that with DRAM requires sending power to the memory periodically to refresh it. This patent talks about using the SSD and some SRAM cache as the working memory instead - I guess SRAM doesn't need this power cycling? End result being lower power consumption.
Last edited:
It doesn't. That's the D in various RAMs, Dynamic (needs refreshed) vs Static (doesn't). Still needs power though, just not as much.
fwiw, your caveat is "hard black", difficult to read in dark mode Era.
Last edited:
OP
OP
Excellent work again GoFreak. The 'On The Spot' loading seems to suggest using SSD as an extension to RAM? Ie not having everything preloaded in ram and just loading it as and when it requires?
Im pretty happy with the adoptation, as long as its 1tb ot higher. Putting in a 500gb one would cripple it big time. Game instals on ps4 al already close to 100gb with patches included.
Also an option to use one of the usb ports for external game installation (ala the the ps4 now) day one would be nice. I dont want to wait 3 years to also do that on ps5
Also an option to use one of the usb ports for external game installation (ala the the ps4 now) day one would be nice. I dont want to wait 3 years to also do that on ps5
User Banned (1 Month): Consistent history of platform warring and trolling.
Leave it for Sony to drive inovations, while the likes of Nvidia and Intel keep putting out the same old products but slightly more powerful with a significantly higher price tag, no wonder pc gaming is dying
Another cool find, thanks!
I can't even tell if this is a joke. Did I fall for the bait?
Edit - Grammar.
Last edited:
The only downside of this for me is it will be very hard decision to remove and replace with higher capacity as the SSD in the market isn't as good as the one came with the ps5 console... hope it has the sane spec but with bigger capacity sold separately for replacement.
imo you will never repalce it, it is not SSD but a NAND soldered to the motherboard.
I am still rocking with a 500GB HDD from the beginning of the gen. I'd be fine with whatever storage they give me. I only play one game at a time, so I can manage what is/isn't downloaded.
I am not even sure it will be possible. I think a secondary is most probable, with game you play being on the 1st mandatory.
First post wins!
I don't think he is though, Rigby was even more difficult to understand.
Gofreak was here on the old forum and this not Rigby... Not at all...
OP
OP
Ah cool, looks like it's the final granted patent. Maybe some changes here and there but largely seems similar to the original application.
I don't know much about Linus and it didn't create a good first impression on me when he went on his ill-informed rant about what Tim Sweeny said about the PS5's SSD solution. Apparently he's trying to make amends by educating himself and said he's reading this thread so good on him for that. If you've got a platform that a lot of people look to for information then it's your responsibility to ensure the information you share is correct.
