PC system impacts from tech like UE5? [Storage, RAM] *spawn*

Hmm imo, They address different things. SSD supports a high I/O rate which enables you to have access to a large catalog of assets. It’s solves the problem of immediacy. If you have really slow I/O your buffering is larger, and the larger it is the more data you need to buffer in random locations. A lot of it wasted. However SSD doesn’t let you render more on screen at once. As you’re limited to 16GB of space. Once filled the I/O must stream textures out to put textures in.

Having more VRAM let’s you actually render more on screen at once. Because you can have more in VRAM. But if you have slow I/O you’ll need to ultimately use some of that spade for buffering depending on how slow that I/O is.

All in All, having super fast I/O coupled with a huge amount of VRAM would net some gnarly results in asset fidelity.

I apologize as i can't find the slide , but during the ps5 ssd presentation from gdc they showed how many duplicate textures and data they could remove from the game because of the ssd and it went down from like 80gigs to 10gigs. On the pc if you have enough ram you'd be able to load all of spiderman into your system ram. You would have no loading time at all. You could double the data set from Spiderman the game to 20gigs and still have left over ram if you have a 32gig ($150usd) ram set up. So instead of streaming from ps5 ssd which gives you what 8 gigs of compressed data transfer rates ? Single channel ddr 3200 ram gives you 25.6GB/s bandwidth and dual channel is 51.2GB.

So if you have a game that exceeds the ram in your system you will then have to hit your hard drive in your pc. So on a pc you'd have graphics ram (4-16gigs?) - main system ram(16 to 128gigs) - hard drive. On console you have main ram 16gigs to ssd 825 to 1tb .

Lets not forget that while currently the ssd's in a pc are slower it doesn't mean it will stay that way. But even still you'd be fetching data from a 3-6GB/s ssd

And then there was DDR5
https://www.anandtech.com/show/1591...sed-setting-the-stage-for-ddr56400-and-beyond

The pc is an ever changing platform and its hard to judge it against a fixed platform.
 
But b/w is b/w however you use it right? So it’s not less b/w in the PC scenario (again unless I’m misunderstanding how tech works).

I'm not sure I fully understand the question so apologies if my answer has nothing to do with what you said! My point was that a PC with lots of DRAM wouldn't need as much IO bandwidth because the IO would be used less. To give a really simplified example, say the PS5 in order to show some new scene needs to stream 100 assets into VRAM from the SSD requiring 8GB/s of bandwidth. But the PC already has 50 of those 100 assets stored in system memory thanks to pre-caching, so it only needs to bring in the other 50 from the SSD with a requirement of 4GB/s.

Consoles only need 16gb because SSD compensates and isn’t the console setup is currently superior to what PC can currently offer?

Ideally they'd have RAM that exceeded the full game size. Anything less than that is a compromise even with fast IO because the IO is still much, much slower than the RAM. They don't only *need* 16GB, they only *have* 16GB because more would have been uneconomical.

As stated games will only get bigger so that’s a smaller and smaller % of the game in RAM and that’s why having the whole game on SSD with ultra fast IO is the better solution...unless you up the RAM each time.

The PS5 only has an 825GB SSD so games aren't going to get that big! I think 100GB on average is likely. And of course, bigger games can have bigger RAM requirements. Don't forget we're only talking about what's needed here to match the console experience, not to run the game. For a huge game that allows the player to access huge swathes of the game content at the drop of a hat, developers could for example recommend either 64GB RAM or a 7GB/s NVMe drive. However for people that don't have one of those, the game would scale back texture resolution, LOD's, draw distance, or simply add loading screens so that the minimum requirement is far lower. That's how it's always been in the PC market. There have always been games that if you want to max them out you need bleeding edge technology, but they always scale down to lesser systems too.
 
He talked about future hw and noted the comparison is not fair. His argument was that he can only go by what we have today and this would have to be revised later. His thinking was that what does it cost today to buy a console killer.
I’m with pbjliverpool on this. Building a console killer today is pointless, and the only ones that that would even suggest the idea are people who have a vested interest in sales. It becomes relevant if/when there are software titles out that depend on the new features of the consoles enough that the PC experience is compromised without a system build that takes the architecture of the consoles into consideration. That time is not now.
Arguably, PCs would have a harder time adjusting to titles that assume that the CPU has access to RAM at 400GB/s in order to achieve good frame rates. That might cause some issues...
 
I apologize as i can't find the slide , but during the ps5 ssd presentation from gdc they showed how many duplicate textures and data they could remove from the game because of the ssd and it went down from like 80gigs to 10gigs.
I think you’ll have to find the slide, because this is wrong from what I recall, it was more like a 10gb saving on space.

edit - yep, it’s a 10gb saving;
https://www.psu.com/news/ps5-ssd-could-cut-game-sizes-to-save-valuable-space-on-your-console/


I'm not sure I fully understand the question so apologies if my answer has nothing to do with what you said! My point was that a PC with lots of DRAM wouldn't need as much IO bandwidth because the IO would be used less. To give a really simplified example, say the PS5 in order to show some new scene needs to stream 100 assets into VRAM from the SSD requiring 8GB/s of bandwidth. But the PC already has 50 of those 100 assets stored in system memory thanks to pre-caching, so it only needs to bring in the other 50 from the SSD with a requirement of 4GB/s.



Ideally they'd have RAM that exceeded the full game size. Anything less than that is a compromise even with fast IO because the IO is still much, much slower than the RAM. They don't only *need* 16GB, they only *have* 16GB because more would have been uneconomical.



The PS5 only has an 825GB SSD so games aren't going to get that big! I think 100GB on average is likely. And of course, bigger games can have bigger RAM requirements. Don't forget we're only talking about what's needed here to match the console experience, not to run the game. For a huge game that allows the player to access huge swathes of the game content at the drop of a hat, developers could for example recommend either 64GB RAM or a 7GB/s NVMe drive. However for people that don't have one of those, the game would scale back texture resolution, LOD's, draw distance, or simply add loading screens so that the minimum requirement is far lower. That's how it's always been in the PC market. There have always been games that if you want to max them out you need bleeding edge technology, but they always scale down to lesser systems too.
It’s probably because I’m not asking the right question.

But essentially my brain thinks that in your previous PC example (22gb usable RAM vs 12gb useable RAM for console ) the RAM still needs to utilise bandwidth to fill it up. In fact in my brain the PC is less efficient with bandwidth because it’s loading 22gb of assets it may only use 10% of.

Due to the IO speed the consoles don’t need to preload a lot of assets because they can load them really quickly, wasn’t it 2gb in half a second or something? So it preloads into RAM just the bits it will need really quickly and the rest can just sit on the SSD until needed.

Like I said though, that’s just the way my brain is thinking the tech works.

And regarding SSD space, it will be ok due to being able to customise the downloaded/installed game, I suspect we may have options to have just the story or online part - or even maybe just the level (and next level) of a story driven game. Then we will get plug in SSDs too.
 
Last edited:
IIRC one of the slide also mention to preload ~2 seconds of gameplay

I recall Cerny went into detail regarding how much more efficient it will be vs more RAM - I will need to rewatch.

https://www.google.com/url?sa=t&source=web&cd=&ved=2ahUKEwjexeSl0dnqAhWSTcAKHdnAB9sQFjAAegQIBRAB&url=https://venturebeat.com/2020/03/18/playstation-5-mark-cernys-deep-dive-into-ssd-memory/amp/&usg=AOvVaw3sxrivETbATFZPWMv4Jl4K
With a faster SSD, system memory (random access memory, or RAM) can now be used more efficiently. On the PS4, system memory had to store all the memory that could be used in the next 30 seconds of gameplay. So data was sitting in memory on the potential to be used. With 16GB of main memory, or random access memory, the loading is fast enough to get the data and load it far more quickly, without needing to store so much in a cache. A lot of the PS4 data never got used. That’s not the case with PS5.
 
Let's just move back to spinning disks. His problem is that he would need to keep spinning disk alive for users without having to tell them how to configure power settings(i.e. control the disk activity state from inside the game)


In all seriousness, the stuff that dude is doing looks great

This guy has one of the most exciting potential for games I’ve seen in a long time. Pure vozel based engine. Looks great. Very dynamic. Lots of potential here for great gameplay.
 
It’s probably because I’m not asking the right question.

But essentially my brain thinks that in your previous PC example (22gb usable RAM vs 12gb useable RAM for console ) the RAM still needs to utilise bandwidth to fill it up. In fact in my brain the PC is less efficient with bandwidth because it’s loading 22gb of assets it may only use 10% of.

Yes the initial load to RAM will be bandwidth heavy on the SDD. In my example it should have actually be 32GB of usable RAM, not 36GB like I initially said (HBCC seems to leave 8GB for the rest of the system). However you'd only need to load in what you're going to immediately use to get the game running and then you can pre-cache the rest in the background. You're only talking about 30 seconds to completely fill that RAM even on a relatively slow 1GB/s drive but you might be able to actually start the game after the first 8 seconds once the 8GB VRAM is filled.

Once the RAM is filled you'd then just be topping up based on player actions in exactly the same way as the consoles would, except in the PC scenario you'd be topping up from SSD less often because you have a wider selection of data already available in RAM. So your ongoing IO bandwidth requirements are lower.

Due to the IO speed the consoles don’t need to preload a lot of assets because they can load them really quickly, wasn’t it 2gb in half a second or something? So it preloads into RAM just the bits it will need really quickly and the rest can just sit on the SSD until needed.

While that's correct in theory, the reality is that they'll still be pre-caching a lot of data in VRAM because actually 16GB of VRAM is pretty huge compared with a game install of say 100GB. So naturally if you can store 16% of your entire game content in RAM at any given time it's probably going to represent more than just the next couple seconds of gameplay. Unless your game is really short that is.

Nevertheless though, the new consoles could technically do this if they weren't constrained by game install sizes. And yes you could argue that it's a more efficient design than having lots of RAM and pre-caching lots of data of which you won't need a large percentage. The RAM option is more a brute force approach which isn't out of character for the PC market. And while the RAM approach has limitations compared with fast IO, it also has advantages because for that portion of the data that you do have stored in RAM, you have much faster, lower latency access to it that even the PS5 would from it's SSD.

To give a real world example of how that might manifest, imagine a scenario where you're using Sampler Feedback to determine what texture blocks you need to load in from memory in the next frame. On the PC, that texture is in system memory because it was pre-cached in case it was needed. On the PS5 it wasn't. But even with the PS5's insane IO speed, there just isn't enough time to load that texture block in readiness for the next frame. So the result is a jarring mip transition on the texture. However on the PC, the texture is fetched from main memory into VRAM much faster and this arrives in time for when it's needed. No jarring mip transition. We know this happens in the real world even with fast IO because the XSX has dedicated hardware built in to mask that transition.
 
Yes the initial load to RAM will be bandwidth heavy on the SDD. In my example it should have actually be 32GB of usable RAM, not 36GB like I initially said (HBCC seems to leave 8GB for the rest of the system). However you'd only need to load in what you're going to immediately use to get the game running and then you can pre-cache the rest in the background. You're only talking about 30 seconds to completely fill that RAM even on a relatively slow 1GB/s drive but you might be able to actually start the game after the first 8 seconds once the 8GB VRAM is filled.

Once the RAM is filled you'd then just be topping up based on player actions in exactly the same way as the consoles would, except in the PC scenario you'd be topping up from SSD less often because you have a wider selection of data already available in RAM. So your ongoing IO bandwidth requirements are lower.



While that's correct in theory, the reality is that they'll still be pre-caching a lot of data in VRAM because actually 16GB of VRAM is pretty huge compared with a game install of say 100GB. So naturally if you can store 16% of your entire game content in RAM at any given time it's probably going to represent more than just the next couple seconds of gameplay. Unless your game is really short that is.

Nevertheless though, the new consoles could technically do this if they weren't constrained by game install sizes. And yes you could argue that it's a more efficient design than having lots of RAM and pre-caching lots of data of which you won't need a large percentage. The RAM option is more a brute force approach which isn't out of character for the PC market. And while the RAM approach has limitations compared with fast IO, it also has advantages because for that portion of the data that you do have stored in RAM, you have much faster, lower latency access to it that even the PS5 would from it's SSD.

To give a real world example of how that might manifest, imagine a scenario where you're using Sampler Feedback to determine what texture blocks you need to load in from memory in the next frame. On the PC, that texture is in system memory because it was pre-cached in case it was needed. On the PS5 it wasn't. But even with the PS5's insane IO speed, there just isn't enough time to load that texture block in readiness for the next frame. So the result is a jarring mip transition on the texture. However on the PC, the texture is fetched from main memory into VRAM much faster and this arrives in time for when it's needed. No jarring mip transition. We know this happens in the real world even with fast IO because the XSX has dedicated hardware built in to mask that transition.

Thanks for a great explanation.

I still think it’s not cut and dry (at 32gb RAM) - firstly the RAM will mean potential performance issues during the initial load (maybe just 30 seconds and after fast travel etc) whilst the system is loading data it doesn’t need.

We know the PS5 was lots of priority feeds so could potentially win out on those scenarios.

Also, if you wanted a random texture that wasn’t preloaded (as given is a super open world where you might use binoculars to look at an area miles away) the IO and priority feeds might help out there.

I get that having more RAM is better (obviously have the whole game in RAM would be awesome), but I think we’re comparing apples to oranges- be interesting to see the first few face offs with these things in mind.
 
From my point of view, I believe what devs are saying and I think it will take time for engines to be tailored around next-generation SSD.

Out of first part exclusives I doubt for the first two to three years we will see the streaming well exploited.

After maybe it will be not true Battlefield 6 will be interesting and the two devs said than it will be easier for custom engine like the Frosbite to pivot.
 
I believe in results. The ue5 demo (showing ssd abilities9 apparently, according to devs, ran even better on a well specced laptop with nvme ssd. Doubt that system was compensating with 32gb ram.

Again they never said it run better we have no res, no details. No one said it runs with the same quality than the PS5 demo and this is important because UE5 scale with streaming capacity and power too

Maybe it runs at40 fps in 1080p maybe the quality of assets was not the same than the PS5 demo.
 
Back
Top