Digital Foundry Article Technical Discussion [2021]
- Thread starter BRiT
- Start date
- Status
I was thinking more in terms of having lower resolution ones and these being the on demand ones as they seem to be pretty much same quality as XSX & PS5.
So less storage and just stream in the higher assets(current ones) for the required maps.
But I guess we don't actually have any details like the storage difference, quality difference (as Tom couldn't spot any) between non stream and stream mode, to give us a little baseline.
My guess is that the assets at lower quality isn't saving much storage space, even though for XSS any saving is nice.
I'm talking about going the opposite direction than you are.
I'm saying the current textures on XSS would be the upper limit, the ones that get streamed. Starts off lower to save storage space if using streaming.
snc
Veteran
Yes according to df talking with devs its thx to higher clocks on ps5
Granted for the higher clocks but what about the "difference in memory set-up" ? What are they talking about?
https://www.eurogamer.net/articles/digitalfoundry-2021-the-touryst-8k-60fps-on-ps5
https://www.eurogamer.net/articles/digitalfoundry-2021-the-touryst-8k-60fps-on-ps5
snc
Veteran
Maybe at 8k there is need for slightly more mem than 10gb with full speed access for gpu
Off the top of my head, the PS5 unified memory RAM and the XBSX semi-unified (some of the 16GB has a lower throughput)?
At higher resolutions fill rate becomes a larger issue which increases pressure on ROPs and footprint (8K buffers are very large) and bandwidth to perform the function.
Series X compromises in ROPs, split pool bandwidth in favour of cost savings and being compute heavy make 8K for simple games out of reach for it unless you intend to make games purely driven by compute shaders. Maybe there is a chance then. But a game like this where geometry is so simple will take full advantage of the fixed function units. So really hard place for series X to perform as well as PS5 here. Not really sure if it’s possible.
Series X compromises in ROPs, split pool bandwidth in favour of cost savings and being compute heavy make 8K for simple games out of reach for it unless you intend to make games purely driven by compute shaders. Maybe there is a chance then. But a game like this where geometry is so simple will take full advantage of the fixed function units. So really hard place for series X to perform as well as PS5 here. Not really sure if it’s possible.
This is my take. PS5 definitely has more fillrate than XSX, and I suspect is the biggest thing.
Was the XS just BC with 120fps and resolution modes?
PS5 seems to actually had work done to it.
I'm all for each platform being pushed even if it's just due to more time if it came out later.
It's possible they could go back and optimize XS more, but whether it could get to 8K maybe not, and not sure if there would be much point in porting across the updates anyway. (DOF)
Yea I’m not really sure; but pushing the bounds of 8K60 you’re bound to run into a bottleneck somewhere.
If 8K requires more than 10GB of memory, then you’ve spilled into the slower pool. If the bandwidth required for 8K is greater than the slower pool can provided; it’s not going to happen.
If you don’t have the fill rate; it just won’t happen either.
it’s one of those scenarios where you’re just pushing the raw limits of hardware, and it’s a complete bypass of ALU requirements.
I don’t know much about the game. I don’t suspect it can be solved via better programming necessarily.
The gap is big here, PS5 is pushing twice more pixels and it has only ~20% higher clocks.
Yes the downgraded RDNA2 ROPs indeed. PS5, on top of the higher clocks, has twice more Z/stencil ROPs. So with clocks it has actually 140% more Z/stencil ROPs resources which could be very helpful here.
Well, I guess this has nothing to do with the memory. This game has almost no textures to speak of. So 10GB of fast memory should be more than fine in this game.
One thing they mentioned is, that they have rewritten the game for the PS5 native API. That can already make a big difference. And yes, this game has not really much to process so I guess the shaders have not much to do, but the ROPs should have plenty to do and here the PS5 is an advantage if it can maintain the highest clock-speeds (and as the rest of the game is really simplistic this should be the case).
Btw, great game, who didn't try it should really try it, if you like such platform puzzle games.
One thing they mentioned is, that they have rewritten the game for the PS5 native API. That can already make a big difference. And yes, this game has not really much to process so I guess the shaders have not much to do, but the ROPs should have plenty to do and here the PS5 is an advantage if it can maintain the highest clock-speeds (and as the rest of the game is really simplistic this should be the case).
Btw, great game, who didn't try it should really try it, if you like such platform puzzle games.
davis.anthony
Veteran
I think the comment regarding memory is likely because the game could, in certain situations, require more then 10Gb of VRAM.
It doesn't have to require that much all the time but even a handful of times where the game slows to a call because it's above 10Gb would enough to make the devs drop down to 6k.
It doesn't have to require that much all the time but even a handful of times where the game slows to a call because it's above 10Gb would enough to make the devs drop down to 6k.
RDNA2 runs RB+ which is a 2x increase for colour rops provided you don’t exceed the 32bpp colour rate, and it supplies the shader engine at 2x as well.
right. 2x depth rops. Not sure in practice how much can be additionally exploited. Thinking out loud I suspect they weighted towards colour rops over depth on purpose; likely because there wasn’t enough colour rops.
but at the same time; AMD never designed a single RNDA2 GPU with so few rbes; so perhaps xsx is deficient in this area.
Last edited:
- Status
Similar threads
