Digital Foundry Article Technical Discussion [2021]

[Tkumpathenurpahl]

Its native BC, not native to PS5.
Native BC implies it' runs on PS5 boost rates. Not that it's leveraging the IPC of RDNA2.

Frankly, I suspect both of these consoles are likely running the slower APIs, so DX11 and GNM+ respectively given the age of the engines in question (I doubt they converted to a low level API). So performance output is dismal compared to say a 3060 or something that eats DX11 titles.

Taken alone, native may imply that BC games run at the PS5's full clockspeed, but there's already evidence that PS5 developers may choose from multiple fixed profiles if they don't wish to deal with the dynamic aspect of Smartshift. I seem to recall someone from Sony confirming the existence of profiles, although they didn't get into specifics (I couldn't find a reference for that one, although I admit I didn't look for long.)

I wouldn't be shocked if there are profiles available for BC games too. It's probably a lot easier to QA test your game across PS4/Pro/5 if the PS5 is locked at, for example, 3.2GHz and 1.8GHz.

Edit: I've found this interview with Cerny.

He confirms profiles on dev kits:

Regarding locked profiles, we support those on our dev kits, it can be helpful not to have variable clocks when optimising.

But goes on to say:

Released PS5 games always get boosted frequencies so that they can take advantage of the additional power.

So I'm sort of at an impasse, purely because a PS4 game running on the PS5 may not qualify as a "released PS5 game." Is there any information confirming the use of full PS5 clockspeeds on BC games?

 

[iroboto]

Taken alone, native may imply that BC games run at the PS5's full clockspeed, but there's already evidence that PS5 developers may choose from multiple fixed profiles if they don't wish to deal with the dynamic aspect of Smartshift. I seem to recall someone from Sony confirming the existence of profiles, although they didn't get into specifics (I couldn't find a reference for that one, although I admit I didn't look for long.)

@Dictator did. Though that's only for making development easier. The developer has no control over the clocking profile on release.

So I'm sort of at an impasse, purely because a PS4 game running on the PS5 may not qualify as a "released PS5 game." Is there any information confirming the use of full PS5 clockspeeds on BC games?

None that I know of. But assumptions should be made here considering that developers must test their PS4 title against PS5 to ensure that it will run in native mode without crashing. So boost is certainly involved, and there is a variant of the title that only exists for PS5 - so logic dictates that boost is what enables this mode to occur, as a Pro Legacy title would be locked to Pro clockspeeds.

I have no issue calling it full boost as per any PS5 title would have access to. We use this term for all PS5 titles, until there is evidence to the contrary that we find out the boost profile for BC titles is less than the 2230Mhz that a PS5 binary has. There are a variety of reasons here why both consoles aren't performing well, and a majority of it being GCN combined with likely, running a high level API like DX11 and GNM+ (unless EA in the goodness of their heart wants to re-write Unreal engine 3 for DX12 and GNM). That's going pull back a lot of performance in itself and I have no reason to suspect that clockspeeds are not at maximum. The game is railed at 4K@60 for 99.98% of the time, with some awkward dips here and there. Those awkward dips are not performance issues with the GPU, but likely some of the pain points caused by a variety of factors around memory management. It should be noted that real frame graphs do not actually look like those shown on VG Tech or any site really, these are smoothed values and not the raw figures. If you decide to smooth the values over the last second (60 frames), you will see bigger swings, when in reality, it might be a temporary hiccup for a split second. There are so many algorithms for smoothing, so I can't tell you which one each site uses. But I can tell you that the values are definitely smoothed as real frame time graphs are likely to look much noisier when they drop below the clamp.

That being said, when you look at PS5 results, the results show me roughly near perfect. Not sure what the complaint is here. If the complaint is that it should be doing better, well, yes. Both consoles should be doing better. They're getting the pants beat off them by a 3060. But this is natural considering one is running DX11 native, and is designed to do it natively Vs 2 consoles that are emulating GCN2.0.

There is no controversy here for me, we have no idea how emulation affects either system in terms of performance loss, and it may or may not be symmetrical loss either.

However, it is clear clockspeeds are the enabler for 2x the frame rate over 4Pro, and it outputting 2x the frame rate while emulating the 4Pro seems to be fairly reasonable, in fact very good.

 

[PSman1700]

hey're getting the pants beat off them by a 3060.

Not all that surprised, a 3060 is a 13TF GPU on a very modern architecture, with 360gb/s of bw all to itself (12gb). Thats not thinking ray tracing. It should be faster then both, but not either very large difference either though when normal rendering using no reconstruction tech etc.

 

[Deleted member 11852]

I don't think many other companies would succeed as quickly as this. This is a very well run studio. Honestly I'm really impressed.

I agree but I think you misunderstand. This is a good studio and it still took then six months. All I'm saying is six months work is not an insignificant undertaking. Six months is a long arse-time. Metro was always a game predicated on it's use of lighting and I imagine it took quite a while to replace all the fake lighting with RT lighting that didn't look weird where you needed to do a bit of fakery.

It's not a small game.

 

[iroboto]

I agree but I think you misunderstand. This is a good studio and it still took then six months. All I'm saying is six months work is not an insignificant undertaking. Six months is a long arse-time. Metro was always a game predicated on it's use of lighting and I imagine it took quite a while to replace all the fake lighting with RT lighting that didn't look weird where you needed to do a bit of fakery.

It's not a small game.

exactly and I'm in total agreement here. They had to go back and re-light the whole thing without changing geometry, without changing anything else about the game and still get similar desired effect. If you're making a new game from scratch and you are building the game ground up with RTGI and dropping the older static method off completely, the geometry and levels will be designed with RTGI in mind as per what their editor will show. And instead of going back to figure out how you did it before, you won't need to, because you're doing everything for the first time around RTGI as opposed to designing a level for static lighting, and then redoing it for RTGI.

And so, once again, to do this all over again, while building out the technology, it's still extremely fast. I get that six months work is not insignificant, but they did just re-write their entire lighting pipeline and redid the whole lighting for the whole game. That's not insignificant in the least considering there are still performance implications that lighting artists and developers must meet for consoles. Luckily they won't need to redo their lighting pipeline _again_ for their next title.

 

[iroboto]

Not all that surprised, a 3060 is a 13TF GPU on a very modern architecture, with 360gb/s of bw all to itself (12gb). Thats not thinking ray tracing. It should be faster then both, but not either very large difference either though when normal rendering using no reconstruction tech etc.

It’s not equivalent. PS5 and XSX have to emulate PS4 Pro and X1X respectively. While running on what could be a port or wrap over or Dx11 over DX9.

DX11 and older DX releases are much more CPU bound than Dx12, so PC with powerful CPUs can crunch through this. Consoles will lose some to emulation and additionally to the API.

 

[PSman1700]

It’s not equivalent. PS5 and XSX have to emulate PS4 Pro and X1X respectively. While running on what could be a port or wrap over or Dx11 over DX9.

DX11 and older DX releases are much more CPU bound than Dx12, so PC with powerful CPUs can crunch through this. Consoles will lose some to emulation and additionally to the API.

True, on the other hand, the 3060 is technically the more capable GPU. It should perform somewhat better, but not by huge amounts either.

 

[iroboto]

True, on the other hand, the 3060 is technically the more capable GPU. It should perform somewhat better, but not by huge amounts either.

It’s performing 4K@130fps with the right CPU. XSX can only do 1440p and below 120fps at the same points. This is just reality of it

 

[dobwal]

I agree but I think you misunderstand. This is a good studio and it still took then six months. All I'm saying is six months work is not an insignificant undertaking. Six months is a long arse-time. Metro was always a game predicated on it's use of lighting and I imagine it took quite a while to replace all the fake lighting with RT lighting that didn't look weird where you needed to do a bit of fakery.

It's not a small game.

Hard to know as we don't know the number of resources allocated to support the work. Would six months be a long time if the majority of the devs were devoted to the project? Probably so, but what if only a small subset were devoted to the project?

A4 has been working on the next Metro since last year, which started with a complete overhaul of the engine that targets new-gen console hardware and the latest PC GPUs with the incorporation of RT. I imagine a lot of that work showed up in the enhanced version. So how do you parse out the amount that simply went into the enhanced version as a bulk of the work is dedicated to development beyond that endeavor?

 

[PSman1700]

It’s performing 4K@130fps with the right CPU. XSX can only do 1440p and below 120fps at the same points. This is just reality of it

Ye the XSX sure underperforms there yes, should be closer to the 3060. Besides software, CPU could play a role too seeing the high framerates (at high resolutions aswell).

 

[Tkumpathenurpahl]

@Dictator did. Though that's only for making development easier. The developer has no control over the clocking profile on release.


None that I know of. But assumptions should be made here considering that developers must test their PS4 title against PS5 to ensure that it will run in native mode without crashing. So boost is certainly involved, and there is a variant of the title that only exists for PS5 - so logic dictates that boost is what enables this mode to occur, as a Pro Legacy title would be locked to Pro clockspeeds.

I have no issue calling it full boost as per any PS5 title would have access to. We use this term for all PS5 titles, until there is evidence to the contrary that we find out the boost profile for BC titles is less than the 2230Mhz that a PS5 binary has. There are a variety of reasons here why both consoles aren't performing well, and a majority of it being GCN combined with likely, running a high level API like DX11 and GNM+ (unless EA in the goodness of their heart wants to re-write Unreal engine 3 for DX12 and GNM). That's going pull back a lot of performance in itself and I have no reason to suspect that clockspeeds are not at maximum. The game is railed at 4K@60 for 99.98% of the time, with some awkward dips here and there. Those awkward dips are not performance issues with the GPU, but likely some of the pain points caused by a variety of factors around memory management. It should be noted that real frame graphs do not actually look like those shown on VG Tech or any site really, these are smoothed values and not the raw figures. If you decide to smooth the values over the last second (60 frames), you will see bigger swings, when in reality, it might be a temporary hiccup for a split second. There are so many algorithms for smoothing, so I can't tell you which one each site uses. But I can tell you that the values are definitely smoothed as real frame time graphs are likely to look much noisier when they drop below the clamp.

That being said, when you look at PS5 results, the results show me roughly near perfect. Not sure what the complaint is here. If the complaint is that it should be doing better, well, yes. Both consoles should be doing better. They're getting the pants beat off them by a 3060. But this is natural considering one is running DX11 native, and is designed to do it natively Vs 2 consoles that are emulating GCN2.0.

There is no controversy here for me, we have no idea how emulation affects either system in terms of performance loss, and it may or may not be symmetrical loss either.

However, it is clear clockspeeds are the enabler for 2x the frame rate over 4Pro, and it outputting 2x the frame rate while emulating the 4Pro seems to be fairly reasonable, in fact very good.

There are approximately 875 conversations all involving you or directed at you in this thread at the moment, so I can understand why it might be a bit tricky to remember who's saying what.

My position isn't at all predicated on the Mass Effect Remaster. I never clicked with those games so I've not paid much attention to the performance. A cursory glance out of interest, but that's all. As you said, performance is pretty rock solid across all current generation consoles. That seems about in line with what we should expect at this point.

I'm only querying the matter of the PS5's BC being full 2.23GHz variable clockspeeds. I'm leaning more towards it being in the realm of ~1.8-1.85GHz as that would still give ample performance to double framerate. Couple that with the fact that I've mostly played PS4 games on my PS5, and I've found that even the more demanding PS4 games don't seem to produce as much heat as even the less demanding PS5 games. For example, I'm playing the PS5 upgrade (it was free, the bloody legends!) of Subnautica at the moment, and it seems to result in more heat than Ghost of Tsushima.

That strikes me as PS5 native games utilising Smartshift, with the SoC getting a full workout, while PS4 games don't, even when their framerate is doubled.

 

[Metal_Spirit]

As there seems to be some confusion on how the BC works on playstation, I will leave this:
BC has 3 modes:

BC mode 1 - PS5 clocks on both CPU and GPU are locked to the original PS4 console (1.6 GHz on CPU and 800 MHz on GPU, with 18 CU). - In this mode a game will act as if played on a standard PS4.

BC mode 2 - Same as above, but with PS4 Pro clocks and CU (2.1 GHz on CPU, 911 MHz on GPU, with 36 CU). n this mode tha game will act as if played on a standard PS4 Pro

BC mode 3 - This is the most complex. Games are given the full PS5 clocks, meaning a 10.28 Tflops GCN PS4. This works as a PS4 Pro wih extra clocks, and cannot do anything the original console would not do if, by any chance, it were overclocked. Meaning that since PS4 Pro cannot output 120 Hz, the PS5 cannot do it either. - This mode will act as a PS4 Pro with extra clocks, but withing the coding limitation imposed on the original console, never thought for 10,28 Tflops.

This means that if the Pro code allows for dinamic resolution up to 4k, the PS5 can go to better resolutions, but if there is a limit of some sort though because of the original console constrains, PS5 will not overcome them. The 120 Hz is an example, since the PS4 Pro cannot do it, then PS5 will not do it either, even if it has available power.

For no old generation imposed limits at all, games must be created and compiled as native PS5 games, meaning they will have a separate version.

But although this is the only way to remove limits, a separate version doesn't necessarily mean real improvements, just the possibility for them to exist.
If a game is just recompiled for PS5 native mode, will still act as a BC mode 3 game, although now you can take advantage of the PS5 hardware feature set. This means that at worst you can have a PS4 Pro game with some new features, like with RT, although the constrains of the original game for the PS4 Pro Will remain, but taking advantage of RT. Same goes for the SSD or any extra new features. At best, you can have a full fledged PS5 game, but that will probably not occur so soon, since with Cross Gen development there are advantages on some code sharing using dinamic situations!

For real advantage on the PS5 hardware, a game has to be though for it from it´s creation, removing restrains that could exist on old hardware. Even the new Ratchet will probably not use PS5 to it´s fullest since the base (although it might be heavily changed by now) it has it´s origin on a highly optimized PS4 engine. And it is too soon to believe the engine is now fully modified for the best PS5 optimization.

 

[Deleted member 11852]

Hard to know as we don't know the number of resources allocated to support the work.

Imagine 4A games will clarify that with their second interview that DF says is with CTO Oles Shishkovstov and senior rendering programmer Ben Archard who will "go into extreme depth on the development of the 4A Engine's brand new ray tracing features.". This update had an executive producer (who DF just interviewed) and they have their CTO and senior rendering guy on it which makes it sounds like this was a considerable project.

This wasn't a "put the interns on it" project.

 

[mr magoo]

Imagine 4A games will clarify that with their second interview that DF says is with CTO Oles Shishkovstov and senior rendering programmer Ben Archard who will "go into extreme depth on the development of the 4A Engine's brand new ray tracing features.". This update had an executive producer (who DF just interviewed) and they have their CTO and senior rendering guy on it which makes it sounds like this was a considerable project.

This wasn't a "put the interns on it" project.

there are no more interns, everybody senior since covid

 

[dobwal]

Imagine 4A games will clarify that with their second interview that DF says is with CTO Oles Shishkovstov and senior rendering programmer Ben Archard who will "go into extreme depth on the development of the 4A Engine's brand new ray tracing features.". This update had an executive producer (who DF just interviewed) and they have their CTO and senior rendering guy on it which makes it sounds like this was a considerable project.

This wasn't a "put the interns on it" project.

True. But this isn't an all-hands-on-deck project either. This is just revisiting an old game that allows leveraging a bunch of prior effort while also benefitting from the work aimed at the next Metro.

It would much harder to give away the upgrade to existing Metro owners if the cost of development dedicated specifically to this project was significant.

 

[eastmen]

looks good , would have been better with reworked textures

 

[BRiT]

DF Article @ https://www.eurogamer.net/articles/digitalfoundry-2021-days-gone-pc-port-analysis

Days Gone PC: a quality conversion that elevates the console experience
And performance is solid too.

Sony promised us more PC conversions of their stellar first party development output and while Horizon Zero Dawn illustrated that this is far from a simple process, Days Gone is on another level. The upgrades are strategically chosen but effective - and performance is solid on both Nvidia and AMD hardware. In fact, there are one or two touches here and there included in this game that really hope to see other developers bring to their own titles, especially when it comes to configurability. While we're not getting the ultimate package here - there's no ray tracing and disappointingly, no DLSS - there's no doubt that this is a solid, impressive port.

Booting up the PC version of Days Gone, it's immediately apparent that this isn't just a basic PS4 Pro conversion. The game is based on Epic's Unreal Engine 4 and I was happy to see that Bend Studio had updated the PC version to include a more recent UE4 innovation - software-based screen-space ray traced global illumination - available as an upgrade over the standard ambient occlusion tech deployed on the console versions. Essentially, the appearance of light bouncing around a game scene is emulated in screen-space, delivering a richer presentation with more realistic lighting and better, more realistic shadows. If light hits a red surface, for example, some measure of that 'redness' will illuminate the surroundings where appropriate. If this sounds familiar, The Coalition made a similar upgrade to the Xbox Series and PC versions of Gears 5, again by dipping into more recent UE4 engine updates and adding it to their existing codebase. The old SSAO is still there though for that authentic console look - it's now simply the 'high' setting for the in-game lighting.

...

 

[snc]

one remark to df days gone video, there should be info that 4k cb is not just rendering at around 1512p resolution as cb upscaling is not performance free

 

[Deleted member 86764]

Can this really be called "the PS5 version"? It's a PS4 game running in backwards compatibility.

Stating that "the PS5 version has lower than low textures" strikes me as a misnomer.

Would the Xbox Series X running 360 code in BC be considered the "Xbox Series X version"?

 

[PSman1700]

Nice analysis as usual! Its the PS5 version on steroids (higher settings and even faster loading using a fast nvme drive). The ultimate Days Gone experience for those that want the best version out there.
For console-like settings you'd want a RTX2060 or RX5700XT.