Curious, but did you have a chance to match PS5's settings to see how it stacks up to a mid-range PC like one with a 3060/2070S? This game runs extremely well on PS5 and the performance bottoms out at 1440p I believe and never dips below 60. That a 4060 cannot maintain 60 even with DLSS Quality at 1440p (base 1080p) suggests that the PS5 runs the game much much better than PC equivalent parts.
Digital Foundry Article Technical Discussion [2024]
- Thread starter Shifty Geezer
- Start date
8GB VRAM cap on PC may be playing a large factor here for the 4060 coupled with 262GB/s of bandwidth; it's extremely low, I suspect it's hitting bandwidth limits here.
Flappy Pannus
Veteran
Fantastically detailed as usual @Dictator.
So like most Sony ports - relatively solid, but still needing some patches. As Alex mentioned, a lot of those are rather bizarre (especially when the PS4 version can look superior) and relegated to specific areas/configs, so I'm sure they're bugs. The devs have previously been responsive online so I'd expect most of them to be addressed.
Yeah, that's the kind of comparison I was expecting and was surprised to not see it. How a midrange PC stacks up like for like with the PS5 version in raw performance is a part of the ports quality, I think we all expect at this point for Sony studio games with custom engines to require more GPU grant than you would expect and frametime consistency is key, but the use of DLSS in the 4060 at 1440p would imply the PS5 version is running far, far better - at least in that config.
Here's a 4070 running the game at 4k DLAA Ultra for comparison:
Now the PS5's Quality mode has a 40fps option, with an 1800p to 2160p dynamic resolution window - and that 40fps is a solid lock from all reports I've seen. So yeah, very GPU demanding - from these numbers the PS5 Pro version will probably deliver better results than a 4070 super.
Not fantastic sure, but as I've always said with these demanding Sony ports I've far more concerned with frametime consistency - you can brute force/tweak your way with bottlenecks that are purely hardware but shader/traversal stutter, you're fucked. They got some areas to iron out (UW displays) sure, but overall I certainly prefer that focus.
On the issue of the cubemap reflections not matching the PS5's quality mode btw, perhaps one of the bottlenecks is they spent a bunch of time trying to get DXR to work but couldn't do it, so they're falling back to 'basic cube maps'. The intention may have been to replace the more accurate cubemaps Quality with just DXR, but the result of that failure also meant they didn't have to time to add in the more accurate cubemap version. There is still the possibility than DXR may come to the PC version later though:
From Eurogamer's interview:
I could try and run PS5 like settings in a one of test for you guys next week maybe, but for the video I did not include it as it is impossible to run PS5 settings due to things like DRS and the shadow setting not matching. One thing to mention is TAAu is way cheaper than FSR, XeSS or DLSS in this game.
I'mDudditz!
Newcomer
There is no DRS in the PS5 VRR performance mode. Just match all the settings to PS5 performance like you have done in the past.
This would be greatly appreciated. I'm really curious to see how a pure console game does in a PC environment using DX12.
There is a 1440p high frame rate mode, which according to John has the same visual make-up as the performance mode but with an unlocked frame rate on 120Hz screens. The resolution is static so no need to worry about DRS there.
The shadows are a bigger problem because if they are indeed between low and medium across the board, that's a big difference. Low seems to run about 20% faster than medium and if the PS5 has a custom value below medium and above low, it could claw back anywhere from 3-18% of the performance. That's a huge deal.
Also, disappointed in this port. Not only does it not scale above the console version, it's actually worse in terms of visuals. Those PS5 to PC ports are seriously underwhelming. I'm glad that they all include DLSS and ultrawide support, but they're often plagued by performance problems, bugs, and other issues. They also don't put any effort into scaling them much beyond their console version outside of Rift Apart and Returnal.
Last edited:
That statement may be true but, I have no way of verifying that on my own. For performance deltas, it should be without DLSS. If DLSS data is provided in addition, that's fine but I'm not a fan of presenting performance aided by DLSS as actual performance. It's not the same thing but, that's just my opinion.
Flappy Pannus
Veteran
Eh...on that. They've dropped a recent patch and apparently made some significant changes to how VRAM is handled, and it's breaking a lot of things. Yikes.
Eh...on that. They've dropped a recent patch and apparently made some significant changes to how VRAM is handled, and it's breaking a lot of things. Yikes.
Yea. They messed with analog settings on controller so now the camera judders as you recenter the stick and Kratos as well when you're moving sometimes. That might be a Steam Input thing though, never tested with it off. Textures also unload as you're in the menus after a few seconds and when you return back to the game all the LODs have to stream in.Eh...on that. They've dropped a recent patch and apparently made some significant changes to how VRAM is handled, and it's breaking a lot of things. Yikes.
This is quite broken now.
Eh...on that. They've dropped a recent patch and apparently made some significant changes to how VRAM is handled, and it's breaking a lot of things. Yikes.
Don't worry guys Nvidia GPUs with 32GB VRAM are coming soon! And affordable
Not entirely sure what that has to do with XBOX performance issues, or VRAM in general. Care to expound?Don't worry guys Nvidia GPUs with 32GB VRAM are coming soon! And affordable
Looks like the issue is resolved, and with the amount of time between hotfixes, it seems like something was pretty easy to resolve.
Flappy Pannus
Veteran
Ok, had to check it out myself for brief tech impressions on a midrange system - i5-12400f, 3060 12GB, 32GB DDR4 3200.
First off, I'm wondering if the Zen2 performance bottleneck extends beyond the gameplay. On my system from a clean cache, the entire PSO compilation process took just under 5 minutes. They're both 6 core, 12 thread CPU's, and sure the 12400F is indeed a fair bit zippier than the 3600, but the 3600 in DF's video took 10 minutes! I certainly haven't seen a 2X performance bump for a 12400f vs 3600 reflected in any other benchmark.
Also while I can't recall if it was mentioned in DF's review (it's probably assumed as most Sony ports thankfully operate this way) but just to be clear yes, the shader optimization notification is just that - a notifier, not a blocker. You can jump right into the game and it will compile in the background on low priority threads. In fact on my first run before I wiped my shader cache to time it in full, I said "fuck it" and started the game at 20% shader compilation. I then went through about 3 minutes of the opening sequence, nary a blip. Now this doesn't mean if you're halfway through the game, get a driver update and jump right back into it with no waiting you won't get shader stuttering as perhaps the process is geared towards compiling assets for the starting area first, but at least in my brief test in the opening sequence it was not a bottleneck at all in maintaining 60fps - hell I'm barely over 50% CPU.
I didn't take much time tweaking the settings before getting the framerates of the opening sequence chase seen (after a fully compiled cache this time, but again it seemed to make little difference), just to see if I could get 4k 60 with DLSS Performance, so was shooting for some comfortable headroom at the opening scene. These are the settings I went with:
(BTW of note: A quirk was discovered though (I've seen this with a number of DX12 games) - changing the graphics settings may degrade your performance until you restart the game. I was futzing around with high/ultra/medium, and putting everything back to those above after that, I couldn't get above 90% GPU utilization and performance was just over 50fps. Reloading the game fixed it, so be aware.)
(Also I really like the settings menu showing PSO compilation/VRAM status in the corner)
With those settings, during the entire opening chase sequence of 5+ minutes, maybe had 3 little 'blips' - not stutter, just moments where the frametime was ever so briefly interrupted that RT's default polling frequency couldn't even catch it as a frame drop, I just saw a slight tick on the graph. GPU% was usually around 80-85%, the highest it got was getting near the fur of the wolves, at one brief camera pan I did get a drop to 58fps. Otherwise seemingly plenty of headroom for my settings and 4k DLSS/medium/high, albeit again this is just the opening sequence. So at least with this early look, this doesn't seem to be a title that's exorbitantly GPU demanding, or at least it can scale well. I certainly can't get 4k DLSS Perf @ 60fps with Horizon Forbidden West or Ghost of Tsushima without going to the absolute lowest settings, and even that's not enough for Forbidden West.
A familiar youtube comment on DF's video seems to be "lol he says it's a great port but then spends 20 minutes pointing out the flaws". I mean granted, youtube so what do you expect, but from my experience the point of Alex's review was solid - the fundamentals of this title seem to be quite strong, and in the case of shader stutter, may be even superlative. There are some definite quirks but they do seem to be that - specific areas with specific weird stuff that definitely needs to be brough to attention, but this is not a case where the foundation of the engine's PC version seems built on shaky ground as it can be with other ports.
(Oh btw, all of these tests were run with the game installed to a 5400 rpm HDD. )
First off, I'm wondering if the Zen2 performance bottleneck extends beyond the gameplay. On my system from a clean cache, the entire PSO compilation process took just under 5 minutes. They're both 6 core, 12 thread CPU's, and sure the 12400F is indeed a fair bit zippier than the 3600, but the 3600 in DF's video took 10 minutes! I certainly haven't seen a 2X performance bump for a 12400f vs 3600 reflected in any other benchmark.
Also while I can't recall if it was mentioned in DF's review (it's probably assumed as most Sony ports thankfully operate this way) but just to be clear yes, the shader optimization notification is just that - a notifier, not a blocker. You can jump right into the game and it will compile in the background on low priority threads. In fact on my first run before I wiped my shader cache to time it in full, I said "fuck it" and started the game at 20% shader compilation. I then went through about 3 minutes of the opening sequence, nary a blip. Now this doesn't mean if you're halfway through the game, get a driver update and jump right back into it with no waiting you won't get shader stuttering as perhaps the process is geared towards compiling assets for the starting area first, but at least in my brief test in the opening sequence it was not a bottleneck at all in maintaining 60fps - hell I'm barely over 50% CPU.
I didn't take much time tweaking the settings before getting the framerates of the opening sequence chase seen (after a fully compiled cache this time, but again it seemed to make little difference), just to see if I could get 4k 60 with DLSS Performance, so was shooting for some comfortable headroom at the opening scene. These are the settings I went with:
(BTW of note: A quirk was discovered though (I've seen this with a number of DX12 games) - changing the graphics settings may degrade your performance until you restart the game. I was futzing around with high/ultra/medium, and putting everything back to those above after that, I couldn't get above 90% GPU utilization and performance was just over 50fps. Reloading the game fixed it, so be aware.)
(Also I really like the settings menu showing PSO compilation/VRAM status in the corner)
With those settings, during the entire opening chase sequence of 5+ minutes, maybe had 3 little 'blips' - not stutter, just moments where the frametime was ever so briefly interrupted that RT's default polling frequency couldn't even catch it as a frame drop, I just saw a slight tick on the graph. GPU% was usually around 80-85%, the highest it got was getting near the fur of the wolves, at one brief camera pan I did get a drop to 58fps. Otherwise seemingly plenty of headroom for my settings and 4k DLSS/medium/high, albeit again this is just the opening sequence. So at least with this early look, this doesn't seem to be a title that's exorbitantly GPU demanding, or at least it can scale well. I certainly can't get 4k DLSS Perf @ 60fps with Horizon Forbidden West or Ghost of Tsushima without going to the absolute lowest settings, and even that's not enough for Forbidden West.
A familiar youtube comment on DF's video seems to be "lol he says it's a great port but then spends 20 minutes pointing out the flaws". I mean granted, youtube so what do you expect, but from my experience the point of Alex's review was solid - the fundamentals of this title seem to be quite strong, and in the case of shader stutter, may be even superlative. There are some definite quirks but they do seem to be that - specific areas with specific weird stuff that definitely needs to be brough to attention, but this is not a case where the foundation of the engine's PC version seems built on shaky ground as it can be with other ports.
(Oh btw, all of these tests were run with the game installed to a 5400 rpm HDD.
)
Last edited:
Cappuccino
Newcomer
I feel Metro Exodus EE looks better than like 90% of new releases and runs with RT enabled on consoles at 60 fps. 4A isn’t exactly a huge studio, why were they able to manage this but nobody else can?
snc
Veteran
On ps5 ? Nah, blurry and underwhelming
Charlietus
Regular
Similar threads
