Ratchet & Clank technical analysis *spawn

Remij said:
Ok, so there's a good stress test I've found which is in the Bronze cup battle arena.. it's basically a speedle ride through multiple rifts.. very similar to the canned portal sequence.. but this time you're in control. The PS5 undoubtedly provides a much smoother experience through this challenge than my PC does. That goes without question... BUT I'm not sure if it is because of an issue with the port itself (maybe this challenge didn't get as much QA as the other stuff?) or if perhaps it's my middling 3900X CPU (which I will be replacing soon by switching to the AM5 platform and a Ryzen 7800X3D).. so perhaps someone with a higher end CPU could test it out and record it and post it? I'd like to see if any PC out there can actually run this section more smoothly.

Here it is:
Click to expand...

Not a high-end CPU of course, but my 12400f gets very similar delays, even if it's run a second time and therefore should be cached in my 32GB. It doesn't seem to be an issue of loading speed, at least from storage media.


Edit: Tried it with Low textures - no difference (!).

 
Last edited:
Flappy Pannus said:
Not a high-end CPU of course, but my 12400f gets very similar delays, even if it's run a second time and therefore should be cached in my 32GB. It doesn't seem to be an issue of loading speed, at least from storage media.
Click to expand...
Yea, I was pretty certain it wasn't anything bandwidth related. Seems like that specific part could have used some more attention. Maybe I'll make a ticket and post my vid and see what they say.
 
I haven't actually played the game yet past the opening level so can't test that, but agree it won't be a CPU limit. It seems to me there is some software/API side issue that is causing stalls in the process. I do believe there is a hard limit placed within the game for how fast these sequences can complete (possibly tied to completing a full animation transition) but there is also clearly a system level issue causing the stalls as well and in my experience they happen whether you are using Direct Storage or not.

It would be interesting to see how the sequence performs on a fast UMA based PC to see if it has anything to do with swapping memory between the pools.

Also @Remij have you found frame generation to work as expected in this game? I was playing around with it last night and noticed basically no frame rate increase when turning it on. In the opening area of the second level (just after the first big portal transition sequence) I'm getting around 80fps maxed out with DLSS Q at 3840x1600, and then when I turn FG on it gets maybe 1 or 2 fps higher after a short period of settling down. I can switch it on and off at will and after leaving for a short period (fps will drop to around 60 ish for 10 second or so after a change), it always settles around 80fps regardless of whether FG is on or off.
 
pjbliverpool said:
@Remij have you found frame generation to work as expected in this game? I was playing around with it last night and noticed basically no frame rate increase when turning it on. In the opening area of the second level (just after the first big portal transition sequence) I'm getting around 80fps maxed out with DLSS Q at 3840x1600, and then when I turn FG on it gets maybe 1 or 2 fps higher after a short period of settling down. I can switch it on and off at will and after leaving for a short period (fps will drop to around 60 ish for 10 second or so after a change), it always settles around 80fps regardless of whether FG is on or off.
Click to expand...
That's not right as I get a good 40%+ from FG.
 
davis.anthony said:
That's not right as I get a good 40%+ from FG.
Click to expand...

Maybe I should try a fresh/clean driver install. I can't say I tested this particular aspect that closely prior to the hotfix driver but I also don't recall noticing anything out of the ordinary with regards to FG behavior back then. Also couldn't find anything on Google so it's probably a local issue.
 
So I tried a fresh clean driver install and hey presto, I immediately saw a healthy frame rate boost after starting the game and turning FG back on. Problem solved!

Or not.

Soon after I started to see the same issue of performance varying wildly again, sometimes much lower than it should be regardless of setting, sometimes performing as it should without FG and getting no gain from turning it on (all in the exact same scene/fixed view). Eventually I noticed a link to power draw. I can be hitting 99% GPU usage one moment with <200w power draw, and then the next I can be drawing 220w+ with 20% more frames in exactly the same scene. It's completely bizarre. I even managed to capture it on a video at one point. Simply loading the map screen and then going back into gameplay tanked both my frame rate and my power draw.

Also I can reproduce these strange results immediately after a game, or even full system restart so it's not down to toggling settings in game. I'm a bit baffled at the moment. When it's working as it should it flys, but more often than not it's massively underperforming (which shows in the power draw despite high GPU usage) and I've no idea why.
 
I'd upload the videos as I've captured it twice now. It's random though so tough to capture efficiently which means the vids would.need editing first. Also my HDR makes the videos look horrible and W11 tells me I need to buy a new codec to play them back even though they were just recorded in GF Experience. So quite a bit of work to actually get them online.
 

Ratchet and Clank Rift Apart | PC CAN Load The Portal Sequence Faster Than The PS5 Afterall​



From the YT description:

Compusemble said:
It turns out that the reason for the slower portal sequences completion on PC compared to the PS5 was due to the presence of a file in the game folder. If you remove the cache.pso file, the loading times improve. Without this file, PC takes the lead.

Keep in mind that first you have to run the game with cache.pso file in the game folder, and then you will have compiled shaders in AppData\Local\NVIDIA\DXCache.

After doing so, you can delete the cache.pso file, and you should not be seeing any shader compilation stutter.

Thanks to @moeezmalik3598 for bringing this to my attention.
Click to expand...

My guess is the cache.pso file are the gathered pipelines for on-the-fly shader compilation, but even after they are compiled by the driver, the cache.pso file is still being referenced to ensure the cache is valid and this can result in a very slight performance penalty (...?).

I'll test later with my 12400/3060 system to see if it has a larger effect on a more budget CPU.
 
Last edited:
Ok, here's my cache.pso test. It's not as perfectly synced up at the start as I would like, but they're not out of sync enough where it's the reason for large disparity.


BTW, no video but I did also test it with the bugsnax race trial, unfortunately it has little effect on that - the bottleneck is somewhere else with that, at least for 2 of the transitions which have ~1 sec pauses.
 
Flappy Pannus said:
Ok, here's my cache.pso test. It's not as perfectly synced up at the start as I would like, but they're not out of sync enough where it's the reason for large disparity.


BTW, no video but I did also test it with the bugsnax race trial, unfortunately it has little effect on that - the bottleneck is somewhere else with that, at least for 2 of the transitions which have ~1 sec pauses.
Click to expand...

I sent them a ticket about that issue specifically. Got the typical "We've passed your feedback onto the team, thanks!" response.. but they've done a good job so far in addressing player feedback. Hopefully they can fix it up!
 
Flappy Pannus said:
Ok, here's my cache.pso test. It's not as perfectly synced up at the start as I would like, but they're not out of sync enough where it's the reason for large disparity.


BTW, no video but I did also test it with the bugsnax race trial, unfortunately it has little effect on that - the bottleneck is somewhere else with that, at least for 2 of the transitions which have ~1 sec pauses.
Click to expand...
Wow. That really adds up. Great editing. Appreciate the effort here
 
Flappy Pannus said:

Ratchet and Clank Rift Apart | PC CAN Load The Portal Sequence Faster Than The PS5 Afterall​



From the YT description:



My guess is the cache.pso file are the gathered pipelines for on-the-fly shader compilation, but even after they are compiled by the driver, the cache.pso file is still being referenced to ensure the cache is valid and this can result in a very slight performance penalty (...?).

I'll test later with my 12400/3060 system to see if it has a larger effect on a more budget CPU.
Click to expand...
Nice find. Yea, makes sense and I agree with your hypothesis. Compusemble said hopefully they'll be able to fix it.. and the way to fix it would be to pre-compile the shaders upfront beforehand. Perhaps they could compile just the shaders required for those sequences upon initial load (would be quite short I imagine), and then have the game ignore referencing the cache.pso file during them specifically.

It's not really the end of the world or anything, but compiling on the fly WILL have a cost when it comes to getting load times down as low as possible.

In this game specifically, unless you have a fast setup, the transitions can be a little janky.. I would imagine a game build for both PS5 and PC from the start would more gracefully handle any slight differences between the two.

Nixxes does an amazing job though, and ideally we want games to be able to compile on the fly to not have long compilation processes at the beginning of a game. The best solution though is likely a combination of a small upfront compile, and on the fly. Again though.. this is why I advocate to have an option in the menu to just pre-compile essentially everything upfront.
 
You must log in or register to reply here.

Similar threads

D
Is PS4/XONE CPU a bottleneck?
2 3 4
Replies
68
Views
22K
Nisaaru
N
A
Holy Wall of Text: My impressions post Playstation 4 (PS4) unveil
2 3 4
Replies
70
Views
20K
(((interference)))
I
A
Summed-Area Variance Shadow Maps (Demo)
2 3 4
Replies
64
Views
53K
Andrew Lauritzen
A
D
Complete Details on Xenos from E3 private showing!
6 7 8
Replies
149
Views
27K
nAo
N
F
  • Locked
The PLAYSTATIOn3 and the game industry !!! Part1...
Replies
4
Views
4K
overclocked
O
Back
Top