-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Cyberpunk 2077 Ultra performance
- Thread starter g67575
- Start date
Soldato
This game is a mixture of ridiculously complicated and un optimised, no one is playing this how it should look until 2 gens time imoI get 45/60fps on my system on a mix of medium to high even with DLSS on performance and it sucks!!
The CPU/MEMORY certainly isn't the problem but I guess my 2070 Super is lacking in the game
Played at release with a decade old 3770k and a 3080. Had everything maxed out at 1440p. Looked fantastic.
It's worth noting that this isn't a twitch shooter, so playing ~60FPS is perfectly acceptable.
Soldato
No dlss? All raytracing? Full everything?
4k no dlss maxed out every setting is maxed out, which I don't think is playable for a while
Soldato
Its the new Crysis dont you know.
Of course I used DLSS, it looks better than native.
Most people realise 4k is still a silly resolution, hence the very low uptake.
Caporegime
I've got DLSS set to performance, everything as high as it'll go (not Psycho though) and HDR on, ray tracing on ultra, looks even better now for some reason, I can't tell the difference between DLSS Quality and Balanced on my 4k TV aside from frame rate.
Sadly they completely borked the lighting with RT as far as characters go, they damn near glow now. Gotta keep that off, so shadows + reflections only atm. Another classic case of pushing more accuracy in the rendering tech and forgetting all the other pre-requisites. Sigh.
See for ex: https://youtu.be/8O9FuU93j2I?t=700
See for ex: https://youtu.be/8O9FuU93j2I?t=700
Associate
- Joined
- 23 Oct 2019
- Posts
- 687
I know they redid the cars and people disappearing when you turn around. I'm not sure about the 2d versions of cars though, that's a pretty common optimisation technique I'd imagine but it is a bit aggressive and obvious in cyberpunk
Soldato
I disagree that dlss looks better than native, but politely agree that your opinion is valid
Also a lot of people that have the highest end cards are very much using 4k
You will also notice that the people who have the highest end cards are also very much in the minority. TVs are the 4k driving force now, where people sit with the lights on around 6-12' away.
Caporegime
Maybe now it's a good time to have a Cyberpunk 2077 benchmark thread considering they've added an actual benchmark to the game, it would be interesting to see how this game performs over multiple systems using the same benchmark rather than snippets from in game where framerate can vary wildly.
Soldato
I agree, that would be interesting
Caporegime
@Kaapstad You're good with those sort of benchy threads? I'd be tempted to set it up but I know I'd slack at ever updating it.
Decided to do some more CPU tests, because why not..
This time I did 70 seconds walks (35s to, 35s back to start), at Arasaka Tower where they have the mostly glass ramps, walking towards the three monks, done from 8am to 10am.
i7 6800K @ 4 Ghz, RX 6800, 16 GB RAM @ 3000 Mhz, 0.5 GB/s Sata SSD
The most interesting results are just how absolutely staggering the CPU impact of CD is (70% more performance with CD Low vs High!!!), even above maxing out ray tracing, and then when turning on RT and having CD High it's not as much of a loss anymore, but that might just be me running into a scaling problem with my current CPU (as well as GPU). This is pretty much as reduced of a GPU bottleneck as possible given that 720p FSR UP is about 240p
But, given that CDPR just announced a new expansion isn't coming this year, it gives me time to wait out Zen 4 + RDNA 3/ADA and hopefully at a reasonable price. Not much else to look forward to this year in terms of AAA, maybe Avatar but we'll see (Massive's games usually require a few months for patching before they're decent), plus I wouldn't be surprised at a delay.
This time I did 70 seconds walks (35s to, 35s back to start), at Arasaka Tower where they have the mostly glass ramps, walking towards the three monks, done from 8am to 10am.
i7 6800K @ 4 Ghz, RX 6800, 16 GB RAM @ 3000 Mhz, 0.5 GB/s Sata SSD
When a RT effect is in brackets it means it is used solo, so RT (Ref) means only ray traced reflections are used. RT Psycho or RT Med, means all RT effects are on but the RT Lighting is either at Psycho or Medium. CD = Crowd Density.
The most interesting results are just how absolutely staggering the CPU impact of CD is (70% more performance with CD Low vs High!!!), even above maxing out ray tracing, and then when turning on RT and having CD High it's not as much of a loss anymore, but that might just be me running into a scaling problem with my current CPU (as well as GPU). This is pretty much as reduced of a GPU bottleneck as possible given that 720p FSR UP is about 240p
But, given that CDPR just announced a new expansion isn't coming this year, it gives me time to wait out Zen 4 + RDNA 3/ADA and hopefully at a reasonable price. Not much else to look forward to this year in terms of AAA, maybe Avatar but we'll see (Massive's games usually require a few months for patching before they're decent), plus I wouldn't be surprised at a delay.
I get 45/60fps on my system on a mix of medium to high even with DLSS on performance and it sucks!!
The CPU/MEMORY certainly isn't the problem but I guess my 2070 Super is lacking in the game
I got 45 - capped 60 FPS using a 3080 at 1440p with everything maxed, including RT Psycho, and DLSS Quality on a 10year old 3770k system. Getting the itch to play again since I've upgraded.
More testing...
This time I've decided to isolate them by zones, so as to get a bit more granularity as to where the stress points are. Initially I got the idea because I noticed how heavily some sections were with RT, moreso than most of the game, but instead of focusing on just RT I figured I'd do more, but I did do a bit more in-depth testing for the various RT effects - some useful findings there in terms of what RT effects to turn on/off.
Test Location #1 Driving Test. From Masala Studios towards Med Center; Watson -> Little China.
Test Location #2 Raytracing Stress-Test Zone. Westbrook -> North Oak -> Kerry's Mansion.
Test Location #3 City Area Test. Behind Embers, from Senate & Market Fast Travel Point. Heywood -> The Glen.
Test Location #4 Side-Zone Test. Just Past NID Docks Fast Travel Point. Watson -> Northside.
Test Location #5 Lightest Zone Test. Eastern Badlands -> From Across The Sunshine Motel.
Results:
Here I did tests for raster vs raytracing, but also scaling across various FSR 2.1 modes (through the mod that latches unto DLSS; Q = quality etc.), as well as scaling for all the RT effects, combined as well as independently, at Kerry's Mansion (one of the hardest stress test areas for RT). The values you see in the table are in milliseconds, not frames per second. Raster settings were all ultra, except fog & clouds at medium.
From the last test (and some other data I have) we can conclude the following for RT:
Don't turn just one on. If you're already building the BVH structure then might as well stack them together. Granted, I should have added at least 2 more locations to properly isolate performance for f.ex. Local Shadows, but this is still one of the worst case scenarios. The following combinations are ideal:
This time I've decided to isolate them by zones, so as to get a bit more granularity as to where the stress points are. Initially I got the idea because I noticed how heavily some sections were with RT, moreso than most of the game, but instead of focusing on just RT I figured I'd do more, but I did do a bit more in-depth testing for the various RT effects - some useful findings there in terms of what RT effects to turn on/off.
Test Location #1 Driving Test. From Masala Studios towards Med Center; Watson -> Little China.
Test Location #2 Raytracing Stress-Test Zone. Westbrook -> North Oak -> Kerry's Mansion.
Test Location #3 City Area Test. Behind Embers, from Senate & Market Fast Travel Point. Heywood -> The Glen.
Test Location #4 Side-Zone Test. Just Past NID Docks Fast Travel Point. Watson -> Northside.
Test Location #5 Lightest Zone Test. Eastern Badlands -> From Across The Sunshine Motel.
Results:
Here I did tests for raster vs raytracing, but also scaling across various FSR 2.1 modes (through the mod that latches unto DLSS; Q = quality etc.), as well as scaling for all the RT effects, combined as well as independently, at Kerry's Mansion (one of the hardest stress test areas for RT). The values you see in the table are in milliseconds, not frames per second. Raster settings were all ultra, except fog & clouds at medium.
From the last test (and some other data I have) we can conclude the following for RT:
Don't turn just one on. If you're already building the BVH structure then might as well stack them together. Granted, I should have added at least 2 more locations to properly isolate performance for f.ex. Local Shadows, but this is still one of the worst case scenarios. The following combinations are ideal:
- All RT, with Lighting at medium, but no reflections
- Reflections + Sun Shadows
- If you use both reflections & lighting, then might as well turn Sun&Local Shadows on as well, the hit is low.
- No point in just using Sun Shadows (here there's also ini tricks you can do to boost just the raster shadow cascade ranges to mimic the RT effect).