50% gains in Cyberpunk 2077 at 4K with DLSS3, FI & Co

[J7SC_Orion]

source

 

 

 

Edited February 5, 2023 by J7SC_Orion

[J7SC_Orion]

After experiencing DLSS3, Frame Insertion and NVidia Reflex on my RTX 4090 in MSFS 2020 as a flawless suite of tools that rejuvenated the 5950X setup and with 4K ultra settings hitting the limit of 120 fps on the 4K120 OLED in Flight Simulator, my expectations were already raised re. what the long-promised DLSS3/FI/NVRe update would do for Cyberpunk 2077...answer:

 

 - a heck of a lot. With virtually no stuttering or visual oddness, the same CPU/mobo/4090 oc setup went from an already decent 90 fps in the CP2077 internal benchmark (DLSS2, 'Quality') to over 130 fps (min fps are even more impressive) in DLSS3, 'Quality'. I didn't even oc to what the card is capable of because with the C1 OLED at max 120, it makes no difference...and I didn't think that I would be able to exceed the 4K120 level so soon. 

 

 

 

FYI, the same system with the same settings managed a respectable 45 - 50 fps (DLSS2, 'Quality') with an overclocked 3090 Strix at close to 2200 MHz.  That card is now hooked up to an older IPS HDR panel at the other end of our place and that is a happy combo - be it for work  or play - in its own right. For certain work tasks, the 3090 is paired with 2x 2080 Ti (all water-cooled)

 

Still, I recall the early days of DLSS 1 - leaving much to be desired on the visuals. DLSS 3 / Frame Insertion / NV Reflex really are game changers, IMO, in the visual fidelity department as well. 

 

The irony is that the RTX4K series, especially the 4090, is probably the card that least required such boosts though apparently, there is also some extra hardware trickery involved on Ada Lovelace compared to Ampere.  

 

 

 

 

[Andrew]

Too bad DLSS3 is only available on RTX 40xx cards.

[UltraMega]

DLSS3 is an impressive technology, one that will no doubt be emulated by other vendors in the coming years. I think the price premium for these Nvidia only things is starting to break down a bit. The 7900XT and XTX are more price compelling even with RT considerations than anything Nvidia is currently offering. I'm not sure how much longer they can float on these upscaling techs. 

[schuck6566]

2 minutes ago, UltraMega said:

DLSS3 is an impressive technology, one that will no doubt be emulated by other vendors in the coming years. I think the price premium for these Nvidia only things is starting to break down a bit. The 7900XT and XTX are more price compelling even with RT considerations than anything Nvidia is currently offering. I'm not sure how much longer they can float on these upscaling techs. 

Depends if they're willing to pay game companies to throw that "Nvidia, the way it's meant to be played" logo on the games like they use to. Just pay them to use DLSS primarily and others as later add in options and guess who keeps market share?

[UltraMega]

1 minute ago, schuck6566 said:

Depends if they're willing to pay game companies to throw that "Nvidia, the way it's meant to be played" logo on the games like they use to. Just pay them to use DLSS primarily and others as later add in options and guess who keeps market share?

In an age where both GPU makers are "under shipping", it's clear they don't care about market share. 

[schuck6566]

2 minutes ago, UltraMega said:

In an age where both GPU makers are "under shipping", it's clear they don't care about market share. 

ROTFL true, but when fewer cards are selling, even AT higher prices, you want to make sure yours are the ones people are buying.

[J7SC_Orion]

8 hours ago, schuck6566 said:

ROTFL true, but when fewer cards are selling, even AT higher prices, you want to make sure yours are the ones people are buying.

 

Point well taken; volume at highest margin is what an economist would recommend to maximize profits (not revenue). In any case, as important as we might think we are as gamers and 'fan-boys-and-girls', the real money for the big GPU manufacturers is in the workstation and enterprise markets. With the huge growth of AI, (primarily) Nvidia has been sitting pretty. It is also known to ride roughshod over some of its board partners by keeping them on a short leash re. designs and cost-plus pricing (see 'EVGA', compared to Palit - the latter owning all kinds of brands including Galax - and is the largest GPU vendor).

 

AMD and Intel are desperately trying to catch up to NVidia in the lucrative workstation and enterprise markets with their latest multi-tile offerings. Still, for now, NVidia rules with the Ada Lovelace and Hopper designs, and well-developed related software with years of a head-start. Some permutations and combinations from those will hit the 'high-end' consumer market...see other threads about a potential 4090 Ti/Titan with 48 GB of VRAM; if/when it hits, it will be priced between the 4090 and RTX 6000 Ada...from that perspective, the US$ 1,619 I paid for my 4090 is outright '''cheap'''. Put differently, those are not upgrades of GPU chips from the gamer market, but downgrades from the workstation and enterprise markets...

 

Who knows what the future will bring ? Multi-tile HD GPUs with tiles disabled for consumer models ? How about paying a subscription fee to unlock some tiles on the GPU you already own ? It boggles the mind...

 

So DLSS3 for RTX4K was easier for NVidia with its massive AI prowess compared to their competitors. All I can say is that the 4090 with its 76.3 Billion transistors is a monster when compared to the 3090 at 28.3 billion and a 2080 Ti at 18.6 billion.  DLSS3 / FrameIns/ NVReflex is hugely impressive from my vantage point re. not only speed in fps and frame-times, but also visual quality...I use a C1 48 OLED 4K120 for comparison, and at that size and resolution, imperfections would show more dramatically.

 

A final point concerns the sheer gain in efficiency. Below are screenies from a workstation bench program called Octanebench, and while the oc'ed 3090+2x2080 Ti setups just nudge ahead of a single 4090, I could try to oc the 4090 a bit more - but they are certainly close. HOWEVER, the 4090 used around 580W for that while the 3xGPU combo was at about 1,260W...at the end of the day, DLSS3/FI/NVR is an incredible technology; I just wish it would also be available on my 3090 Strix, at least 'partially, given some additional hardware requirements with DLSS3. At least we can hope that some of it is used to improve DLSS2 some more.