Support NeoGAF

[MikeM]

First benchmark of the 5090 using CP2077:



Seems to up latency from about 35ms to about 45ms going from DLSS only to DLSS + 4x FG max settings including pathtracing. Not bad…

Tested at 4k:
No DLSS: ~28fps, ~65ms latency
DLSS Quality: ~66fps, ~33ms latency
DLSS frame gen 3x: ~168 fps, ~46ms latency
DLSS frame gen 4x: ~209fps, ~48ms latency

Nvidia did work on frametime consistency with FG. No graphs provided though.

 

[HeisenbergFX4]

 

[diffusionx]

waiting for a summary

 

[reinking]

Most complaints will be by people that will not be buying one.

To get this party started….

 

[Guardian Monkey]

Is that ~28 fps with nothing added? No frame gen/ dlss.

 

[Guardian Monkey]

 

[Puscifer]

The thing that caught my ear was him mentioning Nvidia saying that at 4K DLSS 4.0 Performance is more like current DLSS balanced in terms of quality. Amazing stuff if true!

 

[TintoConCasera]

The thing that caught my ear was him saying Nvidia saying that at 4K DLSS performance is more like DLSS balanced in terms of quality. Amazing shit!

So, better image reconstruction even from lower resolutions? Very nice, should be a nice push for the most demanding games.

 

[Puscifer]

Full path tracing, though right? How's that compare to a 4090?

 

[Nankatsu]

That latency. Welcome to Console gaming PC gamers.

 

[TintoConCasera]

No DLSS: ~28fps, ~65ms latency
DLSS Quality: ~66fps, ~33ms latency

Not bad latency considering the base is just 28fps.

More base fps should mean lower latency with FG, right?

 

[DenchDeckard]

Is that ~28 fps with nothing added? No frame gen/ dlss.

Full path tracing and 0 DLSS at native 4K....

 

[MikeM]

Full path tracing, though right? How's that compare to a 4090?

That shot is just from switching settings. It settled down to about 68ms.

 

[3liteDragon]

So they did manage to reduce latency with the new FG model

 

[Nvzman]

Full path tracing, though right? How's that compare to a 4090?

Granted Phantom Liberty is more taxing than the base game, so my guess is the 4090 averages something closer to 23-25fps in the area that the guy was showing off.

 

[analog_future]

Granted Phantom Liberty is more taxing than the base game, so my guess is the 4090 averages something closer to 23-25fps in the area that the guy was showing off.

Looks like we're consistently seeing ~30-40% gains from the 4090 to the 5090 in pure hardware.

 

[HeisenbergFX4]

Why didn't you take the screenshot when latency settled down to around 70?

While still not great its what it really was not at over 100 right after he changed settings

 

[Buggy Loop]

That latency. Welcome to Console gaming PC gamers.

Lol no

With reflex that latency even for MFG 4x is still a football field lower than console latencies.

 

[TintoConCasera]

Why didn't you take the screenshot when latency settled down to around 70?

That wouldn't be near as inflammatory ofc.

 

[Sept Terrion Of Nothing]

Look, I agree that no Nvidia card should be 2k dollars. No excuse, end of story.

But "native" is so unnecessary right now. If they achieve higher FPS with AI, DLSS, or FG without compromising the image quality I'm all for it. I've seen some bad examples of FG and DLSS. But I've seen so much good examples as well. When I play with these good implementations I really enjoy the game.

That's the most important thing: Enjoying the game with good executed technologies.

 

[Magic Carpet]

Granted Phantom Liberty is more taxing than the base game, so my guess is the 4090 averages something closer to 23-25fps in the area that the guy was showing off.

My 4070 all the way down in the single digits. I have to put CP77 in 1080P mode to make it playable with Path Tracing and DLSS, NO Frame Gen.

 

[Lucifers Beard]

Raster performance is kind of what I expected and the uplift from 4090 looks about right I guess.

Although I'm sure some will hate it, gaming is clearly going in the direction of using A.I to help smooth out frame rates and provide as close to native image quality as possible with upscaling. I'm fine with it to be honest, frames are frames and I couldn't give a shit if they're 'fake frames'.

If this allows developers to focus more of their time on creating interesting games, it's worth it to me.

I'm looking forward to seeing more on the 5080 and 5070ti since these are cards are likely what the majority of people will be looking to upgrade to.

 

[ThisIsMyDog]

With Reflex 2 this latency can still be cut in half am i right? Or maybe this Cyberpunk build has it already.

 

[Lethal01]

Why didn't you take the screenshot when latency settled down to around 70?

While still not great its what it really was not at over 100 right after he changed settings

think he was just showing the fps since this is with no FG or DLSS

 

[MikeM]

Raster performance is kind of what I expected and the uplift from 4090 looks about right I guess.

Although I'm sure some will hate it, gaming is clearly going in the direction of using A.I to help smooth out frame rates and provide as close to native image quality as possible with upscaling. I'm fine with it to be honest, frames are frames and I couldn't give a shit if they're 'fake frames'.

If this allows developers to focus more of their time on creating interesting games, it's worth it to me.

I'm looking forward to seeing more on the 5080 and 5070ti since these are cards are likely what the majority of people will be looking to upgrade to.

Im curious to know if less TOPS means more latency versus DLSS only. I know it will mean less frames but I want to know if TOPS loss means higher latency penalties.

 

[Guardian Monkey]

Why didn't you take the screenshot when latency settled down to around 70?

While still not great its what it really was not at over 100 right after he changed settings

I didn't mention anything about latency.

 

[DirtInUrEye]

The thing that caught my ear was him saying Nvidia saying that at 4K DLSS performance is more like DLSS balanced in terms of quality. Amazing shit!

And now I'm really keen to see what ultra performance looks like.

 

[Mithos]

Anyone using Frameview a lot might be able to tell me, how do you get the statistics against a grey/black background like in this video mine is always transparent, so in some game its hard to see.

 

[BennyBlanco]

I genuinely don't understand how they can pull off the latency at 100 fps when the frame gen is starting at 28 fps. Fake inputs? I just don't get it. But the current frame gen has worked just fine IMO.

 

[Guardian Monkey]

Look, I agree that no Nvidia card should be 2k dollars. No excuse, end of story.

But "native" is so unnecessary right now. If they achieve higher FPS with AI, DLSS, or FG without compromising the image quality I'm all for it. I've seen some bad examples of FG and DLSS. But I've seen so much good examples as well. When I play with these good implementations I really enjoy the game.

That's the most important thing: Enjoying the game with good executed technologies.

I don't have any issues with DLSS and sometimes frame doubling is a bit wonky on the 40series, but the new transformer model seems to be much improved but seems to be the way forward. But naive numbers are still important just as a comparison as a baseline of image quality for comparison performance of prior cards. 4K at full path tracing near 30fps is impressive on its own compared to where we were just a few years ago. Just noting an observation.

 

[moogman]

That's pretty decent although I'd like to see how well that scales down the 50 series. Also, after watching most of the video with path tracing on I was a bit shocked how bad it looks in comparison when he turned RT off.

 

[Bojji]

I genuinely don't understand how they can pull off the latency at 100 fps when the frame gen is starting at 28 fps. Fake inputs? I just don't get it. But the current frame gen has worked just fine IMO.

It's not starting FG from that 28fps. First DLSS SR brings it around ~60 - THEN frame gen can work correctly.

I didn't watch the video but NV always included DLSS SR in their comparisons vs. "DLSS off".

 

[KyoZz]

No DLSS: ~28fps, ~65ms latency
DLSS Quality: ~66fps, ~33ms latency
DLSS frame gen 3x: ~168 fps, ~46ms latency
DLSS frame gen 4x: ~209fps, ~48ms latency

 

[The Cockatrice]

Offscreen trash again with nvidia in the room hiding again. Il wait for the real benchmarks.

 

[//DEVIL//]

looks like I am buying a 5090... I think..

I just dont feel like spending that much lol. I still feel like if I can get a 4090 used for half of that price or a bit more then its worth it ? but the AI thing.... I am 100% sure the 5090 will be compatible with whatever 6090 will offer in terms of AI for having that hardware needed while the 4090 will struggle.

sigh..

 

[hinch7]

That latency. Welcome to Console gaming PC gamers.

Yeah, no. Consoles have significantly more latency. Taken from DF -

And Reflex 2 reduces latency even further from the above. Actually impressed with those figures considering its FGx4.

 

[darrylgorn]

I genuinely don't understand how they can pull off the latency at 100 fps when the frame gen is starting at 28 fps. Fake inputs? I just don't get it. But the current frame gen has worked just fine IMO.

My guess is that these FG benchmarks are including the other AI scaling options with DLSS 4. If that's the case, there could be lower latency due to parts of image being downscaled to lower resolutions.

 

[Boss Mog]

That's the true performance of this card, the rest is smoke and mirrors.

 

[Zathalus]

That's the true performance of this card, the rest is smoke and mirrors.

FG can qualify as smoke and mirrors, but regular DLSS quality sure doesn't. And it clears over 60fps with that on, which is quite a bit step up from the 4090.

 

[TintoConCasera]

That's the true performance of this card, the rest is smoke and mirrors.

Dude, it's a videogame. Everything is smoke and mirrors.

 

[Myths]

This shouldn't be surprising.

My guess is that these FG benchmarks are including the other AI scaling options with DLSS 4. If that's the case, there could be lower latency due to parts of image being downscaled to lower resolutions.

It would necessarily have to have a lower internal raster in the buffer compared to native, otherwise there would be no increase in more frame rate or frame time. That coupled with the integration of synthesized images results in increased frames.

 

[CompleteGlobalSaturation]

Seems like the level of artefacts increase as you go higher on the FG multiplier (as expected, and observed with the 'Lossless Scaling' app).

 

[Danny Dudekisser]

Those raw performance figures are really underwhelming. I didn't realize PL was that demanding of a game.

More than anything, this is discouraging from the standpoint of: frame gen and all of this other garbage that's destroying image quality is going to become the standard. It's going to get worse before it gets better again.

 

[ResilientBanana]

Full path tracing and 0 DLSS at native 4K....

This is an important point. Not all games use full path tracing at 4K but also, no other cards can get close to the number. So really, yes, the video card is worth the money unless there is another card that can beat it, and significantly so. People don't realize the taxing nature of running 4K natively, not only that but throwing Path Tracing on top of it you are never going to hit 30, let alone 60fps on any card. Meanwhile, without Path Tracing, the game runs ultra at 66fps which is rather significant given the quality of the assets. Ray Tracing in general diminishes all returns and is probably best used with DLSS. This is all very reasonable.

In order for consoles to achieve similar results they use medium to high assets, reduce the post processing effects, rely on scaling, and never truly hit native 4K. Same goes for lower spec video cards.

 

[flagdaddy]

With Reflex 2 this latency can still be cut in half am i right? Or maybe this Cyberpunk build has it already.

do we know yet if Reflex 2 frame warping works with DLSS MFG? or is it going to be 1 or the other

 

[Garibaldi]

I genuinely don't understand how they can pull off the latency at 100 fps when the frame gen is starting at 28 fps. Fake inputs? I just don't get it. But the current frame gen has worked just fine IMO.

They aren't. It's using dlss quality as a start. Which isn't 28fps. Native is 28fps.

 

[cinnamonandgravy]

Looks like we're consistently seeing ~30-40% gains from the 4090 to the 5090 in pure hardware.

dont think itll be that high when the big boy benchies come out
probably 20-30%

 

[Puscifer]

That's the true performance of this card, the rest is smoke and mirrors.

Harsh but true lol

To me, though, it would be acceptable if every have had this level of technical prowess but they don't. Screw the new Monster Hunter

 

[sendit]

Tested at 4k:
No DLSS: ~28fps, ~65ms latency
DLSS Quality: ~66fps, ~33ms latency
DLSS frame gen 3x: ~168 fps, ~46ms latency
DLSS frame gen 4x: ~209fps, ~48ms latency

This is what i wanted to see. This kicks the living balls off Lossless scaling. Granted Lossless scaling is only 10 dollars.

 

[cinnamonandgravy]

I still feel like if I can get a 4090 used for half of that price or a bit more then its worth it ?

i might just go 4090 too

i was friggin jazzed for the massive 512 bus, but everything else seems fairly lackluster.
FG seems dumb... looks like 200fps but feels like 40fps... ehhh. id probably rather just have 60.
really want a massive jump in RT performance.