Neon Noir benchmarks

Quartz · 13 Nov 2019 at 18:04

The Neon Noir benchmark is now available from Crytek.

I'm sure @Kaapstad will be along shortly with a proper thread but I thoght I'd kick off with my initial results:

i7-8700, RTX 2080 Ti stock, 32 GB RAM.

1080p / full screen / ultra: 12594 and 120-150 fps
2160p / full screen / ultra: 4176 and 37-59 fps, mostly 45-50.

I had stuff running in the background and didn't look at CPU or GPU usage.

willhub · 13 Nov 2019 at 18:07

Is this utilising tensor cores? I notice how it supports non RTX cards which potentially indicated it's not?

Quartz · 13 Nov 2019 at 18:11

willhub said:
Is this utilising tensor cores?

I don't know and really, I don't care. All I care about is the end result.

willhub · 13 Nov 2019 at 18:13

Quartz said:
I don't know and really, I don't care. All I care about is the end result.

Well you should care then, the end result is scuppered due to this, if tensor cores were utilised then actually the performance would be just about ok as opposed to your sub par 2160P.

https://wccftech.com/neon-noir-crye...ic-ray-tracing-demo-tested-and-available-now/

Quartz · 13 Nov 2019 at 18:20

willhub said:
as opposed to your sub par 2160P.

That article doesn't list 2160p scores, only 1080p and 1440p. And my 1080p frame rates are bang on.

willhub · 13 Nov 2019 at 18:22

Shame it's not running tensor cores as mentioned, the RTX cards would be much ahead.

Quartz · 13 Nov 2019 at 18:46

willhub said:
Shame it's not running tensor cores as mentioned, the RTX cards would be much ahead.

Umm.. we don't know if it is or isn't.

willhub · 13 Nov 2019 at 18:51

Umm... "We used the tool that we were given to benchmark the game engine performance across a variety of graphics cards from the Radeon RX Vega 56 to the GeForce RTX 2080Ti to see how it all performed in an engine that is supporting Real-Time Ray Tracing features but not using specific hardware acceleration"

Quartz · 13 Nov 2019 at 18:58

I took that to mean that it doesn't care about the presence or otherwise of hardware acceleration. The use or otherwise is at a lower level. You'll note that the top-performing cards are indeed RTX cards.

willhub · 13 Nov 2019 at 19:00

If we don't know if the RTX would be ahead with the hardware, lets just think, the other cards don't have the dedicated hardware, the RTX have unused hardware, so it's pretty clear that the performance would increase if the engine was programmed to use the extra hardware.

Dicehunter · 13 Nov 2019 at 19:15

Quartz said:
Umm.. we don't know if it is or isn't.

Actually we do know, The 5 minute dev diary says it is not using RTX and is completely hardware agnostic.

Rroff · 13 Nov 2019 at 19:24

No RT cores or Tensor hardware is used for this - it uses a variety of techniques to do ray traced reflections and some GI features largely by creating a simplified representation of the geometry using Voxel techniques and as such isn't comparable to the feature set of a "proper" ray tracing solution - if it was doing the full feature set it would have another big drop in performance again similar to trying to run RTX features on a Pascal card.

arc@css · 13 Nov 2019 at 19:28

3100 was my score (including overclocking

).

Poneros · 13 Nov 2019 at 19:54

According to the devs, they will implement usage of RT cores as well (eventually) and they believe it will significantly speed up performance. This is from one of the early interviews on it (where they talked about resolutions), don't have a link saved.

Dicehunter · 13 Nov 2019 at 20:22

Benched on a friends rig using a 9900K and 2080 Ti at 1440P max preset I got 8400, FPS was damn smooth too.

Quartz · 13 Nov 2019 at 20:43

Dicehunter said:
Actually we do know, The 5 minute dev diary says it is not using RTX and is completely hardware agnostic

THen I stand corrected.

Rroff · 13 Nov 2019 at 21:11

Poneros said:
According to the devs, they will implement usage of RT cores as well (eventually) and they believe it will significantly speed up performance. This is from one of the early interviews on it (where they talked about resolutions), don't have a link saved.

The approach isn't an exact match for how structure wise nVidia have implemented the hardware in Turing but will be interesting to see results - in theory it should significantly reduce the time taken to process some components of this.

arc@css · 13 Nov 2019 at 21:27

arc@css said:
3100 was my score (including overclocking ).

Edit: now 3482 (remembered to max the case fans out and add a bit more of an overclock). Can't see me getting any better than that though.

shankly1985 · 13 Nov 2019 at 22:47

bru · 13 Nov 2019 at 23:00

ok colour me surprised
1080p ultra score 3192

Xenon 5650 @4.4GHz
MSi 970 gaming 4 @1458 MHz

And I must say it ran sweet as a nut and looked superb, with FPS between 30-40 with the lowest I noticed of 26 during the bullet section.