Quantum Break тест GPU (Gamegpu.com)

[TheELF]

This hunch you speak of. What is it based on exactly?

Hm,maybe on exactly what he talks about in his post?
Aots improved a lot on nvidia (or vice versa) and aots is the goldielocks formula for dx12 games,nice distributable compute workload, so if this game can have good improvements for nvidia than any game can.

 

[PowerK]

Bought the game from Steam last night. I knew this game was a mess before I took the plunge.

Anyway, a brief initial impression after playing for two hours.
Upscaling (or temporal reconstruction or whatever they want to call it) : I personally dont understand people hating this so much. I think the game even with the this off looks really not good either and in terms of textures and aliasing and if you turn the other settings down it will look ugly. This game even with Ultra textures, does not show high detail anyway.

So far, mediocre game play with average level of graphics, I don't understand some people making a fuss about this game other than DX11 vs. DX12 fiasco.

 

[2is]

Hm,maybe on exactly what he talks about in his post?
Aots improved a lot on nvidia (or vice versa) and aots is the goldielocks formula for dx12 games,nice distributable compute workload, so if this game can have good improvements for nvidia than any game can.

Except AOTS doesn't perform better on nViida, so how does that example give him (or you) a hunch that it's simply a matter of optimization and time before NVidia had the advantage over AMD in DX12? And lets keep the discussion apples to apples... Current hardware and percentage of performance gained from DX11 to DX12.

 

[tviceman]

This hunch you speak of. What is it based on exactly?

Improvement over time in AMD's poster-child DX12 AoTS. With current benchmarks, Nvidia gains more performance when moving to the DX12 render path. It's now proven that Nvidia can gain 25+% when moving to DX12, but it also took this long to do so.

 

[Bacon1]

Improvement over time in AMD's poster-child DX12 AoTS. With current benchmarks, Nvidia gains more performance when moving to the DX12 render path. It's now proven that Nvidia can gain 25+% when moving to DX12, but it also took this long to do so.

Why don't any other sites show those gains and such poor AMD performance?

https://www.computerbase.de/2016-08.../#diagramm-ashes-of-the-singularity-2560-1440

Fury X is tied with OC'd 1070.

http://www.guru3d.com/articles_pages/zotac_geforce_gtx_1080_arcticstorm_review,16.html

Again Fury / Fury X are above 980 TI and close to 1070.

Does this mean AOTS is a valid game now that Nvidia is performing better?

 

[Unreal123]

Why don't any other sites show those gains and such poor AMD performance?

https://www.computerbase.de/2016-08.../#diagramm-ashes-of-the-singularity-2560-1440

Fury X is tied with OC'd 1070.

http://www.guru3d.com/articles_pages/zotac_geforce_gtx_1080_arcticstorm_review,16.html

Again Fury / Fury X are above 980 TI and close to 1070.

Does this mean AOTS is a valid game now that Nvidia is performing better?

It does not matter who is performing better. However, the thing only matter is that does anyone getting advantage from it, does developer gain anything from it, does game sales improve, does it perform better for majority of PC gamers. DX12 currently is a over head, extra resources, sales loss at the moment for developers. Any game related to DX12 is flopped badly in terms of reviews, quality port, and sales.

 

[dogen1]

Yes 4 previous frames that are rendered at 720p and then upscaled to 1080p.
Although I do agree with you,doesn't look that bad.
(other then the faces that are really badly lit,at lest at lower qualities where I play it at)

No, if it was upscaling it wouldn't be called reconstruction. What would the point of keeping 4 frames around be if you weren't gonna use any of that data and just upscale?

 

[Piroko]

No, if it was upscaling it wouldn't be called reconstruction. What would the point of keeping 4 frames around be if you weren't gonna use any of that data and just upscale?

Because without the higher resolution part in "rerconstructing into a higher resolution canvas" you would still end up with a very soft 720p image. You're free to call it temporal reconstruction, but at the end of the day it is used as an elaborate method of upscaling.

Some example clips:
https://www.youtube.com/watch?v=BSg3opLozkw

 

[antihelten]

You're free to call it temporal reconstruction, but at the end of the day it is used as an elaborate method of upscaling.

It might perhaps be slightly more accurate to say that it is an elaborate method which amongst other things* also includes an upscaling step.

*I imagine that their choice of a temporal solution may also have been guided by a desire to get rid of temporal aliasing.

 

[Piroko]

It might perhaps be slightly more accurate to say that it is an elaborate method which amongst other things* also includes an upscaling step.

*I imagine that their choice of a temporal solution may also have been guided by a desire to get rid of temporal aliasing.

Yeah but it doesn't do that all that well. As soon as you have noisier textures a lot of shimmering shows through. It's quite visible on the concrete posts in the video I posted. The wooden ceiling a bit later is also rather irritating to the eyes during motion. I'd be hesitant to call this filter a good choice at native resolution.

I still think the emphasis was on the motivation to make the source resolution less obvious to the eye for the same reason that they put a layer of film grain on top of the post processed image. It works, but it's still a very soft image and you can easily get the same performance at native resolution and with an image that looks a lot more "PC like" by turning down two options.

 

[tviceman]

Why don't any other sites show those gains and such poor AMD performance?

https://www.computerbase.de/2016-08.../#diagramm-ashes-of-the-singularity-2560-1440

Fury X is tied with OC'd 1070.

http://www.guru3d.com/articles_pages/zotac_geforce_gtx_1080_arcticstorm_review,16.html

Again Fury / Fury X are above 980 TI and close to 1070.

Does this mean AOTS is a valid game now that Nvidia is performing better?

Different graphics settings (4x MSAA, crazy preset) different drivers, and different CPUs.

The comparison wasn't AMD vs. Nvidia. The comparison was to show Nvidia is reaping the same performance benefits of DX12 as AMD. And the game was never invalid, I challenge you to find a post in my history that says otherwise.

Get off your high horse and have a beer Silverforce 3DVagabond. No one will judge you when you admit Nvidia is showing big gains in a DX12 game.

 

[Bacon1]

. No one will judge you when you admit Nvidia is showing big gains in a DX12 game.

I've been saying for a long time that DX12 would help both AMD and Nvidia. I've been hoping that devs start going pure Vulkan / DX12 and ditching their old DX11 engines.

It was other people saying that Nvidia had nothing to gain from DX12 and that it was all a conspiracy by AMD to hurt their performance and that Nvidia cards were already being used 100% so how could it help them that should admit that DX12 is good.

What I was questioning was why those results don't match most sites. Sadly most stop doing DX11 and DX12 for both so its hard to compare.

Back on topic, people found FRAPS logo showing up during the ending cutscene... great job Remedy:

https://youtu.be/It0sVpRx7AA?t=1522

 

[dogen1]

Because without the higher resolution part in "rerconstructing into a higher resolution canvas" you would still end up with a very soft 720p image. You're free to call it temporal reconstruction, but at the end of the day it is used as an elaborate method of upscaling.

Some example clips:
https://www.youtube.com/watch?v=BSg3opLozkw

I guess I'm just missing where they need to upscale. They reproject the previous frames and use a subpixel jitter to produce the additional samples needed for the higher resolution.

I don't know the entire process though..

 

[Piroko]

I guess I'm just missing where they need to upscale. They reproject the previous frames and use a subpixel jitter to produce the additional samples needed for the higher resolution.

I don't know the entire process though..

I have to admit, I didn't know about the subpixel jitter. But a temporal subpixel jitter helps to create an image with a high spatial resolution as long as the movement vector between frames is zero. If it isn't zero then you lose, well, depending on how it's handled, spatial resolution and/or temporal accuracy which you have to interpolate. Either one is still its own form of upscaling/upsampling.

 

[Carfax83]

Different graphics settings (4x MSAA, crazy preset) different drivers, and different CPUs.

Not only that, but according to the graph for AotS, NVidia is using the DX11 path and not the DX12 one. Might be a mistake, but I thought I should mention it.

 

[Bacon1]

Not only that, but according to the graph for AotS, NVidia is using the DX11 path and not the DX12 one. Might be a mistake, but I thought I should mention it.

Thats because they use whatever API they saw was faster

 

[Bacon1]

https://twitter.com/JokerReview/status/781594157101875200
https://twitter.com/JokerReview/status/781595281230036992
https://twitter.com/JokerReview/status/781833960984809472
https://twitter.com/TheJohnMan1980/status/781826254391488513

Bunch of IQ differences and missing incorrect lighting / shadows

 

[Azix]

"Optimizations"

love to see how this turns out. If they fix it with no perf loss, cool. if not, hmmmmm....

 

[dogen1]

I have to admit, I didn't know about the subpixel jitter. But a temporal subpixel jitter helps to create an image with a high spatial resolution as long as the movement vector between frames is zero. If it isn't zero then you lose, well, depending on how it's handled, spatial resolution and/or temporal accuracy which you have to interpolate. Either one is still its own form of upscaling/upsampling.

Yeah, that's why it looks better when nothing is moving.

Anyway, the technique they use was mentioned here a bit.

https://forum.beyond3d.com/posts/1941893/

https://forum.beyond3d.com/posts/1916756/

And there's a presentation linked.

 

[Unreal123]

"Optimizations"

love to see how this turns out. If they fix it with no perf loss, cool. if not, hmmmmm....

Yes very good optimization ,where AMD is missing lightning effects, and other effects.

RX 480 vs GTX 1060 from other youtuber.
https://www.youtube.com/watch?v=DL7OknOJZJg

 

[Bacon1]

Game is in a sad state of affairs. Another broken port and they haven't even shipped the physical copies people have ordered

 

[PowerK]

I don't understand the direction Remedy took in rendering technology and artistic approach.
Their so called "temporal reconstruction" (aka, upscaling) seems another way of saying we failed in optimization hence base resolution is 1280x720.

I bought a Steam version (DX11) on the weekend, I knew the game was a mess. But I wanted to check it out.
No matter upscaling is ON or OFF, the visual fidelity is blurry. Upscaling OFF actually does provide a cleaner look. Not much difference, however. (Unless you take a screenshot and do a ON/OFF comparison).
Upscaling on = blurry
Upscaling off = less blurry.

Lighting does look very impressive sometimes. Especially in time-laps scene with realtime lighting looks amazing.
Another thing I noticed was that this game showed one of the highest CPU utilization (hovers around 25~35% with 6950X overclocked to 4.4GHz) I've seen in games. Those visual quantum effects with objects much hit hard on the CPU. (or is the game running in XBOX One emulation mode. )

 

[dogen1]

I don't understand the direction Remedy took in rendering technology and artistic approach.
Their so called "temporal reconstruction" (aka, upscaling) seems another way of saying we failed in optimization hence base resolution is 1280x720.

I bought a Steam version (DX11) on the weekend, I knew the game was a mess. But I wanted to check it out.
No matter upscaling is ON or OFF, the visual fidelity is blurry. Upscaling OFF actually does provide a cleaner look. Not much difference, however. (Unless you take a screenshot and do a ON/OFF comparison).
Upscaling on = blurry
Upscaling off = less blurry.

Lighting does look very impressive sometimes. Especially in time-laps scene with realtime lighting looks amazing.
Another thing I noticed was that this game showed one of the highest CPU utilization (hovers around 25~35% with 6950X overclocked to 4.4GHz) I've seen in games. Those visual quantum effects with objects much hit hard on the CPU. (or is the game running in XBOX One emulation mode. )

Well, their temporal reconstruction and anti-aliasing allow them to shade 50% less pixels, meaning they were able to massively increase quality in other areas.

It's not a failure of optimization, it's a trade off they chose to make. Reduced sharpness(and whatever other artifacts that come with their solution) in exchange for more headroom for other gpu heavy effects.

 

[PowerK]

Well, their temporal reconstruction and anti-aliasing allow them to shade 50% less pixels, meaning they were able to massively increase quality in other areas.

It's not a failure of optimization, it's a trade off they chose to make. Reduced sharpness(and whatever other artifacts that come with their solution) in exchange for more headroom for other gpu heavy effects.

Probably. However, I think choosing MSAA which is one of the inefficient forms of AA with rendering techniques these days, to save GPU resources is beyond me. I mean they could also opt for more efficient form of AA such as SMAA rather than MSAA if their intention is to to save fillrate (& GPU resources) and increase quality in other areas.

 

[Erenhardt]

More than 1000 words: