007 First Light PC Review - DLSS Finally Fixed, But Other Issues Still Need Fixing

[00:02] Cyberpunk 2077 First Light, it's one of

[00:04] the biggest gaming highlights of the

[00:06] year from my perspective. Now, we've

[00:08] talked in depth about the game's new

[00:10] technology and the mostly successful

[00:12] console versions. We've also shared

[00:14] initial thoughts on the PC build in DF

[00:16] Direct, but there's more to discuss

[00:19] here. There have been developments. One

[00:21] of the reasons this review was delayed

[00:23] was that a crucial feature of the game,

[00:25] DLSS upscaling, was essentially broken.

[00:28] With the latest patch, 1.05,

[00:31] that is now addressed, making First

[00:34] Light that much more accessible to those

[00:37] with lower-end systems using lower-end

[00:40] RTX graphics cards. And beyond that, IO

[00:43] Interactive, well, they were generous

[00:45] enough to share their console equivalent

[00:47] settings with us. But the question

[00:49] remains, how much further does the game

[00:51] scale, both downwards and upwards? And

[00:54] beyond console quality, are we looking

[00:56] at diminishing returns? And have our

[00:58] thoughts on the PC version changed much

[01:01] at all since the direct beyond the

[01:03] upscaling issues? But before we go any

[01:05] further, a big thanks to Alienware for

[01:07] sponsoring our PC performance reviews.

[01:10] >> [music]

[01:10] >> And with all of our high-end testing,

[01:12] we'll be using the firm's newly revised

[01:14] Alienware Area-51.

[01:17] Yes, the flagship gaming PC has received

[01:19] a full refresh with a refined focus on

[01:22] high-quality design paired with high-end

[01:24] cooling, taking care of the best gaming

[01:27] silicon on the market. In addition to

[01:29] supporting high-end RTX 5000 GPUs up to

[01:32] the RTX 5090, the new Area-51 also

[01:35] supports AMD's top-of-the-line Ryzen

[01:38] 9000 processors, from the brilliant

[01:40] Ryzen 7 9800X3D

[01:43] through to the latest Ryzen 9 9950X3D

[01:46] II. In our tests, Ryzen 9000 X3D

[01:50] processors offer up by far the best CPU

[01:53] gaming performance available, a key

[01:55] recommendation for a high-spec gaming

[01:57] PC.

[01:58] Keeping thermals in check is the new

[02:00] gasket architecture using a combination

[02:03] of 180 and 140 mm fans to create

[02:06] positive pressure, moving 25% more air

[02:09] while being 45% quieter than prior

[02:12] Area-51 generations. There's a 1500 W

[02:15] platinum power supply and more than

[02:18] enough thermal headroom to handle the

[02:19] extreme demands of the RTX 5090, which

[02:23] pulls 600 W just by itself. Put simply,

[02:26] the new Area-51 is built for the

[02:28] absolute limit of modern PC gaming, the

[02:30] perfect hardware for testing the

[02:32] high-end PC experience in our

[02:34] performance reviews. Check out the link

[02:36] in the video description below to find

[02:38] out more.

[02:39] But of course, that kind of hardware is

[02:41] offering gigantic levels of CPU and GPU

[02:44] power. And so in addition to testing

[02:47] with that kit, we're keeping things

[02:48] grounded as well by looking at more

[02:50] modest components. So, let's kick off by

[02:53] looking at the first boot experience on

[02:56] a Ryzen 5 5600 test system, shader

[02:59] compilation took around 3 minutes to

[03:01] complete. Thankfully, gameplay

[03:03] afterwards was largely hitch-free with

[03:05] no obvious PSO related stutter

[03:07] encountered during normal play. Now, as

[03:10] I mentioned in the intro, the good news

[03:12] is that DLSS is basically fixed now with

[03:15] the most recent update. To quickly

[03:17] recap, DLSS was delivering substantially

[03:19] less performance than expected relative

[03:21] to its internal rendering resolution,

[03:24] resulting in surprisingly small gains

[03:26] across the board. Now, according to the

[03:28] release notes, the update fixes an issue

[03:30] where the game could launch with

[03:32] quote-unquote incorrect upscaling

[03:34] settings impacting performance. Whatever

[03:36] the underlying cause, our updated

[03:38] testing shows DLSS performing much more

[03:40] as expected, uh delivering the kind of

[03:42] scaling we'd normally expect from the

[03:44] technology.

[03:45] Some limitations remain, however. The

[03:47] game still lacks support for FSR 4 and

[03:50] XeSS, while the dynamic resolution

[03:53] scaling system used by the console

[03:55] renditions is also absent on the PC

[03:58] version. Frame generation support is

[04:00] similarly limited, with only DLSS frame

[04:03] generation and multi-frame generation

[04:05] available. AMD and Intel users are left

[04:08] without equivalent FSR or XeSS frame

[04:11] generation options, which isn't great.

[04:15] We did uncover one curious result during

[04:17] CPU-limited testing, however. At 1080p

[04:20] and CPU-bound scenarios, FSR 3

[04:23] consistently underperformed. In other

[04:25] words, FSR 3 seems to be carrying some

[04:28] kind of measurable CPU overhead once GPU

[04:31] limitations are removed. And this one's

[04:33] a bit of a mystery, bearing in mind that

[04:36] FSR 3 doesn't have much, if any, CPU

[04:39] overhead in our experience. It's an

[04:41] entirely GPU-focused task. And as for

[04:45] CPU performance itself, First Light can

[04:48] become quite demanding in scenes with

[04:50] large numbers of NPCs. Areas such as

[04:53] this hotel level are enough to push

[04:54] frame rates below 60 FPS on the Ryzen 5

[04:58] 5600, while chase sequences that heavily

[05:01] test the game's streaming systems can

[05:04] also make it difficult to sustain a

[05:06] consistent 60 FPS. Uh by comparison, the

[05:09] Ryzen 9 7950X3D delivers more than

[05:13] double the performance of the 5600 under

[05:15] the same CPU-limited conditions. That

[05:17] gives you some idea of the kind of

[05:19] contrast between mid-range and high-end

[05:21] processor performance in this game. Um

[05:23] but beyond that, you're looking at frame

[05:25] generation to push yourself further into

[05:27] high refresh rate display territory.

[05:31] Now, with that out of the way, let's

[05:32] take a look at what the higher PC

[05:33] settings actually bring to the table.

[05:35] Going to start with a volumetric effects

[05:37] quality, which governs the quality of IO

[05:40] Interactive's new Smolder volumetric

[05:42] system, one of the key rendering

[05:44] upgrades discussed in our recent

[05:46] developer interview. It's also one of

[05:48] the most visually prominent additions to

[05:51] Glacier's rendering technology while

[05:53] simultaneously being one of the game's

[05:54] most demanding rendering features across

[05:56] the board. When this feature is prolific

[05:59] on screen, frame rates can tumble.

[06:01] What's surprising, however, is just how

[06:03] little changes visually as we move down

[06:06] at the preset ladder. Ultra does produce

[06:09] a slightly cleaner result while lower

[06:11] settings introduce a somewhat grainier

[06:13] appearance, but the underlying behavior

[06:15] effect remains largely unchanged. Smoke

[06:18] density persistence lighting

[06:20] interaction, and shadowing all remain

[06:22] intact across the range of presets,

[06:25] making the differences surprisingly

[06:26] difficult to spot during normal

[06:28] gameplay. Performance, however, tells a

[06:31] different story with high improving

[06:33] frame rate by around 16 percentage

[06:35] points over Ultra. Medium and low

[06:37] provide only modest additional gains

[06:39] beyond that, making Ultra difficult to

[06:41] justify given the relatively minor

[06:44] visual differences. Volumetric fog

[06:46] quality controls the fidelity of the

[06:48] game's atmospheric fog and volumetric

[06:51] lighting effects, and it's an effect

[06:53] used extensively throughout First Light.

[06:56] Visual differences between Ultra and

[06:58] high are generally difficult to spot

[07:00] during normal gameplay while medium and

[07:02] low introduce a slightly coarser

[07:04] appearance to the effect. Despite its

[07:06] heavy use throughout the game,

[07:08] performance scaling remains fairly

[07:10] reasonable with medium improving frame

[07:12] rate by around 10 percentage points over

[07:14] Ultra test scene while low extends that

[07:17] lead slightly further. Unlike volumetric

[07:20] effects quality, however, the visual

[07:22] reductions here are more proportionate

[07:24] to the performance savings on offer.

[07:26] Next up is texture quality, and this

[07:28] scales consistently across the range

[07:30] with each step down reducing texture

[07:32] detail on surfaces and environmental

[07:34] assets. Moving from Ultra down to medium

[07:37] results in fairly subtle reductions to

[07:39] texture quality, though Ultra does

[07:41] provide the sharpest presentation. Low,

[07:44] however, is a much more obvious

[07:45] downgrade with noticeably blurrier

[07:48] textures and some signage becoming

[07:50] illegible. 8 GB cards may be better

[07:52] served by medium at 1440p

[07:55] as it retains most of the visual quality

[07:57] while reducing memory usage. If you

[08:00] prefer to play at 4K, however, further

[08:02] reductions may be necessary to keep VRAM

[08:04] usage in check. Moving on, let's look at

[08:07] level of detail which controls the draw

[08:09] distance of distant objects and foliage.

[08:12] Visually higher on ultra noticeably

[08:14] reduce pop-in compared to the console

[08:16] equivalent medium preset by keeping the

[08:19] higher detailed geometry visible much

[08:21] further from the camera. This setting

[08:23] has very little impact when GPU limited

[08:26] with only around 1 to 2 percentage

[08:28] points separating ultra from low.

[08:30] However, at 1080p where the game becomes

[08:33] more CPU limited, uh the gap grows to

[08:35] around 6 to 7 percentage points on a

[08:38] Ryzen 5 5600 making this primarily a CPU

[08:42] scaling related setting.

[08:44] The focus now shifts to shadow quality

[08:47] and this one scales pretty much as you'd

[08:49] expect. Higher settings produce sharper,

[08:51] more clearly defined shadows while each

[08:53] step down results in progressively

[08:55] softer shadowing. In the scene we tested

[08:57] performance differences were fairly

[08:59] minor overall with only the low setting

[09:01] delivering a measurable gain of around 5

[09:03] percentage points over medium.

[09:06] Much of that performance uplift comes

[09:07] from the removal of self-shadowing on

[09:09] vehicles and foliage as well as a

[09:11] general reduction in shadow detail.

[09:14] Global illumination next which controls

[09:16] the quality of the game's screen space

[09:18] GI elements and radiance probes. Ultra

[09:20] and high look largely identical

[09:22] maintaining indirect light bounce and

[09:24] scene depth while medium and low

[09:25] noticeably reduce bounce lighting

[09:27] particularly in the distance alongside

[09:30] some of the subtle shadowing that helps

[09:31] ground objects within the scene with

[09:34] only a single percentage point

[09:35] separating the highest and lowest

[09:37] settings in our testing, the visual

[09:39] reductions on medium and low are

[09:41] difficult to justify. Reflection quality

[09:43] primarily affects the quality of the

[09:45] game's planar reflections. While these

[09:47] are used exclusively on mirrors, mirrors

[09:50] themselves appear frequently throughout

[09:52] First Light, making this a more relevant

[09:54] setting than it might initially seem.

[09:57] Ultra and high produce near identical

[09:59] results, while medium noticeably reduces

[10:01] reflection clarity, resulting in a

[10:04] softer and lower resolution appearance.

[10:06] Low removes planar reflections

[10:08] altogether, leaving mirrors to rely

[10:10] solely on cube map reflections.

[10:12] Performance costs are largely tied to

[10:14] these planar reflections, with medium

[10:16] improving frame rate by around 10

[10:17] percentage points over ultra in our

[10:19] mirror test, while low extends that lead

[10:22] further.

[10:23] Screen space reflections tell a

[10:25] different story. Here, visual

[10:27] differences are limited to slightly

[10:28] reduced reflection coverage at lower

[10:30] settings, while performance remains

[10:32] effectively unchanged. As a result, the

[10:35] primary trade-off of this setting is

[10:37] really centered around the quality and

[10:39] cost of planar reflections.

[10:41] Taken as a whole, the console equivalent

[10:43] settings provided by IO Interactive also

[10:46] happen to align closely with what we'd

[10:48] consider our optimized settings. For

[10:50] most users, this is where we'd start, as

[10:52] they deliver a visual experience that

[10:54] remains remarkably close to maximum

[10:57] settings while avoiding much of the

[10:59] associated performance cost. But, what

[11:02] happens if we simply max out everything

[11:04] then?

[11:04] >> [music]

[11:05] >> Using the console equivalent settings as

[11:06] our baseline, moving every option to its

[11:08] highest setting reduced performance by

[11:10] around 18% in our testing. So, the

[11:13] improvements are certainly there if you

[11:15] go looking for them, but they're spread

[11:16] across the numerous smaller refinements

[11:19] rather than any single transformative

[11:21] upgrade. And that perhaps explains why

[11:23] the absence of the promised path tracing

[11:25] updates is felt so strongly.

[11:27] >> [music]

[11:27] >> The current ultra settings improve the

[11:29] visuals in a number of ways, but they

[11:31] don't fundamentally alter the visual

[11:33] experience in the same way a more

[11:34] advanced rendering feature potentially

[11:37] could. Now, path tracing, it's coming.

[11:39] So, let's just say that we're very much

[11:40] looking forward to seeing how the game

[11:42] evolves once that update arrives. At the

[11:45] same time, it's good to see IO

[11:46] Interactive responding to feedback. The

[11:49] latest update does appear to have

[11:51] resolved the DLSS scaling issues we

[11:53] previously highlighted with our updated

[11:56] testing showing performance behavior

[11:58] much more in line with expectations. But

[12:01] that said, there are still areas where

[12:02] we'd like to see improvement. IO

[12:04] Interactive told us that the use of the

[12:06] static FSR 3.1.5 library for PC is down

[12:11] to commonality with the console versions

[12:13] of the game. Essentially, it streamlines

[12:16] development for them. But my view at

[12:18] this point is that if a new game

[12:20] supports DLSS, the dynamic library FSR

[12:23] framework should be in there, too. And

[12:25] there needs to be support for Intel's

[12:27] upscaling technology, as well. FSR 4 and

[12:31] XeSS are both absent right now, while

[12:34] frame generation support is currently

[12:36] limited just to Nvidia hardware. IO's

[12:38] prior work on Hitman was exemplary in

[12:41] this regard. All technologies embraced

[12:44] and supported, and I kind of want to see

[12:45] that here, too.

[12:47] So, I think we got a pretty decent PC

[12:49] port here for 007 First Light, but there

[12:51] are areas where we'd like to see the

[12:53] feature set expanded. The lack of

[12:55] broader upscaling support is one, while

[12:58] path tracing remains the obvious missing

[13:00] component for higher-end hardware. We'll

[13:03] be reporting back on the path tracing

[13:05] side of things soon, but for now, if you

[13:07] found the content interesting or useful,

[13:09] please do like, subscribe, share, and

[13:11] ring bells and such for instant

[13:13] notifications. DF Supporter Program,

[13:15] >> [music]

[13:16] >> uh patreon.com/digitalfoundry.

[13:18] Do check that out to support the team

[13:19] and our mission. Get early access to

[13:22] content, closer access to the team,

[13:24] ad-free video downloads, ad-free

[13:26] website. Tons of good stuff there. But

[13:28] that's all for me for now. Thanks for

[13:30] watching and supporting Digital Foundry.