CEOInterviews.AI
Start App
Get an email when Steven Walton gives a new interview
One email per new interview. Unsubscribe in one click.
Steven Walton
Co-founder & Host, Hardware Unboxed, Hardware Unboxed

Has Nvidia Nerfed DLSS 5? - New Model Tested

📅 Oct 19, 2026 Hardware Unboxed 20 MIN 290615 VIEWS 25 SEGMENTS · 1 SPEAKERS
Check out Asus Neo motherboards: https://rog.asus.com/motherboards/rog... Support us on Patreon:   / hardwareunboxed   YT Membership:    / @hardwareunboxed   Buy relevant products from Amazon, Newegg and others below: GeForce RTX 5090 - https://geni.us/JyDFBvM GeForce RTX 5080 - https://geni.us/dRy7y07 GeForce RTX 5070 Ti - https://geni.us/AuxCu GeForce RTX 5070 - https://geni.us/Wg4vPDv GeForce RTX 5060 Ti 16GB - https://geni.us/5Kgs2I GeForce RTX 5060 - https://geni.us/xmaeB Radeon RX 9070 XT - https://geni.us/CH0CxfR Radeon RX 9070 - https://geni.us/3neQ6nF Radeon RX 9060 XT 16GB - https:...

What Steven Walton said

Written from the verified transcript and checked against it. Every figure links to the moment it was said.

Steven Walton reviewed Nvidia's October 2026 DLSS 5 update, which claims to run twice as fast with 24% fewer parameters and 20% less VRAM. Testing on NBA 2K27 across the RTX 50 series, he found render times improved 83% at 4K and 94% at 1440p on average, though not the full doubling claimed. Frame rate gains varied by GPU, e.g., RTX 5080 at 4K rose from 53 to 78 FPS (47% increase), and RTX 5070 Ti reached above 60 FPS at 4K for the first time. Despite these gains, enabling DLSS 5 still imposes significant performance losses, e.g., 58% at 4K on the 5080. Visual quality remains broadly similar, with unresolved artifacts like lighting flicker, color shifts, and skin tone inconsistencies. Walton notes modded versions appear toned down, but NBA 2K27's official implementation doesn't. He calls the update a step forward but says several more doublings are needed for widespread adoption, and he questions why Nvidia didn't launch with this model.

Key takeaways

  1. Nvidia's October DLSS 5 model runs 83% faster at 4K and 94% at 1440p on average, not the claimed doubling.
  2. RTX 5080 at 4K improved from 53 FPS to 78 FPS with DLSS 5, a 47% gain, but still a 58% loss vs. no DLSS.
  3. Modded DLSS 5 versions seem toned down, but the official NBA 2K27 implementation shows no such reduction.
  4. Walton believes several more DLSS 5 performance doublings are needed before most gamers enable it without question.

Numbers and commitments

FigureWhat it refers toTypeAt
24% Reduction in DLSS 5 model parameters metric 0:00
20% Reduction in VRAM usage metric 0:00
1.4x Nvidia's claimed performance increase in NBA 2K27 guidance 0:32
47% FPS increase on RTX 5080 at 4K with new DLSS 5 metric 2:16
58% Performance loss on RTX 5080 at 4K with new DLSS 5 metric 2:16
83% Average speed increase at 4K across RTX 50 series metric 9:42
94% Average speed increase at 1440p across RTX 50 series metric 9:42
7.4 milliseconds New DLSS 5 render time on RTX 5080 at 4K metric 10:41
13.7 milliseconds Old DLSS 5 render time on RTX 5080 at 4K metric 10:41
71% Performance loss at 4K with old DLSS 5 on 120 FPS baseline metric 13:15

Chapters

  1. 0:00DLSS 5 October model claims
  2. 2:16RTX 5080 benchmarks
  3. 3:36RTX 5070 Ti benchmarks
  4. 4:22RTX 5070 and lower benchmarks
  5. 5:08RTX 5060 series benchmarks
  6. 6:40RTX 5050 benchmarks
  7. 7:30RTX 5090 benchmarks and render times
  8. 14:31Visual quality and artifacts
  9. 16:58Modded DLSS 5 comparisons
  10. 17:49Final verdict and future outlook
Steven Walton 0:00 ↗
Late last week, Nvidia released a new version of DLSS5 neural rendering that claims to run twice as fast as the old model. This October update reduces the number of parameters within the DLSS5 model by 24%, which halves the execution time and reduces VRAM usage by 20%, according to Nvidia. All of this has supposedly been achieved while maintaining comparable image quality. So, I thought it was worth revisiting this in everyone's favorite title, and once again the first to officially ship with the latest DLSS5 model, NBA 2K27.
Does the new DLSS5 model really run twice as fast on all GeForce RTX 50 series GPUs with no change to the output quality? Does this new model fix some of the issues we first brought up with DLSS 5 in our analysis video from a few weeks ago? And does this actually make DLSS5 worth using now, considering a shift in the balance between performance and visuals? One thing to clarify before I get into the benchmark results is that when Nvidia says the new DLSS 5 model is twice as fast, that does not mean games with DLSS 5 enabled will run twice as fast. This performance improvement only refers to the rendering time cost of DLSS 5, not all the other stuff that has to be rendered in the game. So frame rates will not be twice as high after this update. However, Nvidia do claim a performance increase of 1.4x in NBA 2K27, so we'll see whether that's accurate.
Today's sponsor spot is brought to you by ASUS and their new Neo motherboard lineup, including the TUF Gaming X870-Plus Wi-Fi 7 Neo, ROG Strix B850 Gaming Neo, and ROG Strix X870E Gaming Neo. Built for AMD's AM5 platform, the Neo range has been refreshed with upgraded power delivery to ensure maximum stability for the latest Ryzen processors. ASUS has also improved the building experience with their easy PC do-it-yourself features, including PCIe Q-Release, which allows you to remove your graphics card quickly without fighting with the retention clip. The Neo series also includes enhanced AI networking features, intelligent CPU cooling optimization, and redesigned memory pathways to improve signal quality, helping users to get more out of their high-speed DDR5 memory. So whether you're putting together a new system or upgrading your platform, learn more about the ASUS Neo motherboard range via the link in the video description.
Let's kick things off with Nvidia's flagship gaming graphics card, the GeForce RTX 5080. Previously, DLSS 5 was quite taxing on this GPU at 4K, sending NBA 2K27 from 185 FPS with DLSS5 disabled to just 53 FPS with DLSS 5 enabled. And as I discovered in the previous investigation, the performance here doesn't change much if you adjust settings or increase the level of upscaling, as the official DLSS5 implementation runs on the output frame at the output resolution, not before upscaling is applied. The new October model does increase performance quite a bit at 4K, going from 53 to 78 FPS, which is a 47% increase in this title. This makes 4K playable on this GPU to some degree, though there's still a 58% reduction in FPS compared to not using DLSS 5 at all. Previously, there was a 71% reduction in FPS. So, that gap has closed a bit. At 1440p, I saw a 42% increase in frame rate using the new version of DLSS 5 compared to the old, which reduces the overall performance loss from 54% to 34%. Enabling this feature still significantly reduces in-game performance, just not to the same degree as we saw previously. Meanwhile, at 1080p, the RTX 5080 is quite capable of running DLSS 5. Though, with neural rendering disabled, we do run into CPU and game engine bottlenecks that cap performance.
The RTX 5070 Ti also benefits from DLSS 5 model improvements. At 4K, the new model delivered 48% better performance and pushed NBA above 60 FPS at this resolution for the first time. However, without DLSS 5 enabled, the game did run at 166 FPS. So, the new model still comes with a massive 61% performance hit, making it hard to justify on this GPU at this resolution. 1440p, though, is a much better situation. The new model delivered 47% higher FPS than the old model and cut the performance loss from 62% to 44%. The game now runs above 120 FPS at this resolution with DLSS quality upscaling enabled, which is very playable in this sort of game. At 1080p, I saw a performance loss of 25% enabling the new model compared to 45% previously.
With the old DLSS 5 model, the RTX 5070 struggled at both 4K and 1440p. The new model provides a performance boost in both scenarios: a 52% higher frame rate at 4K and a 49% higher frame rate at 1440p. While this isn't enough to push 4K rendering above 60 FPS, it does deliver 100 FPS at 1440p. So essentially, this new model makes the 5070 much more usable at this common resolution. The performance loss overall compared to not using DLSS5 is still pretty large though. The frame rate was slashed by 60% at 4K and 54% at 1440p. Granted, this is better than what we were seeing previously. At 1080p on the RTX 5070, the impact of using DLSS 5 was around 39%.
DLSS5 was not usable on the RTX 5060 Ti 16GB at 4K using the old model and remains not usable with the new model despite a pretty significant performance increase. Going from 24 to 35 FPS represents a 46% performance gain at 4K, but without DLSS 5, the game runs at 95 FPS, so it's still a whopping 63% performance loss when enabling neural rendering. 1440p is a better situation, though. The old model ran below 60 FPS at this resolution. The new model pushes above 60 FPS, so that's a nice increase here for what used to be a mid-range GPU option. At 1440p, enabling DLSS 5 reduced performance by 55%. And at 1080p, I saw the frame rate cut in half.
On the RTX 5060, once again, I'm using low settings instead of max settings to avoid VRAM issues. The new model delivers 57%, 54%, and 53% higher frame rates at 4K, 1440p, and 1080p respectively, which makes 1440p gaming in particular a lot more pleasant, now running at 71 FPS. As low settings do significantly increase the base frame rate without DLSS 5 enabled, it does mean a performance loss of 65% at 1440p when turning on the feature. 1080p also becomes more usable on the 5060 with the new DLSS5 model now capable of 112 FPS at this resolution. It technically exceeded our threshold for usability previously as well, but going from 240 to 73 FPS was a bit of a sting. The performance loss has now been reduced to 53% at this resolution.
Then we have the RTX 5050. This entry-level model was previously not fast enough to run DLSS 5 in my opinion. It achieved just 33 FPS at 1440p and 53 FPS at 1080p using the old model. The new model runs a lot faster on this GPU, a 55% increase in FPS at 1440p and 1080p. And while this isn't enough to push it over the 60 FPS threshold at 1440p, it does increase the frame rate to 82 FPS at 1080p. A huge performance loss compared to not running DLSS 5, but playable. One weird thing I did see with the RTX 5050 specifically, and not so much the other graphics cards, is that 1% low performance didn't really improve at all using the new DLSS 5 model, and in some cases went backwards. Not sure why that is. Maybe it's just a testing quirk. But at 1080p, if you get an 82 FPS average but have 1% lows still in the 30s, the experience is not the smoothest.
Lastly, here's a look at Nvidia's workstation GPU, the RTX 5090, which by all reports was recently discontinued. This was clearly the best GPU to use with DLSS5, as even with the old model, you got playable frame rates at all resolutions. The new model works quite well here. At 4K, the output frame rate increased by 45% and the performance loss reduced from 60% to 43%. At 1440p, it's an even better situation, though the performance without DLSS 5 enabled is again limited. The new model delivered a 35% higher frame rate, and the overall performance loss was pretty small, just 18%. So, if you happen to be using an RTX 5090 to play NBA 2K27 at 1440p, you may as well enable DLSS5 given the frame rates we're seeing here. And of course, the 5090 is also very capable at 1080p, though I'm not sure how many people pair a 5090 with a 1080p display.
With this performance data at hand, we can calculate DLSS 5's render time across the full RTX 50 series stack and use that to calculate various performance scenarios when this technology is enabled. Here are the original results I showed in my previous DLSS5 performance video. So, this is the original model release. You can see that at 4K, a lot of performance scenarios were not possible. On the RTX 5070 Ti, for example, DLSS5 had a render time of 17 milliseconds at 4K, which meant that no matter how fast the rest of the game runs, you'd never see a 60 FPS render rate on that GPU. The only way to get the visual smoothness of 60 FPS would be to use heavy frame generation. At 1440p, more cards were capable of rendering DLSS5 at acceptable frame rates, but there were still a lot of scenarios that I felt were not very reasonable. The RTX 5070, for example, could achieve 60 FPS with DLSS 5 enabled, provided the game was already running at 156 FPS before switching the feature on. And if you were running the game at 80 FPS previously, your frame rate would drop to just 44 FPS when using neural rendering, which is a substantial hit in a performance scenario that crops up a lot in single player gaming. The 5060 Ti and below were not really capable of running DLSS 5 at this resolution, at least with an acceptable output.
The new October model of DLSS 5 is much faster, significantly reducing the render time. On average at 4K, the model ran 83% faster in my testing and 94% faster at 1440p. This is an average across all the RTX 50 series models. So, we're not quite at the doubling Nvidia claimed in their announcement, but it's substantial enough that it might as well be a doubling, and an 83% increase in speed at 4K is massive. What's interesting is that the speed up seen from this model does vary depending on the overall performance of the GPU at hand. On the RTX 5090, for example, the new DLSS 5 model was slightly more than twice as fast at 4K. The RTX 5070 was 83% faster, but the RTX 5060 was only 76% faster. I saw similar scaling at 1440p where the speed up was higher on the faster GPU models. This is just an interesting observation because even the RTX 5050 at 4K runs the new model 73% faster, which is an excellent outcome on the performance side of things.
This makes DLSS5 a lot more usable across most of the GeForce 50 series. Previously, the 5080 took 13.7 milliseconds to render DLSS 5 at 4K, which made it very difficult to hit 60 FPS unless the game was already running at a super high frame rate. The new model takes just 7.4 milliseconds to render, meaning that you only need the game to be running at 108 FPS before enabling DLSS 5 to still achieve 60 FPS afterwards. And for games running at 80 FPS previously, the new DLSS5 model now achieves 50 FPS compared to 38 FPS with the old model. There's also good news for the RTX 5070 Ti, which again was mostly unusable at 4K. The old model turned 120 FPS gaming into 40 FPS gaming. The new model is much more capable. A 136 FPS starting frame rate will let the game run above 60 FPS with neural rendering enabled, which is still a huge performance impact, but at least that performance configuration is realistic in games. The RTX 5070 though, it's still pushing it at 4K. Ultimately, I think it will be difficult for gamers to achieve a good frame rate with DLSS 5 enabled on that GPU at 4K.
The news is even better at 1440p. With the old model, the RTX 5080 and 5070 Ti were usable at 1440p, and they become even more usable with the new model. On the 5070 Ti, previously you needed the game to run at 113 FPS to achieve 60 FPS with DLSS 5 enabled. That required frame rate has dropped to just 79 FPS, meaning that a game running at 79 FPS before DLSS 5 will only drop to 60 FPS with DLSS 5 enabled. It's still a large performance hit for a single visual feature, a 24% reduction in FPS for this example, but it's not that crazy. We're now within the realm of other performance taxing settings as opposed to being a huge outlier. I've also added a row here for 120 FPS gaming with DLSS 5 enabled. It's pushing it a bit, but the RTX 5080 and of course the 5090 could realistically do this now. And it's technically possible on a few other GPUs as the overhead now allows it.
The big deal at 1440p though is the new model is usable with acceptable levels of performance right down to the RTX 5060, whereas previously I'd say the cutoff was the RTX 5070. To achieve 60 FPS with DLSS 5 enabled, you only need the game running at 89 FPS now on a 5070, 109 FPS on a 5060 Ti, and 134 FPS on a 5060. Again, in some of these configurations, we're still looking at a huge performance hit using this feature. But previously, some of these cards weren't able to output even 60 FPS under realistic conditions.
With the new DLSS 5 model able to run about 80 to 90% faster than the original model, this reduces the DLSS5 performance hit across the board. Take a game running at 120 FPS as an example, across the RTX 50 series stack. Enabling the old DLSS5 model resulted in a 71% performance loss at 4K. That's been reduced to 58% with this new model. At 1440p, the old model caused a 54% performance loss. That's been reduced to 39% with the new model. In these situations where you were previously getting a nice high frame rate gaming experience, we're still seeing a very substantial performance loss when neural rendering is enabled. The RTX 5070, which is pretty much bang on the average result, goes from 120 FPS to 73 FPS with DLSS 5 enabled at 1440p. Essentially, you have to sacrifice the high frame rate experience and the latency benefits that brings to use this feature, but you don't have to sacrifice as much as you did previously.
For games running at 80 FPS, the impact of turning on DLSS 5 is less because DLSS 5's render time takes up a lower portion of the overall render time. With the old model at 1440p, on average, DLSS5 took games from 80 FPS to 44 FPS, a 45% performance loss. That's been reduced to a 31% performance loss with the new model.
Do these performance improvements come with any impact of visual quality? I'm not going to do a huge deep dive into this, but my general impression from using the new model in NBA 2K27, the official implementation, is that visual quality is broadly the same, potentially slightly worse in some situations. I guess there's two ways to look at this. On the one hand, it's great that Nvidia have been able to improve performance while largely delivering similar visual output. That's a step in the right direction. On the other hand, DLSS5's original model had quite a few visual issues. So, stagnation or even a small backward step on the visual front is not ideal. What we really need to see with DLSS 5 is performance and quality improving simultaneously.
The new DLSS5 model still has noticeable issues with stability, though it's hard to say based on my testing whether these issues have gotten worse. Lighting flicker, where surfaces shift in tone slightly from frame to frame, is still very much present in the updated October model. And I saw these issues in identical scenarios to my original DLSS 5 video. This flicker can be quite distracting, especially as once you see it, it's hard to unsee it. And I found myself spotting it in more and more situations as I continued to play the game. I also found with the original DLSS5 model that neural rendering had a hard time keeping objects and skin tones consistent at different distances from the camera. Zooming out on this Gatorade tub, for example, would make the surface darker and a deeper, different shade of orange when DLSS5 was enabled, whereas it accurately remained the same color performing the same action with DLSS5 disabled. This behavior remains with the new DLSS 5 model. It just isn't able to keep a consistent visual output within the scene, which can give a bit of an uncanny feeling as objects morph in color slightly as you move around.
This effect also happened with skin tones, and this is still a big issue with the October DLSS 5 model. DLSS5 is best equipped to enhance skin when a character is relatively close to the camera, dominating the overall field of view. As you zoom out or get further from a specific person, their skin typically gets darker and shinier with neural rendering enabled, something that does not occur when this feature is disabled. This makes human characters look a bit weird in my opinion, unless the character is close to you. The other major artifact I discovered was related to disocclusion. DLSS5 had a tendency to create a halo-like effect around characters, and I think this has actually gotten slightly worse and more visible in the new DLSS5 model, but it is hard to tell. Basically, none of the artifacts and issues I saw with the original model have been resolved in this update. It's purely on the performance side where we are seeing benefits.
Interestingly, when I was researching the new DLSS 5 model, I came across quite a few posts from people using modded versions of DLSS5 complaining about how the new model is toned down compared to the original model, creating a less intense effect and a smaller overall visual change, especially with the sliders set to high levels. This was interesting to hear because in NBA 2K27 with the official DLSS5 implementation, the effect of neural rendering seems broadly similar to what I captured with the original model, it doesn't seem toned down at all. But then NBA 2K27 does not have a particularly intense neural rendering implementation. It mostly attempts modest improvements to avoid overcooking the image and generating an AI slop output. Perhaps part of the optimizations in the new DLSS 5 model involved nerfing the top parts of the intensity range, which developers weren't going to use in the first place.
Based on what I've seen in NBA 2K27, the new DLSS 5 model is a step in the right direction to making DLSS 5 a genuinely usable feature in games. This new model runs much faster than the old model, and visual quality seems broadly similar. It's a decent level of optimization that makes neural rendering viable on more GPUs than it was at launch.
Unfortunately, doubling the speed of processing DLSS 5 is only one of many improvements that need to be made to this technology. Enabling neural rendering still comes with a relatively high performance cost even after this update. It's just that this performance cost is not as massive as it used to be. We need several more doublings of DLSS 5 performance to get it to the level where I think the clear majority of gamers would enable this feature without question, especially on mid-range hardware and lower. And it's clear that performance was the focus of this latest DLSS5 model, not visual quality. All of the artifacts and issues with the original DLSS 5 release remain here, and these things need to be resolved. Otherwise, neural rendering will continue to produce worse image quality in some aspects than traditional rendering.
Whether or not Nvidia can improve both performance and image quality simultaneously remains to be seen. They haven't done so with the October model. What I find a bit bizarre about this is why Nvidia didn't just wait to make this model the original official release of DLSS5. It's only been integrated into one game so far. There's hardly been a rush of developers jumping on board this technology. So, the official launch could have waited until this newer optimized version was ready. Why push out the slow version that doesn't run well on most GeForce GPUs if it's only going to be used in one game and then superseded within six weeks? From here on, DLSS 5 titles will use the new DLSS 5 model. So, that original model is going to be consigned to history, at least officially. The update to Hogwarts Legacy coming in November, for example, will use the new model. And I'll probably give that update a go when it's ready.
For now, that's it for this one. If you do want to support Hardware Unboxed and the independent testing that we do, the best way to do that is via our Patreon page. Link to that is in the description below. If you sign up, you get access to all the usual benefits like our Discord community, monthly live streams, and BTS content. So, thanks for watching and I'll catch you in the next one.

Cite this transcript

APA, MLA, BibTeX
APA

Walton, S. (2026, October 19). Has Nvidia Nerfed DLSS 5? - New Model Tested [Interview transcript]. Hardware Unboxed. CEOInterviews.AI. https://ceointerviews.ai/interview/3229064/

MLA

Steven Walton. "Has Nvidia Nerfed DLSS 5? - New Model Tested." Hardware Unboxed, 19 Oct. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/3229064/.

BibTeX
@misc{walton2026_3229064,
  author       = {Steven Walton},
  title        = {Has Nvidia Nerfed DLSS 5? - New Model Tested},
  howpublished = {Interview transcript, Hardware Unboxed. CEOInterviews.AI},
  year         = {2026},
  month        = {oct},
  url          = {https://ceointerviews.ai/interview/3229064/},
  note         = {Speaker-attributed transcript with timestamps}
}