A week after Nvidia’s DLSS 5 made its commercial debut inside NBA 2K27, the numbers from independent test labs have settled into a consistent, uncomfortable pattern: turning on the company’s new 3D-Guided Neural Rendering pipeline cuts native frame rates by roughly 50 to 73 percent before Multi Frame Generation (MFG) claws the number back up. That is not a rounding error or an early-driver bug. It is, according to half a dozen outlets that benchmarked the game independently, how DLSS 5 is designed to work.
DLSS 5 launched on September 3, 2026, as the first shipping implementation of Nvidia’s 3D-Guided Neural Rendering technology, built in partnership with NBA 2K27 developer Visual Concepts. One week later, the picture that has emerged from Tom’s Hardware, PC Gamer, DSOGaming, Tweaktown, Digital Foundry, and Tom’s Guide is a technology that trades raw rendering speed for photorealistic 3D scene reconstruction, then leans almost entirely on generated frames to make up the difference. This is not a story about one benchmark result. It is a story about what happens when a GPU maker’s flagship rendering feature ships with a performance cost nearly every major reviewer now agrees is unusually steep, and about what that means for buyers, for Nvidia’s rivals, and for how “performance” gets marketed going forward.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What DLSS 5 actually is, and why NBA 2K27 got it first
DLSS 5 is Nvidia’s branding for 3D-Guided Neural Rendering, a pipeline that differs meaningfully from the upscaling-plus-frame-generation approach used in DLSS 3 and DLSS 4. Rather than simply upscaling a rendered frame and inserting interpolated frames between real ones, DLSS 5 uses geometry and lighting data from the 3D scene itself to guide a neural network’s reconstruction of crowds, court reflections, and lighting in real time. Nvidia and Visual Concepts describe it as the first full commercial deployment of that pipeline, chosen for NBA 2K27 in part because packed arenas, dynamic crowd shading, and reflective hardwood courts stress exactly the kind of scene complexity the technology targets, according to the official Nvidia GeForce Game Ready driver notes.
The feature ships exposed through three distinct toggles in NBA 2K27 and the Nvidia app: DLSS Super Resolution, DLAA, and the new DLSS 5 Neural Rendering mode paired with Dynamic Multi Frame Generation, which offers 2x, 4x, and 6x frame multipliers. Nvidia’s own driver notes promise that with DLSS switched on, an RTX 5090 can hit 370 FPS at 4K Ultra settings with ray tracing enabled. That figure, as reviewers were quick to point out, describes the full stack running with high-multiplier frame generation layered on top of neural rendering. It is not a description of what the neural rendering step itself costs, and that distinction is where the current controversy comes from.
The frame-rate hit, GPU by GPU
Strip away Multi Frame Generation and test DLSS 5’s neural rendering step in isolation, and the picture changes sharply. Tom’s Hardware ran every RTX 50-series card through NBA 2K27 and confirmed what it called a big performance hit across the entire Blackwell lineup, while noting that even the RTX 5060 can still manage close to 60 FPS at 1080p with DLSS 5 enabled. DSOGaming reported a 50 to 60 percent drop in its own testing, watching frame rates fall from 176 FPS to 108 FPS at native 1440p Max settings with ray tracing, and from a higher baseline down to 63 FPS at native 4K. Digital Foundry’s numbers, relayed through a ResetEra benchmark thread, lined up closely: an RTX 5090 fell from 128 FPS native to 63 FPS with DLSS 5 on, a 51.2 percent drop, while an RTX 5070 dropped from 120 FPS to roughly 53 FPS, and an RTX 5060 fell from about 70 FPS to 40 FPS.
PC Gamer’s hands-on testing on an RTX 5080 found the damage could run even deeper in specific scenes, with frame rates dropping by up to 73 percent and settling as low as 49 to 55 FPS without frame generation engaged, a result GamesRadar’s own hands-on testing echoed almost exactly. Tom’s Guide, testing on a mid-range gaming laptop rather than a desktop, found the cost varies wildly by scene type: menu screens dropped from 109.8 FPS to 30.4 FPS (a 72 percent hit), and instant replays fell from 92.6 FPS to 26.9 FPS, while one gameplay sequence showed almost no frame-rate cost at all but a latency jump from 79.8ms to 111.2ms. That inconsistency is itself notable. It means the “performance hit” isn’t a single number so much as a range that depends heavily on what the camera is looking at.
| GPU | Native FPS (no DLSS 5) | DLSS 5 On (no MFG) | Approx. FPS drop | Source |
|---|---|---|---|---|
| RTX 5090 | 128 FPS (4K, Max, RT) | 63 FPS | ~51% | Digital Foundry / ResetEra, DSOGaming |
| RTX 5080 | Not isolated | 49–55 FPS | Up to 73% in specific scenes | PC Gamer, GamesRadar |
| RTX 5070 | 120 FPS (1440p, Max) | ~53 FPS | ~55% | Digital Foundry / ResetEra |
| RTX 5060 | ~70 FPS (1080p High, RT) | ~40 FPS | ~42–43% | Digital Foundry / ResetEra |
| RTX 5060 (with DLSS Quality + 6x MFG) | — | 258 FPS | N/A (full stack) | Tom’s Hardware / Tweaktown |
These figures line up closely with earlier benchmark analysis that clocked a roughly 48 percent FPS tank alongside an 802-watt power draw on high-end configurations. The pattern holds across nearly every independent test: DLSS 5’s neural rendering step alone behaves like it caps performance somewhere in the 40 to 65 FPS range on current-generation hardware at higher settings, regardless of how fast the card is natively. Tweaktown’s analysis went as far as calling DLSS 5 by itself “effectively a 60 FPS technology,” a characterization that lines up with what every other outlet measured independently. The card you buy changes your starting point. It doesn’t change the size of the hit DLSS 5 takes once you turn it on.
Multi Frame Generation is doing the heavy lifting
Nvidia’s marketing headline number, the 370 FPS claim for an RTX 5090 at 4K Ultra with ray tracing, only becomes achievable once Dynamic Multi Frame Generation is stacked on top of DLSS 5. Tweaktown’s breakdown of an RTX 5090 run showed DLSS 5 alone landing around 50 FPS, then climbing to 97 FPS with 2x MFG and 170 FPS with 4x MFG. Tom’s Hardware separately clocked an RTX 5060 at 258 FPS using DLSS Quality mode combined with 6x frame generation at 1080p, a result the outlet noted required stacking every available multiplier to get there. In other words, the headline frame-rate numbers Nvidia is putting in press materials describe a fully assembled stack of upscaling plus aggressive frame insertion, not what the neural rendering feature costs on its own.
That’s a meaningfully different story than how DLSS 3 and DLSS 4 were pitched. PCMag’s hands-on testing found that performance could actually drop slightly even with 6x frame generation engaged, because the GPU was spending its cycles generating synthetic frames instead of rendering real ones, and recommended that anyone gaming at 4K with max settings test with frame generation off or at a lower multiplier to find their actual best frame rate. Nvidia’s own driver documentation adds a technical caveat that reinforces how tightly coupled the system is: DLSS 5’s frame generation mode is explicitly incompatible with frame rate limiters and V-Sync unless every required system component is present and enabled, a constraint that wasn’t nearly as strict under earlier DLSS generations.
How this compares to DLSS 3 and DLSS 4 in earlier games
DLSS upscaling has historically been a performance-positive technology. DLSS 2’s Super Resolution mode typically cost a small, single-digit percentage of GPU overhead in exchange for rendering at a lower internal resolution and upscaling, a trade nearly every reviewer treated as an easy win. DLSS 3’s frame generation, introduced with the RTX 40 series and used in titles like Cyberpunk 2077, added modest latency and a light computational cost while inserting interpolated frames for a net FPS gain. The overhead in both of those earlier generations was treated as background noise next to the frame-rate benefit.
DLSS 5, as implemented in NBA 2K27, breaks that pattern. Because 3D-Guided Neural Rendering is reconstructing lighting, reflections, and crowd geometry from the underlying 3D scene rather than just upscaling a finished frame, the computational cost is far higher and appears to behave like a fixed per-resolution frame-time tax rather than a small overhead layered on top of native rendering. Reviewers who tested both technologies side by side describe DLSS 5’s base cost as something closer to a new rendering pipeline than an optional visual filter, one that only becomes performance-positive once Multi Frame Generation is stacked on top of it. That’s a fundamentally different value proposition than what DLSS 3 buyers experienced two generations ago, and it raises the practical question of who DLSS 5 is actually built for at launch: owners of the RTX 5090 and RTX 5080, who can absorb the hit and still land in playable territory, or the wider base of RTX 5060 and RTX 5070 owners Nvidia has spent two years selling as the “of course you can run the latest DLSS” tier.
Reviewer reaction has been sharper than typical launch coverage
The tone across outlets that tested DLSS 5 independently has been notably more skeptical than Nvidia’s own launch messaging. PC Gamer’s hardware team, after measuring the 73 percent worst-case drop on an RTX 5080, questioned what kind of rig would be needed to justify the hit. GamesRadar’s hands-on coverage concluded that Nvidia appears to be leaning on its multi-frame-gen tools specifically because the underlying neural rendering step can’t stand on its own at playable frame rates without it. DSOGaming’s testing, run across native 1440p and native 4K scenarios with ray tracing, arrived independently at the same 50 to 60 percent range that Digital Foundry and Tom’s Hardware measured, lending the number a level of cross-outlet consistency that’s relatively rare in early benchmark coverage. That consistency is also reflected in a separate GPU-by-GPU verdict that found only 4 of 6 RTX 50-series cards fully proven out for the feature at launch.
Not every reaction has been negative. Beebom’s hands-on impressions described being won over by the visual upgrade once the full DLSS 5 plus Multi Frame Generation stack was running, reporting frame rates averaging around 600 FPS on a high-end configuration once every multiplier was engaged, well beyond what a typical monitor’s refresh rate could display. That split in reception, roughly, image-quality wins on one side and frame-rate-cost concerns on the other, is shaping up to define how DLSS 5 is discussed across the review cycle. Everyone agrees the neural rendering output looks noticeably better in crowd and reflection-heavy scenes. Not everyone agrees the performance cost to get there is proportionate, or that Nvidia’s headline frame-rate claims fairly represent what a player will experience without cranking every generation multiplier available.
Market impact: what this means for RTX 50-series buyers
For consumers weighing an RTX 50-series purchase this fall, the DLSS 5 rollout adds a new variable to an already crowded set of considerations. Buyers eyeing a step up specifically to run the newest visual features now need to budget for the fact that the RTX 5060 and RTX 5070, cards Nvidia has positioned as accessible entry points into its ecosystem, post frame-rate drops in the 42 to 55 percent range with DLSS 5’s neural rendering enabled at native settings, before any frame generation is applied. That’s a materially different proposition than earlier DLSS generations, where even budget-tier cards could enable the newest visual feature with a comparatively small performance cost.
It also changes how buyers should read Nvidia’s own marketing numbers going forward. The 370 FPS figure attached to the RTX 5090 launch messaging is real, in the sense that reviewers were able to reproduce frame rates in that range, but only by stacking the full Multi Frame Generation pipeline on top of DLSS 5’s base neural rendering. Shoppers comparing GPUs on the strength of headline frame-rate claims now have to account for the fact that those numbers assume every available frame-generation multiplier is switched on, a configuration that introduces its own trade-offs in input latency and image consistency that several reviewers flagged as noticeable, particularly in fast camera-cut scenes like instant replays.
Competitive landscape: how AMD and console makers are positioned
DLSS 5’s steep base cost lands at a moment when the upscaling and frame-generation category has become one of the most closely watched fronts in the GPU market. AMD’s FSR line has spent the last two generations trying to close the perceptual gap with DLSS while historically keeping overhead lower, an approach that has resonated with buyers who prioritize a predictable frame rate over the highest possible peak. Sony and Microsoft’s console-side upscaling implementations, including PSSR on PlayStation hardware, have taken a similarly conservative approach to overhead by design, since console hardware doesn’t have the thermal or power headroom to absorb a 50-plus percent frame-rate tax and still hit a target frame rate.
A prior three-way comparison of DLSS 4.5, FSR 4.1, and PSSR 2.0 already found a 115 percent FPS gap between the leading upscaler and the most conservative one, underscoring how differently the major vendors weigh raw frame-rate delivery against visual fidelity. If DLSS 5’s neural rendering cost proves to be a structural feature of the technology rather than something Nvidia optimizes away with driver updates, it opens a real marketing opportunity for AMD to position FSR as the more frame-rate-honest option, particularly for buyers who don’t want to depend on aggressive frame generation to hit playable numbers. It also raises the stakes for how Nvidia rolls DLSS 5 out to future titles beyond NBA 2K27: developers evaluating whether to integrate 3D-Guided Neural Rendering will now be weighing a technology with a publicly documented, independently verified performance cost, rather than adopting it on the strength of Nvidia’s own marketing claims alone.
Historical context: DLSS’s shifting performance math
Nvidia has iterated on DLSS for roughly six years, moving from a first generation criticized for soft image quality, through DLSS 2’s widely praised Super Resolution overhaul, into DLSS 3’s frame-generation era that debuted alongside the RTX 40 series. Each of those transitions came with a consistent underlying promise: enabling DLSS would net the player more performance, not less, with overhead treated as a rounding error against the upscaling gains. DLSS 4 extended that formula further with improved ray reconstruction and broader multi-generation frame insertion.
| DLSS generation | Core technique | Typical overhead vs. native | Primary launch title/GPU tie-in |
|---|---|---|---|
| DLSS 2 | AI upscaling (Super Resolution) | Low, single-digit % cost | Multiple RTX 20/30-series titles |
| DLSS 3 | Upscaling + single-frame generation | Modest, net FPS gain | RTX 40-series launch titles |
| DLSS 4 | Multi-frame generation, improved ray reconstruction | Moderate, net FPS gain with multipliers | RTX 50-series launch |
| DLSS 5 | 3D-Guided Neural Rendering + Dynamic MFG | ~42–73% base cost, offset only by MFG | NBA 2K27, Sept. 3, 2026 |
DLSS 5 represents the first time that formula has visibly inverted at the base-feature level. The neural rendering component costs enough on its own that Nvidia’s entire marketing case now rests on convincing buyers that Multi Frame Generation isn’t a bonus feature layered on top, but a required component for DLSS 5 to be usable at all in demanding scenes. That’s a bigger ask than any previous DLSS generation made, and it’s one the review cycle has been unusually willing to push back on in the week since launch.
What developers and Nvidia are likely to do next
Based on the pattern of reviewer feedback and Nvidia’s own driver documentation, a few near-term developments look likely. First, expect Nvidia to publish updated driver notes or an SDK revision aimed specifically at reducing the base cost of the neural rendering step, since the gap between the marketed 370 FPS figure and the sub-native-without-MFG reality has become a recurring theme in coverage. Second, expect Visual Concepts and other developers evaluating DLSS 5 for future titles to be more explicit in their own marketing about which configuration (with or without Multi Frame Generation) any published frame-rate claim reflects, a transparency step several outlets have already called for directly. Third, AMD is likely to lean into FSR’s comparatively lower baseline overhead in upcoming marketing, framing predictability rather than peak frame rate as its differentiator against DLSS 5.
Further out, expect DLSS 5 support to expand cautiously to additional titles rather than in a broad wave, given how closely reviewers are now scrutinizing the technology’s real-world cost. Expect Nvidia to continue pushing RTX 5090 and RTX 5080 owners as the primary DLSS 5 audience, since those cards have enough native headroom to absorb the base cost and still land at playable frame rates before frame generation is applied. And expect the RTX 4000-series patch precedent, which brought DLSS 5 to Ada Lovelace cards with roughly a 39 percent frame-rate drop according to prior testing, to inform how aggressively Nvidia extends the feature to older hardware generations going forward, since the overhead problem appears to scale down rather than disappear on less powerful GPUs. Buyers weighing whether to jump to current-generation hardware at all can also compare how the RTX 5090 stacks up against the RTX 4090 on raw generational gains before factoring DLSS 5 into the equation.
Timeline: one week of DLSS 5 coverage
The speed at which independent frame-rate data accumulated is itself part of the story. Pre-launch leaks citing testing methodology from Wccftech surfaced around September 1, ahead of the official rollout. Nvidia’s Game Ready driver and the official NBA 2K27 integration went live September 3, alongside Nvidia’s own 370 FPS claim for the RTX 5090. Within 24 to 48 hours, DSOGaming, PC Gamer, PCMag, and GamesRadar had published hands-on benchmark data independently, all converging on frame-rate drops in a similar 40 to 73 percent range. By September 5 and 6, Digital Foundry’s more exhaustive GPU-by-GPU breakdown and Tom’s Hardware’s full RTX 50-series sweep had added further confirmation, and Tom’s Guide’s laptop-specific testing highlighted how scene type (menus, replays, live gameplay) changes the size of the hit. By the one-week mark on September 10, Tweaktown’s deep dive into Multi Frame Generation’s role in recovering frame rate had become one of the most-cited technical explanations of why the base cost is so high. That rapid, cross-outlet convergence on a consistent number is unusual for a brand-new rendering feature, and it’s a large part of why the frame-rate hit has become the dominant storyline around DLSS 5’s launch rather than the neural rendering image-quality improvements Nvidia had hoped to lead with.
Should you turn DLSS 5 on right now?
This launch is also a useful data point for anyone tracking the broader AI chips and GPU pricing landscape heading into the RTX 50 Super refresh. For players on an RTX 5090 or RTX 5080 who prioritize image fidelity in crowd and reflection-heavy scenes, and who are comfortable running Multi Frame Generation to recover playable frame rates, DLSS 5 in its current NBA 2K27 implementation is functional and, per Beebom’s hands-on account, capable of impressive peak numbers once fully stacked. For players on an RTX 5060 or RTX 5070, or anyone gaming at 4K who wants a frame rate close to their native baseline without leaning on frame generation, the current data across six independent test outlets suggests DLSS 5 is not yet the free visual upgrade earlier DLSS generations tended to be. The safest near-term approach, echoed by PCMag’s own recommendation, is to test with frame generation at a lower multiplier or off entirely and compare the actual frame rate and input feel before committing to running DLSS 5 at its highest settings.
Frequently asked questions
What is DLSS 5 in NBA 2K27?
DLSS 5 is Nvidia’s 3D-Guided Neural Rendering technology, which debuted commercially in NBA 2K27 on September 3, 2026, in partnership with developer Visual Concepts. It reconstructs lighting, reflections, and crowd detail using data from the underlying 3D scene rather than simple frame upscaling.
How much does DLSS 5 reduce frame rate in NBA 2K27?
Independent testing from Tom’s Hardware, PC Gamer, DSOGaming, and Digital Foundry found frame-rate drops ranging from roughly 42 percent on an RTX 5060 up to 73 percent in specific scenes on an RTX 5080, with an RTX 5090 typically losing around 51 percent when DLSS 5’s neural rendering is enabled without Multi Frame Generation.
Does Multi Frame Generation fix the performance hit?
It offsets it substantially. Tweaktown measured an RTX 5090 climbing from roughly 50 FPS with DLSS 5 alone to 97 FPS with 2x MFG and 170 FPS with 4x MFG, while Tom’s Hardware recorded 258 FPS on an RTX 5060 using DLSS Quality mode with 6x frame generation. The trade-off is added input latency, which Tom’s Guide measured rising from around 80ms to over 111ms in one test scenario.
Is the Nvidia claim of 370 FPS on the RTX 5090 accurate?
Reviewers were able to reproduce frame rates in that range, but only using the full stack of DLSS 5 neural rendering plus high-multiplier Multi Frame Generation. It is not representative of what the neural rendering step delivers on its own.
Can older RTX 40-series cards run DLSS 5?
Nvidia has extended DLSS 5 support to RTX 40-series (Ada Lovelace) hardware via a driver patch, though prior testing found a roughly 39 percent frame-rate drop on those cards, consistent with the base neural rendering cost seen on RTX 50-series GPUs.
Is DLSS 5 different from DLSS 4 and DLSS 3?
Yes. DLSS 3 and DLSS 4 primarily combined AI upscaling with frame generation and typically carried a modest performance cost relative to their frame-rate gains. DLSS 5’s 3D-Guided Neural Rendering is a more computationally intensive reconstruction step, which is why its base cost before frame generation is applied is significantly higher than prior DLSS generations.
Which GPUs perform best with DLSS 5 in NBA 2K27?
The RTX 5090 and RTX 5080 have enough native headroom to stay in playable frame-rate territory even after DLSS 5’s base cost, while RTX 5060 and RTX 5070 owners are more likely to need Multi Frame Generation engaged to reach comfortable frame rates, particularly at 1440p and 4K.
Will other games get DLSS 5 next?
Nvidia and Visual Concepts have positioned NBA 2K27 as the first commercial implementation, and coverage of the launch suggests Nvidia is treating it as a proving ground before wider rollout. No additional confirmed titles have been named in current reporting as of September 12, 2026.
Related Coverage
- Apple A20 Pro Leak: 3 Outlets Agree, Price Up $200 [2026]
- GTA 6 DualSense Controllers Hit $84.99, PS5-Only [2026]
- Arm's Mali G2-Ultra NX Brings DLSS to Phones [2026]
- NVIDIA RTX 5070 GB203 Rumor: 8,960 Cores Eyed [2026]
- Ryzen 7 9800X3D vs 7800X3D: 14.7% FPS Gap, $150 Price Split [2026]
