Nvidia’s newest upscaling technology just failed its first real-world stress test. Independent benchmarks published by Club386 on September 3, 2026 show that enabling DLSS 5 on a GeForce RTX 5090 can cut frame rates by nearly half while pushing the card’s power draw right up against its board limit. The story landed the day before Nvidia’s own official DLSS 5 launch, and it complicates the company’s pitch that neural rendering is a free performance upgrade.
The numbers are specific and, for a flagship card that already costs thousands of dollars, uncomfortable. Club386 tested a GeForce RTX 5090 Founders Edition and an MSI GeForce RTX 5090 Lightning Z across Cyberpunk 2077, Hogwarts Legacy, and Control, and found frame rate drops in the 39% to 49% range once DLSS 5 kicked in, alongside power draw increases that pushed the Founders Edition to its 575W thermal graphics power limit. This is a DLSS 5 performance drop story, not a marketing win, and it’s spreading fast across hardware coverage sites.
What Club386’s DLSS 5 Benchmarks Actually Found
Club386’s test rig ran the same three games twice per card: once with Nvidia’s existing upscaling stack, once with DLSS 5 switched on. The pattern held across every title. Cyberpunk 2077 dropped 42% on the Founders Edition and 39% on the Lightning Z. Control dropped roughly 42% on both cards. Hogwarts Legacy, run at 4K with ray tracing and DLSS Performance mode, was the worst case: the Founders Edition lost 49% of its frame rate and the Lightning Z lost 42%.
Power draw moved in the opposite direction. In Cyberpunk 2077, the Founders Edition climbed from 482W to 551W, a jump of about 14%. In Hogwarts Legacy, it went further, from 417W without DLSS 5 to 547W with it enabled. In Control, the card was already close to its ceiling and DLSS 5 pushed it to 575W, which is the card’s rated TGP limit. Reviewers at Tom’s Hardware and XenoSpectrum ran overlapping tests and reported the same directional pattern, though their exact figures diverged slightly from Club386’s numbers, which is typical when different labs use different driver builds and test passes.
RTX 5090 Power Consumption Under DLSS 5: The Numbers
The RTX 5090 Founders Edition ships with a single 12V-2×6 power connector and a 575W board limit. That ceiling is central to why DLSS 5 hurts the Founders Edition more than the Lightning Z, a custom card MSI built with a higher power ceiling and extra connectors. When a workload needs more power than the connector or the firmware allows, the card doesn’t crash. It throttles clocks to stay inside its limit, and frame rate pays the price.
A separate Cyberpunk 2077 test at 4K with DLSS Quality mode illustrates the mechanism at the frame level. Before DLSS 5, the test scene ran at 63 fps, drawing 562.0W at 71°C. After DLSS 5, the same scene ran at 42 fps, a 33% drop in rendering speed, while power climbed to 609.2W, temperature ticked up to 72°C, and VRAM usage crossed 15.5GB. The card was doing more work, burning more power, and delivering fewer frames, all at once.
Other test scenarios point the same direction. On an RTX 5070 Ti, one benchmark run showed frame rate falling from 71 fps to 35 fps with DLSS 5 neural rendering enabled, a 51% reduction. A separate RTX 5090 test recorded a drop from 91 fps to 50 fps, a 45% reduction. Across every card and every scene tested so far, DLSS 5 costs somewhere between a third and half of the frame rate it’s supposed to help deliver.
DLSS 5 Frame Rate and Power Draw, Game by Game
The table below pulls together the confirmed per-game figures from Club386’s test pass across both RTX 5090 variants.
| Game | GPU | Frame Rate Drop | Power Before | Power After |
|---|---|---|---|---|
| Cyberpunk 2077 | RTX 5090 Founders Edition | 42% | 482W | 551W |
| Cyberpunk 2077 | MSI RTX 5090 Lightning Z | 39% | n/a | +14% draw |
| Hogwarts Legacy (4K, RT, DLSS Perf.) | RTX 5090 Founders Edition | 49% | 417W | 547W |
| Hogwarts Legacy (4K, RT, DLSS Perf.) | MSI RTX 5090 Lightning Z | 42% | n/a | n/a |
| Control | RTX 5090 Founders Edition | 42% | 563W | 575W (TGP limit) |
| Control | MSI RTX 5090 Lightning Z | 42% | n/a | n/a |
Note the pattern in that last column. Every RTX 5090 Founders Edition result that reports a “power after” figure lands close to or exactly at the card’s 575W limit. That’s not a coincidence, and it’s the strongest evidence that power headroom, not raw compute, is what’s capping DLSS 5 performance on the reference card.
A Second Data Point: FPS Drops Across Different Test Scenarios
Beyond Club386’s three-game test, other outlets ran their own isolated scenarios. The results line up with a broader trend: DLSS 5 in its current form trades roughly a third to half of available frame rate for its neural rendering pass, regardless of which RTX 50-series card runs it.
| Test Scenario | FPS Before | FPS After DLSS 5 | Drop |
|---|---|---|---|
| Cyberpunk 2077, 4K DLSS Quality | 63 fps | 42 fps | 33% |
| RTX 5070 Ti, neural rendering test | 71 fps | 35 fps | 51% |
| RTX 5090, separate scenario | 91 fps | 50 fps | 45% |
| Hogwarts Legacy, RTX 5090 FE | 121-124 fps | 63 fps | 49% |
Four different tests, four different games and cards, and the drop never falls below a third. That consistency is what turned this from a single outlet’s odd result into a story picked up across the hardware press within a day.
Why DLSS 5 Costs So Much Compute
DLSS 5 is not a simple resolution upscaler in the way earlier DLSS generations were. It runs a full neural rendering pass, generating substantially more of the final image through a trained network rather than through traditional rasterization plus a lightweight AI sharpening step. That approach can produce results existing techniques can’t match, but it also means the GPU has to run a much heavier neural workload on top of whatever the game engine is already doing. More compute per frame means more power drawn per frame, and more heat to dissipate at the same clock speed.
This is the tradeoff Nvidia is betting on: accept a real-time performance cost now, in exchange for image quality and frame generation gains that earlier DLSS versions couldn’t reach. Nvidia’s own DLSS product page and the official DLSS 5 launch on September 3 leaned on multi-frame generation to offset the raw rendering cost, claiming hundreds of frames per second in favorable scenarios. Club386’s benchmarks measured the cost side of that equation directly, and it’s steeper than Nvidia’s marketing suggested.
The 12V-2×6 Connector Problem Nvidia Hasn’t Solved
The RTX 5090 Founders Edition’s single 12V-2×6 connector has been a point of scrutiny since the card launched, and DLSS 5 is the first workload demanding enough to expose its ceiling clearly. When a card needs more sustained power than its connector and firmware will allow, the GPU boost clock steps down to stay within spec, and that shows up as a frame rate drop even though the silicon itself could theoretically do more work.
Custom cards like the MSI RTX 5090 Lightning Z, built with extra connectors and higher power ceilings, consistently posted smaller percentage drops than the Founders Edition across Club386’s tests. That gap between reference and custom cards is a market signal in its own right: buyers chasing DLSS 5 performance may need to look past Nvidia’s own reference design toward board partners with beefier power delivery, adding cost on top of an already expensive GPU generation.
Market Impact: What This Means for RTX 5090 Buyers
RTX 5090 pricing has already been under pressure this year as AI demand pulls supply away from gaming channels, a trend shattered.io covered when RTX 5090 prices doubled toward $5,000 earlier in 2026. A benchmark story showing the flagship card losing up to half its frame rate under Nvidia’s newest headline feature doesn’t help that pricing narrative. Buyers paying a premium for the top-end card now have to weigh whether DLSS 5’s image quality gains are worth a performance hit this large, or whether they’re better off sticking with DLSS 4 until Nvidia tunes the feature further.
There’s also a power supply angle retailers and system builders will have to account for. A card that can spike to its 575W limit under a single feature toggle changes the math on recommended PSU headroom, case airflow, and thermal design for anyone assembling a system specifically to run DLSS 5 titles at launch. For more on how GPU pricing and power trends are shaping build decisions this year, see shattered.io’s ongoing hardware coverage.
Historical Context: DLSS Has Always Traded Compute for Quality
Every DLSS generation has asked the GPU to do more work in exchange for a better image or a higher effective frame rate. Early versions ran a comparatively light upscaling pass. Later versions added full frame generation, inserting synthetic frames between rendered ones, which raised the on-screen frame count but added its own compute overhead and input latency considerations. DLSS 5’s neural rendering approach pushes that tradeoff further than any prior version, replacing more of the traditional rendering pipeline with a trained network rather than supplementing it.
What’s different this time is the size of the hit relative to the hardware it’s running on. Previous DLSS upgrades generally improved effective frame rate compared to native rendering, even accounting for their own overhead. Early DLSS 5 benchmarks show the opposite in raw terms: turning it on measurably lowers frame rate compared to running the game without it, before any frame generation is factored back in. That’s a harder sell to reviewers and buyers than “your existing frame rate, but with a sharper image.”
Competitive Landscape: Where AMD and Intel Stand
Nvidia remains the only GPU vendor shipping a neural rendering feature at this scale, and that’s part of why the DLSS 5 performance drop is drawing so much coverage. AMD’s FSR lineup and Intel’s XeSS have followed Nvidia’s upscaling playbook closely in past generations but haven’t announced a directly comparable full neural rendering pass of their own. That gives Nvidia a features lead on paper, but this week’s benchmarks show the risk of being first: DLSS 5 is the reference point every future neural rendering feature from AMD or Intel will get measured against, and the bar it’s setting right now is “significant performance cost for better visuals,” not “free upgrade.”
For readers weighing upscaling technologies more broadly, shattered.io’s existing DLSS vs FSR comparison covers how Nvidia’s and AMD’s approaches have differed generation over generation, a useful baseline for understanding how big a departure DLSS 5’s compute cost represents.
The DLSS 5 Leak Timeline: From NBA 2K27 to Official Launch
This isn’t the first time DLSS 5 has generated headlines this year. The feature first surfaced through a leaked DLL pulled from an early NBA 2K27 build, which modders quickly got running on unsupported hardware, including a community patch that forced DLSS 5 to run on RTX 4090 cards Nvidia never officially supported for the feature. Those early leaked builds already hinted at heavy performance costs on unsupported silicon. Club386’s benchmarks confirm that even on the card DLSS 5 was actually built for, the RTX 5090, the performance cost is real and significant, not just an artifact of running the feature on hardware it wasn’t designed for.
Nvidia’s Response So Far
Nvidia has not issued a detailed public rebuttal to the Club386 benchmarks as of this writing. The company’s official DLSS 5 messaging around its September 3 launch focused on multi-frame generation output and image quality gains rather than raw rendering cost, and Nvidia’s own published performance figures for the RTX 5090 emphasize scenarios where multi-frame generation is active, which can mask the underlying rendering overhead this week’s benchmarks isolate. Until Nvidia addresses the specific power and frame rate figures reviewers are publishing, buyers are left comparing marketing claims against independently measured numbers that tell a less flattering story.
5 Predictions for DLSS 5 Through Late 2026
- Driver optimization will narrow the gap. Nvidia has a strong track record of shaving overhead off new rendering features within a few driver releases, and DLSS 5’s first weeks are likely to be its worst-performing.
- Board partners will market power headroom as a selling point. Expect MSI, Asus, and other partners to lean harder on custom power delivery in RTX 5090 marketing now that Club386’s data shows a real performance gap versus the Founders Edition.
- Per-game DLSS 5 profiles will emerge. Rather than a universal on/off toggle, expect Nvidia and developers to ship tuned presets that balance neural rendering quality against frame rate on a title-by-title basis.
- Scrutiny of the Founders Edition’s single-connector design will grow. If DLSS 5 adoption widens across more titles, the 12V-2×6 bottleneck story is likely to resurface every time a new demanding feature ships.
- AMD and Intel will frame their next upscaling updates around efficiency. A competitor pitching “similar quality gains, lower power cost” would land well against this week’s coverage, giving both companies a clear marketing angle heading into 2027 GPU launches.
Should You Enable DLSS 5 Right Now?
For most RTX 5090 owners, the honest answer is: test it per game before committing. The frame rate and power figures above aren’t universal constants, they vary by title, resolution, and which upscaling mode within DLSS 5 is active. Players chasing maximum smoothness in fast-paced competitive titles may find the tradeoff isn’t worth it yet, while players prioritizing visual fidelity in single-player, cinematic games may accept the hit gladly. Anyone running a Founders Edition card should also keep an eye on power draw specifically, since it’s the card most likely to hit its ceiling and throttle under DLSS 5’s heavier workload.
The broader lesson from this week’s benchmarks is that neural rendering, as currently implemented, is not a free lunch. It’s a genuine tradeoff between image quality and raw performance, and until Nvidia optimizes further or ships hardware built specifically to absorb that cost, buyers and reviewers will keep measuring the gap between what DLSS 5 promises and what it currently delivers.
Frequently Asked Questions
What is the DLSS 5 performance drop on RTX 5090?
Club386’s benchmarks found frame rate drops between 39% and 49% across Cyberpunk 2077, Hogwarts Legacy, and Control when DLSS 5 was enabled on both the RTX 5090 Founders Edition and the MSI RTX 5090 Lightning Z.
Why does DLSS 5 increase power consumption?
DLSS 5 runs a heavier neural rendering pass than earlier DLSS versions, requiring more GPU compute per frame. That extra workload draws more power, pushing the RTX 5090 Founders Edition close to or at its 575W board power limit in several tested scenarios.
Is the RTX 5090 Founders Edition or the MSI Lightning Z better for DLSS 5?
Benchmarks show the MSI RTX 5090 Lightning Z posting somewhat smaller frame rate drops than the Founders Edition in some tests, likely because of its higher power ceiling and additional connectors, which give it more headroom before it has to throttle.
Does the 12V-2×6 connector limit DLSS 5 performance?
Reviewers, including Tom’s Hardware, point to the Founders Edition’s single 12V-2×6 connector as a likely factor. When the card’s power draw approaches its rated limit, the GPU throttles clocks to stay within spec, which shows up as a frame rate drop.
Will Nvidia fix the DLSS 5 performance issues?
Nvidia has not issued a detailed public response to these specific benchmarks. Past DLSS launches have seen driver-level optimizations that reduced overhead over time, so improvement is plausible, but no fix or timeline has been officially confirmed.
Which games were tested for DLSS 5 performance?
Club386’s benchmarks covered Cyberpunk 2077, Hogwarts Legacy, and Control. Other outlets ran additional isolated tests on cards including the RTX 5070 Ti, with results showing similar drops in the 33% to 51% range.
Should I enable DLSS 5 on my RTX 5090 right now?
It depends on the game and your priorities. Test DLSS 5 per title before committing to it, since the frame rate and power tradeoffs vary. Competitive players chasing smoothness may want to wait, while players prioritizing visual quality in slower-paced games may find the tradeoff acceptable already.
