The RTX 5090 Founders Edition (FE) hits its 575W total board power limit when running DLSS 5. In tests published by Tom's Hardware on September 2, 2026, the MSI RTX 5090 Lightning Z—a card with two power inputs—drew up to 802W and posted higher average frame rates than the FE in all three games tested. While the FE hit its power ceiling, a differently designed card with a higher power configuration recorded higher average fps.
However, framing this as "it's slower because it has only one 12V-2x6 connector" overreaches the evidence. Both cards use the same GPU chip, but they differ in boost clocks, cooling, board design, and power limits. What the test actually confirmed is that the FE hit its power cap, and that the Lightning Z, capable of drawing more power, maintained higher performance. The causal role of connector count itself remains unresolved.
Testing unofficial DLSS 5 across three games, at 4K, no MFG
The games tested were Cyberpunk 2077, Hogwarts Legacy, and Control. Output resolution was 4K, with DLSS Super Resolution set to Performance and an input resolution of 1920×1080. Graphics settings were maxed out, with path tracing enabled in Cyberpunk 2077 and ray tracing enabled in the other two titles. Ray Reconstruction was used where available.
Multi Frame Generation (MFG) was disabled. This means the average fps figures compared here directly reflect the rendering performance with DLSS 5 layered in—unlike the generated-frame-inclusive numbers NVIDIA showed for NBA 2K27. Testing toggled DLSS 5 on and off under identical settings, providing some of the earliest third-party data to check NVIDIA's pre-announced 50–60% performance overhead claim.
That said, the game integration used isn't an official release. The tool used, OptiScaler_DLSSNR, isn't a feature NVIDIA has publicly shipped—its own developers explicitly label it "experimental" and "unofficially supported." Users must separately obtain NVIDIA's model DLL and inject it into the rendering pipeline of supported games. There's no guarantee this matches the speed or stability of an official integration where game developers tune input buffers and model settings themselves.
Capping out at 575W, while the Lightning Z drew up to 802W
The clearest example came from Control. With DLSS 5 enabled, the FE's average fps dropped from 87.5fps to 50.9fps, while its measured power draw (via PCAT) rose from 563W to 575W. The Lightning Z dropped from 107.0fps to 61.6fps, but its power draw climbed from 691W to 802W. With DLSS 5 on, the Lightning Z's average fps was 21.0% higher than the FE's.
| Game | GPU | DLSS 5 Off | DLSS 5 On | Power (On) |
|---|---|---|---|---|
| Cyberpunk 2077 | Lightning Z | 99.2fps | 60.0fps | 723W |
| Cyberpunk 2077 | RTX 5090 FE | 88.3fps | 51.3fps | 551W |
| Hogwarts Legacy | Lightning Z | 124.2fps | 72.8fps | 720W |
| Hogwarts Legacy | RTX 5090 FE | 121.1fps | 62.8fps | 547W |
| Control | Lightning Z | 107.0fps | 61.6fps | 802W |
| Control | RTX 5090 FE | 87.5fps | 50.9fps | 575W |
In Cyberpunk 2077 too, enabling DLSS 5 pushed the Lightning Z's power draw from 580W to 723W, and the FE's from 482W to 551W. In Hogwarts Legacy, the figures rose from 480W to 720W and from 417W to 547W, respectively. Because the first two games used power figures reported via NVAPI (NVIDIA's management API), while Control used an external measurement device, wattage values shouldn't be simply averaged across titles. Even so, a common trend emerges: with DLSS 5 enabled, the FE climbs close to its rated 575W across all three games.
This power increase aligns with the process DLSS 5 uses to generate lighting and materials for the final image. According to NVIDIA's technical documentation, a single-stage pixel-space diffusion model takes rendered frames, motion vectors, and temporal state as inputs. Unlike traditional super-resolution features that reduce rendering load, this adds Tensor Core processing to the end of the rendering pipeline. This additional workload plausibly reducing available power headroom is consistent with both the underlying mechanism and the measured results.
Normalizing across three games weakens the connector-only theory
Before comparing the two cards directly, it's necessary to normalize how much performance each card retained after enabling DLSS 5. Dividing the on-state average fps by the off-state fps gives the Lightning Z a performance retention rate of 60.5% in Cyberpunk 2077, 58.6% in Hogwarts Legacy, and 57.6% in Control. The FE's retention rates were 58.1%, 51.9%, and 58.2%, respectively.
| Game | Lightning Z vs. FE (On) | Lightning Z Retention | FE Retention | Retention Gap |
|---|---|---|---|---|
| Cyberpunk 2077 | +17.0% | 60.5% | 58.1% | +2.4 points |
| Hogwarts Legacy | +15.9% | 58.6% | 51.9% | +6.7 points |
| Control | +21.0% | 57.6% | 58.2% | -0.6 points |
With DLSS 5 enabled, the Lightning Z's average fps exceeded the FE's by 17.0% in Cyberpunk 2077, 15.9% in Hogwarts Legacy, and 21.0% in Control. Meanwhile, the performance retention rate after enabling DLSS 5 was 60.5%, 58.6%, and 57.6% for the Lightning Z, versus 58.1%, 51.9%, and 58.2% for the FE—gaps of 2.4 points, 6.7 points, and -0.6 points, respectively. These are not consistent.
In Control, the Lightning Z used dramatically more power than the FE, yet its on/off retention ratio actually trailed the FE's by 0.6 points. Conversely, in Hogwarts Legacy, the Lightning Z's retention rate was 6.7 points higher. The Lightning Z, despite its higher power configuration, didn't consistently show a higher retention rate—meaning the extra power headroom alone cannot be identified as the cause of the real-world frame rate gap.
Even with DLSS 5 off, there's a gap between the two cards. The Lightning Z outpaces the FE by 12.3% in Cyberpunk 2077, 2.6% in Hogwarts Legacy, and 22.3% in Control. Its official boost clock is also higher, at 2775MHz versus the FE's 2410MHz, and it features a liquid-cooling setup and a custom board. A gap of up to 21.0% with DLSS 5 on cannot be explained by power connector count alone.
Separating power limits from the 12V-2x6 standard
According to NVIDIA's official specifications, the RTX 5090 FE has a total board power rating of 575W. Its power requirement is a single 600W PCIe Gen 5 cable, or four 8-pin cables connected via the included adapter. The single 12V-2x6 input is adopted to meet this specific product configuration.
PCI-SIG defines the 12V-2x6 standard in PCI Express Card Electromechanical 5.1, replacing the earlier 12VHPWR spec. But what this test's graphs actually measured were frame rate and card power consumption. It didn't measure per-wire current, terminal temperature, or voltage drop, and it isn't a test of the connector's electrical limits. The fact that the card reached 575W primarily indicates the FE's vBIOS-defined board power ceiling.
The Lightning Z starts from a different design baseline altogether. MSI's specification sheet lists two 16-pin inputs, an OC mode rated at 800W, an Extreme mode at 1000W, and a recommended PSU of 1600W. While dual inputs are one factor enabling a board configuration that exceeds the 600W class, its higher performance also stems from its 2775MHz boost clock, 40-phase power delivery, and liquid cooling. This isn't a comparison of "the same card with a different connector"—it's a comparison of two different designs sharing the same GPU chip.
What an official release should test: power limits on the same card
To isolate the effect of connector count, the more realistic test would hold clock speed and cooling constant on the same Lightning Z unit while varying only the power limit—say, between 575W and 800W-plus. Ideally, testers would use identical board hardware and vary only the input configuration, while recording GPU clock and voltage alongside per-pin current and terminal temperature. Measuring frame time and system latency alongside average fps and 1% lows would also reveal what fraction of the added power actually translates into perceptible improvement.
Official game integration is another item that needs verifying. NVIDIA plans to bring DLSS 5 to NBA 2K27, targeting the RTX 50 series. In the official release, developers will tune the model, strength, and masking to suit the game. Whether the same level of performance drop and power behavior seen in these early OptiScaler experiments holds up will need to be re-measured once it becomes available.
Even if a higher power configuration boosts performance, that doesn't automatically mean efficiency or safety improves as well. The Lightning Z drew 802W with DLSS 5 enabled in Control and posted a higher average fps than the FE. The next piece of evidence needed isn't about the number of connectors on the outside of the card—it's a measurement, under identical conditions, of what fps, latency, and temperature that extra power actually buys.