NVIDIA DLSS 5 performance revealed for RTX 50 GPUs, with huge frame rates in NBA 2K27
NVIDIA has shared the first performance numbers for DLSS 5 running on GeForce RTX 50 series graphics cards, giving PC gamers an early look at what the company’s next-generation AI upscaling and frame generation technology can deliver. The first game shown with DLSS 5 support is NBA 2K27, and the results point to extremely high frame rates across 1080p, 1440p, and 4K when paired with the latest RTX 50 GPUs.
The company is positioning DLSS 5 as its most advanced and complex AI model yet. It is designed to improve image quality, boost performance, and give developers more control over how visuals are presented in their games. At the same time, NVIDIA says it is continuing to optimize the technology, with major performance improvements already achieved since its first public demonstrations.
One of the most important details NVIDIA clarified is the difference between native DLSS 5 support and modified DLSS 5 implementations. In games that officially support DLSS 5, developers decide how the technology is used. They can tune the visuals according to their own artistic direction, ensuring that players experience the game the way the creators intended.
That is different from what some players may have seen online, where modders use developer tools to change DLSS 5 parameters manually. Those custom comparisons may be interesting, and modding remains a major part of PC gaming, but they do not necessarily reflect the official visual target chosen by the original game developers.
For native DLSS 5 games, the experience for most players should be simple. Instead of exposing every advanced setting, games are expected to offer a basic on/off toggle. Developers will handle the tuning behind the scenes. However, as with previous upscaling and frame generation technologies, the PC community will likely find ways to customize and modify settings for those who want more control.
NVIDIA also noted that games using Streamline should be able to integrate DLSS 5 more easily. This gives developers access to the necessary tools while allowing them to decide how the final image should look.
The first DLSS 5 performance tests were conducted in NBA 2K27 with ray tracing enabled, the Ultra preset selected, Multi-Frame Generation set to 6x mode, and DLSS Super Resolution running in Quality Mode. The results show very high performance across the RTX 50 lineup.
At 1080p, even the entry-level GeForce RTX 5060 delivered more than 250 FPS. The rest of the RTX 50 series reached anywhere from over 300 FPS to more than 500 FPS. These are impressive numbers, although 6x Multi-Frame Generation may be unnecessary at lower resolutions unless the player is using an extremely high refresh rate monitor.
At 1440p, NVIDIA recommends the RTX 5070 and above for DLSS 5 in NBA 2K27. In this resolution, the RTX 5070 achieved more than 260 FPS, while the RTX 5090 pushed beyond 500 FPS. That makes 1440p gaming with ray tracing and ultra settings look highly achievable for high-refresh displays.
At 4K, NVIDIA recommends the RTX 5080 and RTX 5090. The RTX 5080 delivered an average of 233 FPS at 2160p, while the RTX 5090 reached around 370 FPS on average. Those numbers suggest that even demanding 4K gameplay with ray tracing can run at extremely smooth frame rates with DLSS 5 enabled. In some cases, players may be able to reduce Multi-Frame Generation from 6x to 4x and still enjoy a strong 240Hz experience on flagship hardware.
DLSS 5 is currently focused on the RTX 50 series because it is built to take advantage of the latest Tensor Core architecture. NVIDIA says the model runs on Tensor Cores and uses FP8, a format also available on RTX 40 series GPUs. However, the company’s official priority is RTX 50 support because older hardware may not deliver the desired experience, especially without the latest Multi-Frame Generation capabilities.
The technology is demanding. NVIDIA says DLSS 5 can have an average performance cost of around 50% to 60%, which explains why Multi-Frame Generation is used broadly across the RTX 50 family to offset that impact. Even so, DLSS 5 does not require other DLSS features as a prerequisite. It can work independently and can be enabled with a native raster renderer.
Latency is another key point. According to NVIDIA, DLSS 5 does not create the same kind of systematic latency associated with traditional frame generation because it does not need to wait for the next frame. Any added latency comes from the cost of running the AI model on the current frame.
NVIDIA also said that developer-defined masks or grouped visual elements do not create performance differences. The model does not scale based on the value inside the buffer, and the different DLSS 5 model options do not have separate performance costs.
Optimization appears to be a major focus for NVIDIA. When DLSS 5 was first shown earlier in 2026, the demo required two RTX 5090 graphics cards. Within two months, NVIDIA improved the model by around 50%, allowing it to run on a single RTX 5090. By July, the model had improved by nearly 2.5x, and by August, NVIDIA said it had improved by 5x, making it capable of running across the RTX 50 GPU family.
These improvements came from a mix of hardware tuning, software kernel optimization, and co-design work. NVIDIA says the current DLSS 5 model is smaller, faster, more realistic, and gives developers greater control than the version first demonstrated earlier in the year.
The company also says optimization will continue at a faster pace. Future updates are expected to make DLSS 5 run faster while improving image quality even further. That means the first performance results in NBA 2K27 may only be the beginning of what the technology can achieve.
In short, DLSS 5 is shaping up to be one of NVIDIA’s biggest upgrades for AI-assisted gaming. With RTX 50 GPUs, Multi-Frame Generation, ray tracing, and DLSS Super Resolution working together, games like NBA 2K27 can reach extremely high frame rates at demanding settings. For players with high-refresh monitors, especially at 1440p and 4K, DLSS 5 could become a major reason to consider NVIDIA’s newest graphics cards.NVIDIA DLSS 5 Explained: Backward Compatibility, Performance, AI Integration, and Developer Control
NVIDIA DLSS 5 is shaping up to be more than a simple upgrade to the company’s existing image upscaling technology. Based on the latest developer-focused details, DLSS 5 introduces a different approach to image creation, giving game studios more flexibility over how scenes are generated, styled, and optimized.
One of the key points clarified is that DLSS 5 does not become slower just because developers use more masks, groups, or object-specific settings. In other words, a game can have multiple masked areas with different values and visual rules applied to each one without causing the DLSS model itself to scale in cost.
The way it works is that the model reads a screen-space buffer. This buffer contains the values created by the developer, and those values can represent different parts of the scene or different rendering intentions. How that buffer is created depends on the game engine, so there may be a small performance cost on the engine side when preparing the data. However, once the information reaches the DLSS 5 model, the workload does not increase based on how many values are inside the buffer.
This is important for developers because it means DLSS 5 can be used as a highly flexible graphics tool without automatically creating a performance penalty every time a studio adds more visual control.
Another major question is whether DLSS 5 will work with older games. The answer is yes, but only if developers choose to integrate it. DLSS 5 can be added to older titles because the integration process relies on familiar rendering data, such as the color buffer and motion vectors. Games that already support NVIDIA’s Streamline framework could be suitable candidates, but the final decision remains with each game developer or publisher.
That means DLSS 5 backward compatibility will not be universal by default. Instead, it will depend on whether studios decide that updating older games is worth the time and resources.
DLSS 5 is also designed to work alongside other NVIDIA AI technologies without conflict. For example, NVIDIA Reflex 2 operates separately and comes into play after DLSS in the rendering pipeline. Because of that, the two technologies do not interfere with each other.
The same applies to AI-based texture compression. Neural texture compression works inside the renderer, while DLSS operates later using the final screen-space buffer. Since these systems handle different parts of the graphics pipeline, they can function independently. This gives developers more freedom to combine DLSS 5 with other rendering and latency-reduction technologies.
A big part of DLSS 5 is the amount of control it gives developers. Studios can decide how detailed they want their object or scene classification to be. Some may choose broad categories, such as indoor and outdoor areas. Others might separate gameplay sequences from cutscenes. More advanced implementations could even apply different settings to specific objects, materials, or asset types.
For example, developers could potentially treat foliage, furniture, metal surfaces, or other scene elements differently depending on the artistic goal. NVIDIA appears to view this system as another tool in the game development pipeline, similar to other rendering techniques that require artists and engineers to decide how best to apply them.
This approach makes DLSS 5 less of a simple toggle and more of a creative rendering system. It allows studios to shape the final image based on artistic direction rather than relying only on raw rendering power.
One of the most interesting details is how DLSS 5 differs from earlier DLSS technologies. Previous versions of DLSS focused on reconstruction. They used sparse input data to rebuild an image toward a known target, often described as a ground truth. That is why earlier DLSS versions are commonly associated with image reconstruction and upscaling.
DLSS 5, however, is described as fundamentally different. Instead of only reconstructing an image based on a fixed ground truth, it moves toward image generation. In this model, the artist’s direction becomes the new target. The goal is not simply to recreate a traditionally rendered frame, but to help generate the image that developers and artists want players to see.
This could have major implications for future game graphics. By giving artists more control over the final output, DLSS 5 may help studios push visual quality beyond the limits of traditional compute and VRAM budgets. In simpler terms, it may allow games to look more ambitious without requiring every detail to be rendered in the most expensive way.
For players, DLSS 5 will likely appear differently depending on how each developer implements it. At launch, it is primarily described as a developer tool. That means when gamers enable DLSS 5 in a supported title, they will see the experience as the developer intended.
There may be more user-facing options in the future, such as presets created by developers, but that will depend on individual games. Some studios may offer simple on and off controls, while others may choose to expose more visual modes. Ultimately, the player-facing settings are expected to be controlled by the game developer.
NVIDIA DLSS 5 is scheduled to become available on September 3 in NBA 2K27. As more games adopt the technology, we should get a clearer picture of how developers use its AI-powered generation tools, how it performs in real gameplay, and how much visual improvement it can deliver across different genres.
For now, DLSS 5 looks like a significant step forward for AI-driven game rendering. It is not just about higher frame rates or cleaner upscaling. It is about giving developers a new level of control over how games are created, optimized, and presented to players.
