NVIDIA KES The 2025 presentation demonstrated its new RTX 50 series graphics cards proudly, with RTX 5090 through PTX 5070 they all offer roughly double the performance of their previous-generation RTX 40 series counterparts, something Nvidia noted in the new Blackwell architecture in its report. However, take a look at Nvidia internal tests shows that the 2x improvement in gaming comes more from the DLSS 4 software package than from bare silicon versus silicon.
Deep Learning Super Sampling, or DLSS, has been a core feature of Nvidia’s consumer graphics cards since the RTX 20 series. The DLSS 4 software suite includes many new and improved artificial intelligence features that improve game performance and image quality over DLSS 3/3.5 last generation. Software and hardware improvements result in updates to DLSS super-resolution scaling, ray reconstruction, and frame generation, a feature where the AI generates one additional frame for every frame rendered.
DLSS 4 moves to multi-frame generation, a process that can generate up to three frames from a single rendered frame. This can effectively increase the speed from 30fps to 120fps by generating and inserting additional frames. But, as with conventional frame generation, it’s really about frame smoothing rather than real performance improvements – user input is still sampled at a much slower rate, and latency issues persist.
DLSS 4, according to Nvidia testing, is better in every way than DLSS 3. Nvidia’s own relative performance landmarks for the RTX 5090, 5080, 5070 Ti and 5070, which compare the new cards to their previous generation counterparts (RTX 4090, etc.), aim to prove this precisely.
All future cards more than double last-gen performance in Cyberpunk 2077, Black Myth: Wukong and Alan Wake 2, all at 4K Max settings with full RT. There is also a D5 Render performance test. However, the fine print reveals that while some DLSS 4 features will be available on 40-series cards and older, when tested, DLSS 3 was used on 40-series cards.
There’s nothing wrong with that; DLSS 4 has not yet been implemented on older cards, and Multi Frame Generation will only be available on 50-series cards thanks to special hardware updates (at least that’s what Nvidia says). But in games where DLSS 4 is not available, the advantage drops to 1.5x or less.
A Plague Tale: Requiem was tested with DLSS 3 on all cards, while Far Cry 6 was tested without DLSS (which it doesn’t support) and showed a marginal improvement of 1.25x per generation.
Any gamer purchasing an RTX 50 series will want to turn all of DLSS 4’s settings and features up to eleven to see real performance gains over the 40 series. When the 50 Series releases starting this month, some DLSS 4 updates will also be rolled out to older cards, improving visuals and performance along with the 50 Series.
DLSS 4’s multi-frame generation is also interesting because of how it deals with input lag. Because of the way multi-frame generation works (after two frames are rendered in the normal way, up to three intermediate frames are created to fill in the gaps), there is some input lag as the generated frames are created without any new input data. . But in this respect it is no worse than an ordinary framegen; Generating multiple frames simply creates intermediate frames faster than generating 40-series frames. There are also Reflex 2technology that seeks to reduce input latency and further help compensate for the extra headroom required by Multi Frame Generation.
The appropriately named follow-up to 2020’s Reflex, Reflex 2, reduces gaming latency by up to 75%, surpassing the 50% reduction of Reflex 1. The standard latency pipeline follows mouse/keyboard input from the CPU to the render queue to the GPU. which then outputs the image after about 58 milliseconds. Reflex is an SDK that aims to better synchronize the CPU and GPU by passing updated inputs from the CPU to the GPU just before they are needed, bypassing the render queue and reducing latency to 28ms. According to Nvidia, 90% of gamers use Reflex in more than 100 supported games.
Reflex 2 performs this improved queuing process even faster than Reflex 1 and adds a bonus layer to its pipeline. “Frame Warp” can perform super-fast mouse movements and, instead of rendering a new angle, simply warp the existing frame in the direction of the input, further reducing input lag by eliminating the need for a full render. The predictive algorithm fills in the gaps at the edges of the new warped frame, and voila; Latency is further reduced to 14ms.
If you’re confused by this voodoo, the simplified explanation is that Reflex 2 further improves input lag and can help overcome the additional frame lag introduced by DLSS 4 multi-frame generation for those who buy an RTX 50-series GPU. For a more in-depth explanation, check out the Nvidia Reflex 2 video below.
Consumers have been waiting for Nvidia’s RTX 50-series on the Blackwell platform for a long time, and its twice the performance figures compared to the RTX 40-series seem incredibly impressive. But while Blackwell may have more memory bandwidth and more powerful ray tracing cores compared to Ada Silicon, the reported real-world performance gap between the two generations is actually due to DLSS 4 software improvements over the last generation. And of course, this is all based on Nvidia’s own tests; Once the RTX 5090 and 5080 arrive for testing sometime this month, we’ll be able to more quantitatively confirm these numbers.
Check back for our full testing of the RTX 50 series after the cards launch sometime this month, and be sure to stay tuned for more updates. Tom’s Equipment‘s Coverage of CES 2025 as the conference continues this week.