Tag: RTX 50

  • CheckMag: DLSS 4.5 Performance Analysis & Visual Quality Insights

    CheckMag: DLSS 4.5 Performance Analysis & Visual Quality Insights

    Key Takeaways

    1. DLSS 4.5 introduces a 2nd generation Transformer model, enhancing image quality through improved temporal stability and sharper details compared to its predecessor.

    2. The Nvidia app now supports DLSS 4.5 in over 400 titles, with new presets (Preset L and Preset M) optimized for different performance levels.

    3. DLSS 4.5 can result in significant performance drops, with frame rates decreasing by up to 44.5% in some games when using the new Preset M.

    4. Users with RTX 20 and RTX 30 series GPUs may experience diminished benefits from DLSS 4.5 due to hardware limitations, making the original DLSS 4 a viable option.

    5. Future updates will include multi-frame generation (MFG) for RTX 50 cards, enhancing frame output dynamically to match monitor refresh rates.


    Nvidia first unveiled DLSS with its RTX 20 Turing series of GPUs. DLSS 1 wasn’t well-known for its image quality because it functioned mainly as a spatial image upscaler and had limited support from games. However, last year marked a significant advancement with the arrival of DLSS 4, which utilized a Transformer model alongside RTX 50 Blackwell GPUs, greatly enhancing upscaled visual fidelity.

    The DLSS 4.5 Update

    This year, Nvidia is rolling out a minor update, DLSS 4.5, which aims to fine-tune DLSS 4 even further. Nvidia claims that the 2nd generation Transformer in DLSS 4.5 provides improved temporal stability, better resolution, and crisper details.

    With the Nvidia app now officially offering DLSS 4.5 upscaling in over 400 titles, we take a detailed look at image quality comparisons between the new 2nd generation Transformer model and its predecessor, along with anti-aliasing techniques like TAA and DLAA.

    Getting Started with DLSS 4.5

    To start using DLSS 4.5, make sure to update the Nvidia app to version 11.0.6.379 or later (it should automatically update on launch if it hasn’t done so already). While it’s not strictly necessary, having the latest Game Ready Driver version 591.74 or newer installed will enhance your experience.

    Developers will take time to natively integrate DLSS 4.5 into their games. In the meantime, you can manually adjust the DLSS version through the Nvidia app.

    DLSS works by rendering at a lower native resolution, which is usually shown as a percentage of the target resolution. For example, the internal render resolutions for a target of 4K would be:

    You also have the option of using DLAA, Nvidia’s AI-based anti-aliasing solution that operates directly at the native resolution.

    New Presets for DLSS 4.5

    Nvidia uses preset settings to choose the appropriate DLSS version for each game based on its rendered resolution, with the new Preset L and Preset M created specifically for DLSS 4.5.

    Preset M is the best option for DLSS Performance and higher modes, including DLAA, while Preset L is optimized for DLSS Ultra Performance. Preset K is the older Transformer from DLSS 4.

    The simplest way to set your DLSS preferences is to choose DLSS Override to Recommended. This will automatically activate Preset M for Performance, L for Ultra Performance, and K for all other modes.

    You also have the option to select your desired model preset from the Super Resolution dropdown in the Custom tab.

    Another method is to choose the Super Resolution Mode to set an input resolution, which will then select the appropriate preset accordingly.

    Once you have chosen your override mode for the game, open the Nvidia overlay (Alt+Z), navigate to Statistics, and choose DLSS in the Statistics view section.

    This will display the model override preset when you press Alt+R while playing a game.

    And that’s all there is to it. Now, let’s dive into some pixel examination.

    Our testing setup consists of an Intel Core i9-14900K processor with 32 GB of DDR5-6400 RAM (32-39-39-80) and a Zotac Gaming GeForce RTX 5090 Arcticstorm AiO GPU, all running on a Gigabyte Z790 Aorus Master motherboard.

    Testing Results with Popular Titles

    For this comparison, we evaluated four popular AAA games: Battlefield 6, Cyberpunk 2077: Phantom Liberty, Doom: The Dark Ages, and Black Myth: Wukong. We compared image quality at 4K native TAA, 4K native DLAA, DLSS 4 Performance, and DLSS 4.5 Performance (Preset M).

    All games were set to maxed-out 4K Ultra settings. Frame generation and ray reconstruction (RR) were disabled where applicable. Notably, RR still utilizes the original Transformer model, so enabling it in-game effectively reverts you back to DLSS 4.

    In Battlefield 6, DLSS 4.5 Preset M shows noticeable visual enhancements compared to the previous Performance mode.

    We observed a general increase in image sharpness, with many previously missing details—particularly in distant objects like mountains and terrain features—now appearing well-reconstructed. However, no significant improvements were seen in direct first-person view.

    That said, these quality upgrades come at the cost of a 7% decrease in average frame rate when compared to the original Transformer model.

    Doom: The Dark Ages takes advantage of the idTech8 engine’s built-in ray tracing, adding depth and gameplay value. The game also supports path tracing with ray reconstruction (RR), but we turned those off for our comparison.

    Immediately, we can see that DLSS 4.5 Preset M provides a noticeably better image than even the native resolution, with details in the dried tree, the statue, and the rocky mound it sits on looking much clearer and sharper.

    However, this improvement comes with a steep 44.5% performance hit, dropping the average frame rate from 175 fps using DLSS 4 Performance to just 97 fps with DLSS 4.5 Preset M.

    Enabling path tracing (PT) with Preset M but without RR significantly enhances lighting quality but further halves the frame rate.

    Despite being over five years old, Cyberpunk 2077’s neon-lit streets are still a great display for Nvidia’s latest tech.

    With DLSS 4.5 Preset M and path tracing on, we noticed a considerable enhancement in scene quality. Reflections, in particular, appear much more realistic with Preset M compared to the pixelated look of the earlier Transformer.

    Distant objects also show a marked improvement in sharpness and cleaner anti-aliasing with Preset M in comparison to TAA, while the average frame rate impact is minimal at around 3%.

    Preset L offers nearly identical image quality to Preset M and a convincingly better picture compared to the older Ultra Performance mode without affecting frame rates.

    That’s quite impressive, considering Preset L has to upscale from significantly fewer native pixels than Preset M.

    Black Myth: Wukong, known for its dense foliage and path tracing effects, pushes Unreal Engine 5 to its limits, making it ideal for any upscaling help.

    At Full RT Cinematic settings, Preset M effectively restores detail at long distances and reduces shimmer effects that were present with the old Transformer.

    However, the overall image can appear a bit oversharpened, with noticeably sharper shadows compared to native TSR 100.

    On average, the frame rate drop with Preset M is about 5% compared to DLSS 4 Performance mode.

    Likewise, Preset L performs well in upscaling from 720p native, reconstructing the image with much clearer quality than the previous Transformer while virtually eliminating performance loss.

    DLSS 4’s transition to a Transformer model from the DLSS 3.x convolutional neural network (CNN) already marked a significant enhancement in upscaled image quality. Nvidia’s new DLSS 4.5 with its 2nd generation Transformer promises even further fidelity improvements, and it mostly delivers.

    Across the games tested, the primary visual impact of DLSS 4.5 is seen in increased sharpness and, at times, enhanced temporal stability, especially in long-distance views and complex lighting situations.

    However, these differences are mostly noticeable when running DLSS Performance or Ultra Performance, as the 2nd generation Transformer is specifically optimized for these scenarios.

    While Nvidia generally recommends Preset M, Preset L is surprisingly effective given the low native resolution it starts from. This can be especially advantageous for users with RTX 4060 or RTX 5060 GPUs (both laptop and desktop) that have limited VRAM.

    Later in the year, Nvidia is set to introduce multi-frame generation (MFG) up to 6x, generating up to five frames for every rendered frame, but this feature will be exclusive to RTX 50 cards. MFG 6x will operate dynamically, producing just as many frames as needed to match the monitor’s refresh rate.

    Conclusion

    Now for the downside. The enhancements of DLSS 4.5 come at a cost. Even with high-end RTX 40 and RTX 50 series GPUs, you will notice a performance overhead, especially with Preset M. This can be quite significant in some games, necessitating the use of frame generation for smoother visuals.

    The RTX 20 and RTX 30 series cards face even more challenges as their Tensor cores lack FP8 acceleration, which the 2nd generation Transformer relies on. While DLSS 4.5 is still functional on these cards, the performance penalty may easily outweigh the slight visual quality increase.

    Ultimately, the advantages of upgrading to DLSS 4.5 are heavily reliant on the specific game, your target resolution and graphic settings, and the RTX GPU you’re utilizing.

    Sometimes, sticking with the original DLSS 4 might still be the wiser option.

     

  • Nvidia DLSS 4.5: 2nd-Gen Transformer Boosts RTX 50, 20, 30

    Nvidia DLSS 4.5: 2nd-Gen Transformer Boosts RTX 50, 20, 30

    Key Takeaways

    1. Nvidia is launching DLSS 4.5 at CES 2026, not the anticipated RTX 50 Super series GPUs.
    2. DLSS 4.5 features a 2nd generation Transformer model that improves temporal stability, reduces ghosting artifacts, and enhances anti-aliasing.
    3. DLSS 4.5 currently upgrades only the Transformer model, with no changes to ray reconstruction or Nvidia Reflex.
    4. Multi-Frame Generation (MFG) technology is enhanced in RTX 50 GPUs, achieving up to 6x frame generation for smoother gameplay, with improved frame pacing and quality.
    5. Nvidia’s app allows users to apply DLSS 4.5 directly to compatible games, but MFG 6x will be available only in Spring 2026 for RTX 50 series GPUs.


    While it might catch some off guard, Nvidia is not revealing the anticipated RTX 50 Super series GPUs today. Instead, the company is presenting a new treat for gamers at CES 2026: the launch of DLSS 4.5.

    Improvements in DLSS 4.5

    The Transformer model in DLSS 4 significantly enhanced the quality of visual upscaling compared to the earlier convolution neural network (CNN) technique utilized in previous DLSS versions. Although the transition to Transformer was generally welcomed, it had its flaws, especially regarding ghosting effects and a drop in temporal stability in certain games.

    Nvidia aims to address these issues with the 2nd gen Transformer in DLSS 4.5 Super Resolution.

    As per Nvidia, the 2nd generation Transformer in DLSS 4.5 offers better temporal stability, which should lead to sharper images, fewer ghosting artifacts, and improved anti-aliasing.

    Some Drawbacks to Consider

    However, there are a few caveats to keep in mind! First, DLSS 4.5 currently only upgrades the Transformer model for upscaling purposes. There are no adjustments to DLSS 4’s ray reconstruction or Nvidia Reflex at this time.

    DLSS 4’s Transformer has some performance costs. Digital Foundry pointed out in their DLSS 4 analysis that the Turing RTX 20 series and Ampere RTX 30 series cards face a substantial performance hit with the Transformer, mainly because their Tensor cores only support FP16 inference natively.

    With the Ada Lovelace RTX 40 generation, Nvidia introduced native FP8 into the 4th gen Tensor cores, which greatly enhances Transformer-based upscaling. The latest Blackwell RTX 50 series GPUs are equipped with 5th gen Tensor cores that support native FP4 inference, making it the optimal choice for running Transformer with minimal performance loss.

    Nvidia suggests that the 2nd gen Transformer model in DLSS 4.5 is five times more compute-heavy and will significantly benefit from FP8 acceleration when used with RTX 40 series cards and newer.

    Thus, it’s reasonable to think that older RTX 20 and RTX 30 series GPUs will incur a relatively larger performance drop when enabling Transformer v2, although the precise amount can’t be determined until proper testing is conducted.

    Frame Generation Advancements

    While Ada brought frame generation thanks to the architecture’s optical flow accelerator, the RTX 50 Blackwell GPUs have taken it further with multi-frame generation (MFG), achieving rates up to four times the base frame rate using a clever feature called hardware flip metering.

    Even though MFG greatly enhanced the smoothness of gameplay and created the sensation of a higher frame rate, it has not been without issues.

    The primary worry has been about increased latency and input lag, much of which can be mitigated with Nvidia Reflex. Another significant concern was the appearance of glaring artifacts, particularly during fast-moving scenes.

    Alongside the 2nd generation Transformer, Nvidia is now rolling out MFG up to 6x on RTX 50 Blackwell GPUs. The company claims that MFG 6x brings better frame pacing and quality aimed at reaching 240 fps, as 240 Hz monitors are becoming more common.

    To facilitate this, DLSS 4.5 incorporates Dynamic Multi Frame Generation, allowing the system to automatically select the appropriate MFG multiplier (up to a maximum of 6x) to match the monitor’s refresh rate.

    According to Nvidia, DLSS 4.5 along with 6x MFG can achieve up to 246 fps on an RTX 5080 at 4K with path tracing enabled in Black Myth: Wukong.

    You won’t have to wait for game developers to implement DLSS 4.5 support in their games. The Nvidia app should let you apply the DLSS 4.5 override directly to any game in your library that supports DLSS starting today.

    Since older RTX cards might see performance penalties, Nvidia is also providing the option to switch between CNN, Transformer v1, and Transformer v2 as needed.

    However, MFG 6x will be available only in Spring 2026 for — you guessed it — exclusively the RTX 50 series GPUs.

    Source:
    Link


     

  • GeForce RTX 50 Laptops Lack ROPs Found in Desktop Models

    GeForce RTX 50 Laptops Lack ROPs Found in Desktop Models

    Key Takeaways

    1. Nvidia’s Blackwell-based RTX 50 graphics cards may have a small portion lacking Render Output Units (ROPs), potentially affecting both desktop and laptop models.
    2. Manufacturing defects are usually caught before shipping, but some issues may slip through, raising concerns about Nvidia’s RMA plan.
    3. The GeForce RTX 5090 and RTX 5080 are likely candidates for the ROP issue due to their use of the GB203 GPU.
    4. The recent earthquake in Taiwan might have damaged up to 20,000 wafers, impacting the supply of RTX 50 laptop variants.
    5. The RTX 50 series is facing supply chain challenges, including scarcity of RTX 5070 cards and performance-related issues, as reported by DigiTimes.


    Even though Nvidia mentioned that only a “small portion” of its Blackwell-based RTX 50 graphics cards are lacking ROPs (Render Output Units), the issue could possibly go beyond just its desktop products. Reports from Heise.de and Hardwareluxx indicate that certain laptop models may also be impacted.

    Production Concerns

    Typically, most defects like these are expected to be caught and fixed during the manufacturing process before the products are shipped. However, it’s possible for some issues to get missed, and we can only hope that Nvidia has a solid RMA plan ready for such situations.

    Potentially Affected Models

    It is still unclear which specific models are impacted, but the GeForce RTX 5090 and RTX 5080 might be likely suspects, as they utilize the same GB203 GPU found in the desktop RTX 5080 and 5070 Ti. Even if all ROPs are intact, the RTX 50 laptop variants face additional challenges. Reports say that the recent earthquake in Taiwan could have damaged as many as 20,000 wafers.

    Supply Chain Issues

    The exact number of these wafers belonging to Nvidia remains uncertain, but the current scarcity of RTX 5070 cards and the delayed launch of RTX 50 laptop variants imply that the company may have experienced some loss in stock. Additionally, a report from DigiTimes reveals that the RTX 50 series is facing some performance and driver-related issues.

    Source:
    Link

  • CyberPowerPC Safeguards Gamers with Full ROPs in RTX 50 GPUs

    CyberPowerPC Safeguards Gamers with Full ROPs in RTX 50 GPUs

    Key Takeaways

    1. Performance Issues: About 0.5% of RTX 5090, RTX 5080, and RTX 5070 Ti cards have 8 fewer Render Output Units (ROPs), causing a performance drop of around 4% on average in games, with some benchmarks showing greater drops.

    2. Customer Support Concerns: Nvidia is directing affected customers to seek replacements from board partners, leaving many unsure of how to identify the problem.

    3. CyberPowerPC’s Commitment: CyberPowerPC is ensuring that customers receive fully operational RTX 50 cards with their pre-built gaming PCs, standing out as the only system integrator making this promise.

    4. Market Availability: Due to stock shortages, buying a pre-built PC remains a viable option to secure an RTX 5090/5080/5070 Ti at a fair price.

    5. Industry Best Practice: CyberPowerPC’s proactive approach to customer satisfaction highlights a valuable practice that other businesses in the industry should consider adopting.


    After dealing with a terrible launch stock that caused the prices of the RTX 50 GPUs to soar and power connectors to melt, the RTX 50 series is now caught in another problem regarding missing ROPs. According to Nvidia, about 0.5% of RTX 5090, RTX 5080, and RTX 5070 Ti cards have 8 fewer Render Output Units (ROPs) than what was promised. This leads to a drop in performance, which Nvidia claims is only about 4% on average in games. However, independent tests indicate that the performance drop can be greater in benchmarks like 3DMark Time Spy.

    Customer Frustration

    Sadly, Nvidia is telling affected customers to look into the issue themselves and reach out to their board partners for a replacement GPU. This situation leaves gamers who aren’t sure how to spot the problem stuck with a less capable RTX 50 GPU, which they likely purchased for much more than the suggested retail price. This is especially true for those who lack the skills to build their own rigs and tend to buy pre-built gaming PCs from companies like CyberPowerPC.

    A Positive Step

    On a brighter note, CyberPowerPC has stepped up to make sure that customers who choose an RTX 50 gaming PC receive one with a fully operational card. As of now, it appears that CyberPowerPC is the only system integrator providing such a promise.

    A Good Move for Gamers

    Putting aside the oddity that a system integrator has to safeguard its customers against Nvidia’s mistake, this is definitely a smart move by CyberPowerPC. With significant stock shortages in the market, purchasing a pre-built PC remains one of the few options to secure an RTX 5090/5080/5070 Ti at a fair price. Therefore, CyberPowerPC’s commitment to giving its customers peace of mind is a commendable practice that other businesses should consider following.

    Source:
    Link

  • RTX 50 Gaming Laptops Delayed Due to Nvidia’s AI Server Focus

    RTX 50 Gaming Laptops Delayed Due to Nvidia’s AI Server Focus

    Key Takeaways

    1. Asus, MSI, and Acer showcased RTX 50 gaming laptops at CES 2025, with pre-orders starting on February 25.
    2. Nvidia has delayed the launch of RTX 50 mobile GPUs, with high-end models now expected in March and mid-range options in April.
    3. There are rumors of performance issues with RTX 50 GPUs, including hardware functionality problems and unfinished drivers.
    4. Nvidia’s focus on AI products is impacting resource allocation, causing delays in the release of RTX 50 laptop graphics cards.
    5. Acer’s Predator Helios Neo 16 AI, featuring up to an RTX 5070 Ti, is set to be available in Q2 2025, supporting the reported delays.


    Laptop manufacturers such as Asus, MSI, and Acer showcased their gaming laptops powered by RTX 50 at CES 2025. The pre-orders for these first RTX 50 gaming laptops are set to begin on February 25. Currently, there are no RTX 50 gaming laptops available in stores.

    Delays in GPU Launch

    A report from DigiTimes has revealed that Nvidia has postponed the release of its RTX 50 mobile GPUs. The high-end RTX 50 laptop graphics cards, including the RTX 5090, are now expected to launch in March. Additionally, the mid-range options like the RTX 5070 are also delayed and are now anticipated in April instead of March.

    Gamers eagerly waiting for the next-generation gaming laptops will have to exercise some patience.

    Performance Concerns

    According to DigiTimes, there are rumors that Nvidia may not be achieving the expected performance levels with the RTX 50 laptop GPUs. Team Green could be encountering issues related to hardware functionality, as the report mentions “screen problems after the hardware is turned on.” This may suggest that the drivers are still unfinished. DigiTimes cites unnamed supply chain sources hinting at these issues.

    These unnamed sources also indicate that the delay of the RTX 50 laptop GPUs is “probably related to Nvidia’s full sprint to AI servers.” This focus on AI products appears to be impacting “the company’s resource allocation,” which in turn is hindering the “debugging efficiency of new products.”

    Resource Allocation Issues

    In simple terms, it seems Nvidia is having a hard time directing enough resources to the consumer segment, which is causing delays in completing the RTX 50 laptop graphics cards. As a result, the release of RTX 50-based gaming laptops may be pushed back by as much as two months.

    While we cannot definitively confirm the accuracy of the DigiTimes report, Acer has announced that its mid-range Predator Helios Neo 16 AI, featuring up to an RTX 5070 Ti mobile GPU, will be available in Q2 2025, which supports the rumors mentioned earlier.

    Source:
    Link

  • RTX 40 GPUs Going Out of Stock as Nvidia Prepares for RTX 50

    RTX 40 GPUs Going Out of Stock as Nvidia Prepares for RTX 50

    Reports about Nvidia reducing the supply of high-end RTX 40 GPUs by up to 50% began in August. At that time, we guessed this cut might be to ready the market for the upcoming RTX 50 “Blackwell” release.

    Supply Cuts on RTX 40 Models

    New information has come from Chinese Board Channels about the discontinuation of several RTX 40 models. The latest details suggest Nvidia is planning to decrease production of the RTX 4070 and higher in the next few months, with current stock likely to run out before the new RTX 50 GPUs arrive.

    According to Chinese Board Channels, the availability of the RTX 4070 and RTX 4070 Super is expected to be limited in Q4’24. Fresh shipments of the new RTX 4070 and RTX 4070 Super models will reportedly continue until December. After that, it is forecasted that the stock of these models at AIBs will be depleted by January, while retail shelves will clear out a bit later.

    Diminishing Availability Ahead

    In simpler terms, the last two months of 2024 are likely to see very limited stock of the RTX 4070 and RTX 4070 Super, and this trend is expected to worsen in the following months.

    Just like the RTX 4070/Super, the RTX 4070 Ti Super and RTX 4080 Super are also being phased out. Information from Board Channels suggests that the existing supply of these GPUs will be gone by November. As a result, consumers might find it extremely difficult to purchase an RTX 4070 Ti Super or an RTX 4080 Super (Available on Amazon) in December.

    Furthermore, rumors state that since the RTX 5080 is anticipated to launch in January, there won’t be any new 80-class GeForce GPUs available for more than a month. This scenario is particularly concerning since the RTX 5080 is likely to cost more than the RTX 4080.

    Focus Shifting to RTX 5090

    Chinese Board Channels report that AIBs currently have no new RTX 4090D cards, which is the China-specific version of the RTX 4090, to offer. This may indicate that Nvidia has ceased production of new RTX 4090 GPUs and is concentrating on the RTX 5090, expected to launch in January with a significant price hike.

    The supply reduction of the RTX 4090 appears to be affecting the US market too, as there are no RTX 4090 cards listed for under $2,000 on Amazon and Newegg at this moment. For example, the MSI Gaming Trio RTX 4090, which was priced at $1,930 in September, is currently selling for $2,130 on Amazon.

    In conclusion, if you were hoping for holiday discounts on upper-midrange and high-end RTX 40 GPUs, it would be wise to start shopping now, as the chance to buy a new RTX 40 card at its original price may vanish in a month.

    Link,Link


    Image 1