The exhibition Computex 2025 AMD company presented FSR Redstone — a major update to FidelityFX Super Resolution technology that will be released in the second half of 2025. The new stage of FSR development is based on machine learning, and includes three key components: Neural Radiance Cache, Ray Regeneration и ML-accelerated frame generation.
FSR Redstone will be a direct competitor to NVIDIA's DLSS 3.5 and RTX Video Super Resolution. The first feature, Neural Radiance Cache (NRC), is responsible for forecasting and caching indirect lighting based on the analysis of reflected rays. This will significantly increase the realism of global illumination, especially in scenes with dynamic lighting and complex geometry.
The second key technology is Ray Regenerationanalog Ray Reconstruction from NVIDIA. It uses trained neural networks for pixel restoration, which cannot be accurately calculated in real time. This helps reduce noise and artifacts in ray tracing, as well as increase productivity without loss of quality. AMD claims that the algorithm already shows excellent results in benchmarks and is ready for integration into games.
The third component is Machine learning frame generation, which complements existing FSR upscaling algorithms and provides more smooth and clear picture. Unlike previous versions of FSR, Redstone is based on neural network models, which allows us to reach a level of quality and rendering speed previously available only on GPU NVIDIA RTX.
It is important to note that At launch, Redstone will only be available for graphics cards with RDNA 4 architecture., including the Radeon RX 9060 and RX 9070. However, AMD is considering adapting certain features for RDNA 3 in the future, especially in the context of frame generation and neural network upscaling.
NVIDIA has released an update 11.0.4 versions for the NVIDIA App branded application on Windows. The new release adds light theme mode, improves accessibility and expands the list of supported games for automatic graphics optimization. A number of bug fixes have also been made critical errors, affecting recording, overlay and HDR performance.
Users can now switch between light and dark interface themes, and also use the system setting Windows for automatic selection. This function is available in the menu "Settings" > "Features" > "Theme". Also added support Windows Narrator — built-in interface voice technology that improves accessibility for users with visual impairments.
The main practical update is Adding optimal graphics settings for eight new games, including blockbusters and new releases. The list includes: — Assassin's Creed Shadows — Deadlock — Grand Theft Auto V Enhanced — Half Life 2 RTX — Indiana Jones and the Great Circle — InZOI — Monster Hunter Wilds — Split Fiction — The Last of Us Part II Remastered
The release also addresses Driver loading errors, problems with setting the recording bitrate, incorrect HDR playback, lack of in-game overlay on RTX 5070and reset microphone gain after reboot.Additionally corrected bug where Highlights couldn't be turned off, and the overall stability of the application has been improved.
NVIDIA App 11.0.4 is now available through the standard in-app update mechanism and is recommended for all GeForce RTX users, especially those planning to launch new games and using recording, optimization, or HDR features.
AMD Unveils Major Ecosystem Update at Computex 2025 ROCm (Radeon Open Compute), focusing on full support Windows and improved compatibility with Linux. The updates are designed to make it easier for both developers and enthusiasts to run neural network frameworks and AI workloads on Radeon and Radeon Instinct graphics cards.
The main news was the addition of support PyTorch and ONNX-EP under Windows. PyTorch for ROCm is coming in preliminary version in the third quarter of 2025, and ONNX Execution Provider - in July 2025 years. Both solutions were previously limited to Linux platforms, but now developers will be able to work with AMD GPU directly to Windows, including via WSL and HIP SDK.
For Linux, AMD introduces ROCm integration out of the box into the major distributions. In the second half of 2025, support will be added to Red Hat EPEL and Ubuntu, and will also appear for the first time in OpenSUSE. Fedora will continue to be a compatible platform. This will free users from manual configuration and simplify the installation of ROCm via system packages.
Support Windows and integration into Linux distributions significantly expands the potential audience of ROCm. Developers will be able to use local GPU for training models, running inference and optimizing performance, without moving to proprietary ecosystems. It also expects improved compatibility with machine learning libraries and LLM runners such as Llama.cpp, Hugging Face Transformers, Diffusers, and others.
AMD is committed to making ROCm an open and fully-fledged alternative to closed CUDA platforms. Universal support for PyTorch, ONNX and HIP on Windows and Linux — is the key to achieving this goal.
AMD has officially announced Radeon AI PRO R9700 video card, the first model in a new series aimed exclusively at AI workloads and AI-accelerated computing. The new product is built on architecture RDNA4, is equipped with 32 GB of GDDR6 video memory and offers performance up to 1531 TOPS (INT4 Sparse) — 2 times more than the previous generation.
Radeon AI PRO R9700 uses the chip Navi 48 with 64 compute units (4096 stream processors) and 128 hardware AI accelerators. The card has TBP up to 300W, 256-bit memory bus and total FP16 performance 96 TFLOPS, making it particularly effective for running advanced local models.
According to AMD, the R9700 is aimed at running neural networks of the level DeepSeek R1 Distill Qwen 32B Q6, Mistral Small 3.1 24B Instruct, Flux 1 Schnel and SD 3.5 Medium, where 32 GB of video memory becomes the necessary minimum. In practice, the R9700 demonstrates 2x performance compared to Radeon PRO W7800 32 GB, and also outperforms the RTX 5080 with 16GB up to 5x in VRAM-sensitive models.
The special feature of the card is support 4-fold multi-GPU-configurations via PCIe 5.0, allowing you to combine up to 128 GB of video memory - this is enough to load models of the level Mistral 123B and DeepSeek R1 70B, which require more than 110 GB.
The model replaces the old Radeon PRO and WX series and marks transition to the new Radeon AI PRO brand line, which is entirely aimed at developers, researchers and enthusiasts of local AI. Radeon AI PRO R9700 is made in dual-slot form factor with a turbine cooler, and will go on sale in July 2025 through partners such as ASUS, ASRock, Gigabyte, PowerColor, Sapphire, XFX and Yeston.
AMD has officially confirmed the pricing and release date of the graphics card Radeon RX 9060 XTwhich will go on sale 5 June 2025 year. The new product will become the company's main offer in the mid-range segment with two configurations: 8GB for $299 и 16GB for $349. Thus, the RX 9060 XT becomes direct competitor to the $5060 RTX 299 и significantly cheaper than the RTX 5060 Ti, which costs $399 with the same 16GB of VRAM.
Radeon RX 9060 XT is built on the architecture RDNA4equipped 32 computing units, 32 hardware ray tracing units and 64 AI accelerators. The card supports FSR4, DirectX 12 Ultimate and is connected via interface PCIe 5.0 x16. Power consumption ranges from 150 to 182 W, and the Boost frequency reaches 3.13 GHz. Also supported are DisplayPort 2.1a and HDMI 2.1b, which ensures compatibility with modern 4K and 8K monitors.
The junior version with 8 GB of GDDR7 at 32 Gbps provides throughput up to 1 TB/s, making it a good choice for 1080p and 1440p gaming, while the 16GB model will be especially useful for texture-heavy games and long-term use — especially considering it costs $50 cheaper than a 16GB RTX 5060 Ti.
The cards will go on sale at all major AMD partners: ASUS, Gigabyte, Sapphire, PowerColor, ASRock, Acer, XFX, Yeston and others. The initial batches are expected to be in sufficient volume to avoid shortages.
Thus, the RX 9060 XT sets the standard aggressive price point in the up to $400 segment, offering more memory and FSR 4 support at a significantly more affordable price.
The exhibition Computex 2025 AMD has officially unveiled Threadripper 9000 series processors, which translates the Zen 5 architecture into the HEDT and professional workstation segment. The updated platform will go on sale in July and will be available in two directions: Threadripper PRO (WX-series) for workstations and Threadripper X (without PRO) for enthusiasts.
The flagship of the new line was Threadripper PRO 9995WX, equipped 96 Zen 5 cores, 384 MB L3 cache and frequency up to 5.4 GHz. It represents the highest specification chip in the history of the series. All PRO models support 8-channel DDR5 memory, as well as up to 128 PCIe Gen5 lanes, making them ideal for systems with multiple GPU and storage devices.
The PRO line includes 64-core (9985WX), 32-core (9975WX) and 24-core (9965WX) models, but motherboard manufacturers based on sTR5 (TRX50) will also be able to use 16- and even 12-core processors. These configurations make it easier to enter the Threadripper ecosystem without sacrificing connectivity.
For enthusiasts we offer Threadripper 9970X (32 cores) и 9960X (24 cores) — chips without PRO branding, focused on HEDT. They support 4-channel DDR5-6400, have limited PCIe layout (compared to PRO), but maintain frequencies up to 5.4 GHz and high performance in rendering and compilation tasks.
AMD has changed physical layout of crystals in the X-series, removing the remote CCD chiplets. In their place are stubs (dummy dies) to maintain thermodynamic balance. This allows to reduce the cost and highlight the differences between the HEDT and PRO segments.
Pricing for Threadripper 9000 has not yet been announced, but both lines will be available at retail, including the DIY market. AMD has also prepared separate packaging for each series, emphasizing their positioning.
At Computex 2025, AMD revealed the full specifications of the new Radeon RX 9060 XT video cards, which opens the RDNA 4 lineup in the mid-range segment. The model is aimed at high-performance gaming and AI computing acceleration, offering a balanced ratio of power and energy consumption.
The RX 9060 XT is based on 32 RDNA 4 compute units, as well as 32 hardware ray tracing units (HW RT Accelerators) and 64 units for AI operations (HW AI Accelerators). In peak INT4 Sparse acceleration mode, the card reaches performance up to 821 TOPS, which makes it suitable not only for games, but also for generative AI, upscaling and video coding tasks involving neural networks.
Maximum frequency GPU is 3.13 GHz, and the memory capacity is 8 or 16 GB GDDR6 depending on the version. The card is connected via the interface PCIe 5.0 x16 and demonstrates power consumption at the level of 150–182 W, which allows it to be used both in gaming systems and in compact assemblies with a limited thermal budget.
Support DisplayPort 2.1a and HDMI 2.1b provides up-to-date capabilities for connecting high-resolution and high-refresh-rate monitors. The RX 9060 XT is also designed to work with FSR 4 and FidelityFX technologies, including AI-based scaling and spatial enhancements.
So AMD is betting on universal video card, suitable for both modern games and AI workloads. The RX 9060 XT promises to be one of the most interesting offers on the market thanks to the combination of RDNA 4, GDDR6, AI accelerators and an affordable power consumption level.
In the latest version of the GeForce driver 576.52 NVIDIA has fixed a critical bug related to premature power limiting (TGP) on mobile video cards GeForce RTX 50 seriesPreviously, this problem could cause a decrease in frequencies. GPU long before reaching the estimated thermal limit, which negatively affected performance in games and professional tasks.
The error was registered under number 5170771 and was specifically related to laptops. Because of her RTX 50 series graphics cards could enter throttling mode prematurely, reducing the load on the chip and, as a result, the frame rate. This was especially noticeable in long gaming sessions and resource-intensive applications, where every watt of power affects the final result.
NVIDIA confirmed that the glitch has been fixed, and TGP will now be fully operational, corresponding to the passport values. This means that GPU in laptops will be able to maintain high frequencies longer, use the entire available thermal budget and provide stable performance without artificial restrictions.
For users, this means a real boost in performance and increased stability under long-term load. The update is especially important for those using RTX 5070, RTX 5080 and RTX 5090 in laptops, where cooling systems are designed for high TGP and full utilization of resources is important.
The driver update is now available via GeForce Experience or NVIDIA official website, and is highly recommended for installation by all owners of mobile systems with RTX 50 video cards.
The exhibition Computex 2025 SilverStone company presented new generation of retro case FLP02, which became a continuation of last year's FLP01. The case design is made in the spirit of desktop PCs of the late 80s, with a characteristic beige color, angular shape and classic arrangement of the Power, Turbo and Lock buttons. Despite the old-school appearance, the body is designed for the most modern components.
The FLP02 is a full-size mid-tower tower, which can easily fit flagship level video cards, including the long three-fan models. They are supported by a built-in stand in the same color to prevent sagging. They also fit inside 360mm SVO, making the case a full-fledged platform for enthusiasts.
On the front panel of the case, under the magnetic cover, there are modern interfaces: USB Type-C, two USB 3.0 ports, and a combo audio jack. And the Turbo button has an unexpected modern use: it now switches fan speeds, displaying the mode on a red digital display, just like in real retro PCs.
The front panel is equipped with three real 5.25" bays, allowing you to install optical drives, card readers or additional controllers. If there is no need, you can close the compartments with false panels for floppy disks - they are included in the kit. It is also possible to mechanical power interlock via a key - as in old office systems.
The SilverStone FLP02 case will be available in third or fourth quarter of 2025 priced at $220 in the US, the company promises large-scale production given the high demand for the first model, which remains in short supply.
The first details about the upcoming video card have appeared online NVIDIA GeForce RTX 5080 SUPER, revealed by a well-known insider kopite7kimi, which has previously shared precise information about upcoming products on multiple occasions. According to the leak, the video card will receive full chip GB203-450 and an expanded memory configuration, which makes it noticeably different from the original RTX 5080.
The heart of the map will be GPU с 10752 CUDA cores, like the regular RTX 5080, but a different revision of the crystal is used - GB203-450 instead of GB203-400. A new PCB design is also indicated - PG147-SKU35. It is claimed that the energy consumption level will exceed 400 W, which is 40 W higher than the original model. This is due not only to the possible increase in frequencies, but also to the new level of video memory.
The video card will receive 24 GB GDDR7 memory with 32 Gbps of memory bandwidth, which is 2 Gbps more than the RTX 5080, which uses 30 Gbps of memory, bringing the total throughput to 1024 GB / s, which is 6,6% higher. The same 256-bit bus, as in the base model.
In this way, gaming performance can increase by 5-6%, but the key argument in favor of the RTX 5080 SUPER will be the memory capacity. With modern games increasingly demanding more video memory, the jump from 16 GB to 24 GB could be a deciding factor for enthusiasts and 4K monitor owners.
While NVIDIA has not announced a release date., and the video card was not demonstrated at the Computex 2025 booths. However, according to sources, the RTX 5080 SUPER could be released in late 2025 or at CES 2026. It will be the first card with 24 GB of GDDR7 among solutions in the sub-flagship class, and it will certainly interest both gamers and professional users.
