At the technology symposium TSMC North America company SK hynix demonstrated its leadership in the field of next-generation memory by introducing a commercial version to the general public for the first time HBM4While competitors like Micron and Samsung are still in the prototype stage, SK hynix is already preparing for mass production HBM4 in the second half 2025 years.
The HBM4 from SK hynix offers impressive specifications: a volume of up to 48 GB on the stack, throughput up to 2.0 TB/s and the input/output speed is at the level 8.0 Gbps. The company's engineers were able to achieve such a high density by using a 16-layer stack architecture connected using technologies Advanced MR-MUF и TSV (Through Silicon Via). This makes SK hynix a pioneer in such complex memory packaging.
At the same time, a new version was demonstrated HBM3E with 16 layers, providing up to 1.2 TB/s bandwidth. This type of memory is planned for use in future NVIDIA AI clusters, codenamed GB300 "Blackwell Ultra", in anticipation of the transition to HBM4 in the architecture Vera Rubin.
In addition to HBM solutions, SK hynix also showcased its latest server memory modules, including RDIMM и MRDIMM, developed on the basis of DRAM standard 1c. They reach speeds of up to 12,500 MB / s, and MRDIMM models already offer throughput up to 12.8 Gbps in 64, 96 and 256 GB capacities. A 256 GB 3DS RDIMM for high-load data centers was also introduced.
It is obvious that SK hynix is confidently ahead of its competitors in the markets HBM and DRAM, through rapid innovation and close collaboration with industry leaders such as NVIDIA.
Company Intel is preparing to launch a new mobile platform Panther Lake, which promises to be the biggest leap in performance in years. The new processors will not only be built on the proprietary process technology 18A, but will also bring significant improvements in the field AI computing, which is especially important against the backdrop of growing interest in neural network applications.
According to fresh data discovered by the user @InstLatX64, the Panther Lake architecture will use Cougar Cove as performance cores (P-Cores) and Darkmont for energy-efficient cores (E-Cores). This change in plans - previously expected to use Skymont - opens up the possibility for Intel to offer excellent performance in the mobile segment.
As for configurations, leaks have revealed several interesting options:
In terms of power consumption, the processors will have maximum PL2 values up to 80 W (PL1 45W), which promises outstanding performance for notebooks and ultrabooks.
One of the most significant innovations will be the fifth generation of specialized AI accelerators (NPU), capable of reaching up to 180 TOPS computing power. For comparison, this is more than twice as much as current Lunar Lake solutions. It also supports memory management LPDDR5X at speeds up to 8533 MT/s и DDR5 to 7200 MT/s, which will further increase the throughput of the systems.
Mass production of Panther Lake is expected to begin in 2026 year, and the first devices will appear on the market already in second half of 2025. At the moment, the processors are being actively tested by Intel partners, which indicates the high readiness of the platform for commercial launch.
In the vastness of Chinese YouTube user Octopus conducted an unusual experiment - he transformed the processor cover Core i9-14900KS into full-fledged water block. Instead of installing a separate cooler, the enthusiast decided to mill it into IHS (Integrated Heat Spreader) a series of channels for cooling liquid, creating a closed water system directly within the processor case.
To implement the project, octppus used a machine CNC and close cooperation with the subscriber. As a result of the modification, neat inlet and outlet channels appeared in the processor cover, and a rubber gasket and screws were used to seal the structure. The entire heat exchanger was connected to a non-standard cooling system: water was supplied from a conventional pails, which served as a reservoir. After each cycle, the liquid was poured into a separate bottle for the accuracy of the tests.
The results were contradictory. With a standard load of 60 W and at idle, the modified IHS provided lower temperatures compared to the classic water block. However, when the pump speed was reduced, the temperature rose sharply, overtaking the indicators of traditional systems. The main reason is the reduced heat exchange area and the lack of channel optimization for uniform heat distribution, unlike well-thought-out industrial solutions.
However, the idea showed that reducing the distance between the crystal and the cooling liquid by 4 times gives a real effect at high flow. However, the risk of damaging the processor with such a modification is extremely high, and the efficiency directly depends on stability coolant flowDespite the success of the experiment, this method remains extremely complex and impractical for most users.
Company Microsoft officially began rolling out three new features based on artificial intelligence for devices Copilot+ PC. The update is distributed via a preview version Windows 11 for April 2025. To access the features, users need to activate the option "Get updates immediately upon release" in the Settings > Update Center section Windows.
One of the main innovations was the service Recall — is a background data capture system that regularly takes encrypted screenshots and stores them locally. The images are indexed by keywords, dates and applications used. To protect privacy, Microsoft has implemented mechanisms Windows Hello to confirm changes to settings, the ability to select exception apps, configure the data retention period, and delete individual snapshots. In a corporate environment Windows E3 administrators will be able to manage Recall via Group Policy.
The second function, Click to Do, works at the window manager level and is activated by a keyboard shortcut Win + Click or swipe right on the touchscreen. The tool automatically suggests functions depending on the context: translation and summary are available for texts, background removal and editing are available for images. Image processing is supported on all Copilot+ PCs, and text functions currently only work on Snapdragon processors, with support for AMD Ryzen AI 300 и Intel Core Ultra 200V planned for the coming months.
The third innovation is an updated search system Windows, which now uses a compact language model that runs directly on the NPU processor unit. The new architecture allows for natural language search by all interface Windows without having to enter exact file names. The feature requires an NPU with a performance of 40 TOPS and provides 70% faster response times compared to searching in Windows 10
New details about future processors have emerged AMD Ryzen Threadripper 9000 и Threadripper 9000 Pro, which promise to be a real breakthrough not only in professional tasks, but also in game scenarios. As reported by Hagal from the forum AnandTechThe company Gigabyte has already released a BIOS update for its motherboard TRX50 Aero D, where support for next-generation processors has been officially added.
The most interesting feature of the new chips will be the introduction of large 3D V-Cache, which will be implemented in configurations with one, two or four stack modules. This will provide a significant expansion of the L3 cache memory volumes - up to 256 megabyte additional cache connected via technology Through Silicon Vias (TSV). This architecture has already proven its effectiveness in the Ryzen 7 5800X3D and 7950X3D gaming processors, and now its capabilities are scaling for high-performance HEDT platforms.
According to preliminary data, the future replacement of the flagship Threadripper Pro 7995WX will combine 384 MB of standard L3 cache (12x32 MB) and another 256 MB of cache memory located in a 3D stack layout. In total, this will give a gigantic volume - 640 megabytes of cache memory, which will be a record figure for general-purpose processors. This solution is designed to significantly speed up the processing of tasks that are sensitive to memory access speed: rendering, simulations, and demanding modern games.
Ryzen Threadripper 9000 Series Processors Will Use a New Socket LGA4844 (sTR5), maintaining compatibility with chipsets TRX50 for enthusiasts and WRX90 for workstations. This approach will allow users to build systems for both heavy multitasking and gaming builds without having to change the platform. It is expected that the new products will receive the architecture Zen 5, which will provide performance gains at the core level and optimized power consumption.
Previous leaks indicated models with core counts from 12 to 96, and now it’s confirmed that some configurations will not only have a huge number of threads, but also an additionally expanded cache for increased performance in real-world tasks. All this data suggests that the Ryzen Threadripper 9000 could be the most balanced solution for both professionals and gamers.
The exact release dates of the processors have not yet been officially announced, but given the appearance of support in the BIOS and the activity of AMD partners, the announcement may take place at the next exhibition Computex 2025.
Forum User Chiphell under the nickname "michelelee" achieved impressive results in overclocking the video card Gigabyte RTX 5070 Ti Gaming OC, equipped with new memory GDDR7 from SK Hynix. Out of the box, the memory worked at 1750 MHz (28 Gbps), but thanks to overclocking, it was possible to stably overclock it to 2125 MHz (34 Gbps), which resulted in a 21% increase in VRAM frequency.
An attempt to flash the BIOS from the AORUS Master model with Samsung memory ended in failure - the card was temporarily "blocked". However, even with the standard firmware, the results were impressive. After overclocking, the video card showed in the test 3DMark Time Spy the final result is 30 426 points against the previous 28 thousand, which corresponds to a productivity increase of approximately 7%.
Importantly, the new SK Hynix memory modules have proven to be on par with Samsung solutions, with no compatibility issues or performance degradation. More models are expected to be available in the coming weeks. RTX 50 series will be equipped with SK Hynix GDDR7 memory, providing NVIDIA with more stable supplies amid component shortages.
Members ROG ASTRAL series noticed an unexpected innovation in the program GPU TWEAK III - a new feature called EQUIPMENT INSTALLATION CHECK. It is designed to check the correct installation of the video card and identify its possible sagging under the weight of the body.
Previously, ASUS did not officially mention this feature in the description of the ROG ASTRAL RTX 50 series. The main focus was on the functions THERMAL MAP for temperature monitoring and POWER DETECTOR+ for power control. However, apparently, with the release of particularly heavy models like the ROG ASTRAL RTX 5090 LC, which weigh more than 3 kg, there was a need for additional protection. The new system monitors the position of the card using a built-in sensor BOSCH SENSORTEC BMI323, operating as an inertial measurement unit (IMU).
In addition, ROG ASTRAL provides triple control of the card's status: temperature, power supply, and position. Thus, the series offers users not only high performance, but also a high level of protection against mechanical damage to the PCIe slot. Against the backdrop of the ever-growing sizes of video cards and their cost, such functions turn into a significant advantage for owners of premium systems.
Details about new professional processors have emerged Threadripper from AMD. While the company has not officially announced the new generation of chips, indirect evidence is still emerging. Thus, Gigabyte released a BIOS update for the TRX50 motherboard that adds support for processors SHP — is an abbreviation of Shimada Peak, the code name for the new Threadripper on Zen 5 architecture.
The F10A BIOS for the TRX50 has restarted the firmware numbering, which may indicate that engineering samples are being prepared for testing. The update was published three days ago, and although there is no confirmation of the start of reviews, the BIOS publication itself indicates a high level of readiness for launch.
Recent leaks have also revealed processor configurations, ranging from models with 12 cores to options with 64 cores, with the possible appearance of a chip on 96 cores, as in the previous Storm Peak series. All new processors will retain compatibility with the sTR5 socket and will have TDP 350W, which will make it easier for professional users to upgrade.
AMD has officially announced a mobile processor Ryzen 5 7533HS, expanding the range of solutions for notebooks. The new product is built on the architecture Rembrandt Refresh and offers users familiar features: 6 cores, 12 threads, base frequency 3.3 GHz and maximum boost up to 4.4 GHz.
Ryzen 5 7533HS comes with integrated graphics Radeon 660M, providing a decent level of performance for modern tasks. The new chip differs from the previously released Ryzen 5 7535H by a slightly reduced maximum frequency (by 150 MHz), which makes it an excellent option for laptops focused on energy efficiency without a serious loss of power.
According to AMD's official database, the processor will be used in the latest Lenovo laptop models. The company has previously released exclusive SKUs for individual manufacturers, and the Ryzen 5 7533HS continues this practice, offering an optimized solution for partner devices. The processor is expected to provide high performance in compact and thin laptops, where the balance between power and battery life is important.
Company Gigabyte officially confirmed the problem thermal gel leaks for video cards of the series GeForce RTX 50. Recall that earlier the user complained that after a month of use his RTX 5080 began to release liquid towards the PCIe connector. In the statement Gigabyte indicated that the problem concerns early parties products that used "a little more gel."
The manufacturer explained that the increase in the volume of the thermal gel was made for reasons of ensuring reliable contact for heat dissipation. However, excess amount the substance resulted in its release beyond the intended area. In this case, Gigabyte argues that appearance, while it may be a concern, has no effect on performance, reliability or life time graphic cards.
Thermogel can withstand temperatures up to 150°C before turning into liquid. However, Gigabyte did not specify whether this problem would be considered a warranty case. Changes have already been made: in the next RTX 50 and RX 9000 series, the company adjusted the volume of thermal gel application to eliminate such incidents.
Users should be aware that the defect only affects early games video cards and should not be repeated in future deliveries. There is no official explanation regarding warranty obligations yet.
