The exhibition Computex 2025 Enermax has demonstrated one of the most advanced cooling systems for high-performance workstations — Cirrus Mk1 immersion unit, designed for a thermal load of up to 3300 W. The new product uses two-phase cooling with immersion in dielectric liquid, which allows for efficient heat removal from components at the RTX 5090 and Threadripper levels.
The key feature of the Cirrus Mk1 is the use of low boiling point liquids, which evaporates quickly when heated GPU and CPU. The resulting steam condenses in the upper part of the tank, and then the heat is transferred to the external heat exchanger through the second thermal circuitThis dual-circuit solution ensures stable cooling and refrigerant isolation, which is especially important when using ozone-free PFAS cooler with low greenhouse effect.
The demo configuration included AMD Threadripper 7960X and four RTX 5090, installed on the board Gigabyte TRX50 AI-TOP. The entire system weighs 70 kg and takes up a decent amount of space: 700 mm in height, 500 mm in diameter. It is powered by two units Platimax II 2400 W each, and the total power consumption reaches 4800 W - exceeding the capabilities of conventional sockets, which makes special electrical wiring necessary.
Enermax emphasizes that the Cirrus Mk1 is designed not for gamers but for professional use — rendering, simulations and AI processing. The company claims that the coolant is 100 times more efficient than air and can significantly reduce carbon footprint of the system. Thanks to this approach, users will be able to achieve stable high performance over a long period of time.
The cost of the system itself is about $50,000, and that's without taking into account the components. In reality, such a workstation will cost no less than $100,000, but for studios and research centers it may be a worthwhile investment.
The exhibition Computex 2025 MSI has announced a new initiative to improve the ease of PC building — PinSafe Design technology, which should save enthusiasts from cuts and pricks when installing a motherboard. The essence of the innovation is in the complete redesign of the back side of the PCB: instead of the usual sharp pins, it uses flat contact pads, not protruding from the board.
While this may seem like a minor improvement at first glance, it was a solution that was rarely used before due to concerns about component reliability and durability. Now MSI claims that PinSafe Design not only increases comfortbut improves protection against electrostatic discharges, reduces the risk of short circuits and makes the system physically more resistant to external influences.
The first motherboard to feature the new approach will be B850MPOWER — a model for AMD processors, focused on overclocking and professional use. The MPOWER series is traditionally intended for enthusiasts, and MSI has integrated into it not only flat pins, but also proprietary functions: EZ Dashboard with power, reset, clear CMOS buttons and diagnostic LEDs, as well as 2 DIMM slots to improve stability DDR5 overclocking.
Also worth noting is that new soldering method PinSafe may also be useful in terms of cooling and compactness of components. The absence of protruding metal elements on the bottom side makes the board compatible with denser layouts and reduces the chance of damage when installed in a case.
Thus, MSI takes another step towards safer, more convenient and modern PC assembly. If PinSafe lives up to expectations, we can expect this technology to spread to other series in the future.
At the Foxconn conference, Computex 2025 NVIDIA CEO Jensen Huang said the company no longer relies on Moore's Law and achieves unprecedented success in the field of artificial intelligence thanks to its own developments. According to him, doubling the power of chips is now achieved not by reducing the technological process, but through architectural and packaging innovations, such as TSMC's CoWoS and proprietary interface NV Link.
Huang pointed out that Moore's Law, which states that chip performance doubles roughly every two years, stopped working in modern conditions. Reducing the process technology no longer brings the same gains as before. Instead, NVIDIA is betting on new generation crystal layout, tight integration of components and scaling via NVLink, allowing the combination of GPU into unified computing systems at the rack level.
This strategy allowed the company to move to 6-month release cycle for new architectures, while the industry is accustomed to two years. Vera Rubin and Blackwell architectures Ultra were announced within half a year of each other, and NVIDIA is already considering switching to four-month food roadmap, which was previously considered impossible.
All of this means that NVIDIA is planning to roll out AI solutions even faster—and not just by increasing the power of individual GPU, but also through a complete restructuring of the approach to hardware development. The company continues to move forward confidently, without slowing down even in the face of restrictions, sanctions and competition.
The exhibition Computex 2025 Thermal Grizzly has introduced an unusual product with a characteristic name der8enchtable — a modular, open platform for building and testing PCs. Developed in collaboration with the legendary overclocker Roman 'der8auer' Hartung and ElmorLabs, the new product is aimed at content makers, testers and enthusiasts, for whom quick access to components and flexibility in system configuration are important.
der8enchtable is an open source platform, which combines the functions of a case, an assembly station and a debug panel. In the center of the design is horizontal PCB tray, around the perimeter of which there are stands for the motherboard. In addition, there are many interfaces built into the edges: power connectors, USB, SATA, microSD, control buttons and fan and pump connection areasSome of these functions are usually assigned to the case, but here more complete control over the entire system is implemented.
The frame itself is made of metal guides and racks, allowing the installation of massive Coolers, Radiators and AIO Systems. Located at the end support "gate" mechanism, which holds graphics cards or other PCIe cards upright without putting any strain on the slot. This is especially true for massive GPU last generation.
In addition to technical functionality, the design provides for D-RGB backlight and active PCB zone, which makes der8enchtable not only a working tool, but also a bright demonstration platform. The product looks unusual, but well thought out: every detail is tailored to the convenience of quick assembly, acceleration and content shooting.
Expected that details about price and release date will be announced later - for now, der8enchtable is in the final prototype stage, but has already attracted the attention of professionals at the exhibition.
Company Seasonic is preparing a new line of Prime power supplies, in which a built-in will appear hardware protection against overheating and melting of 12V-2×6 and 12VHPWR cables. A prototype demonstrating the technology was presented at the exhibition Computex 2025, and the final models will go on sale in first quarter of 2026.
The new protection system includes temperature and current sensors, installed in the power connectors near the power supply case, microcontroller and firmware, which manage their interactions. Seasonic says this will allow real-time tracking overload or local heating of cables, and if the permissible values are exceeded, activate the built-in protection against overheating and overcurrent with automatic system shutdown.
The company cites one of the main reasons for cable overheating: imbalance of loads on power lines, which occurs depending on the video card operating scenario. This can cause local current surges and overheating of individual cable cores. This is especially true for video cards with a connector 12V-2×6, including models RTX 4090, 5090 and the upcoming RTX 50 SUPER line.
In addition to hardware protection, Seasonic will offer external indicator, which, when connected to a PC, will notify the user about cable overheating in real time. However, if the user is not at the computer, the unit itself will turn off the system at critical values, which can prevent damage video cards worth thousands of dollars. However, there is a risk of losing unsaved data or damaging the drive if the session terminates abnormally - and this is a potentially negotiable trade-off.
New Prime Series PSUs Expected to Get More Expensive, as protection will require additional components, firmware, and quality control. However, for owners of flagship graphics cards, this could become the de facto standard in 2026.
At Computex 2025, ASUS announced ROG Strix ACE XG248QSG — a unique monitor for professional eSports with refresh rate 610 Hz (in overclocking mode) and response time 0,1 ms. This is the world's first display based on the new Super TN-matrices, specially designed for ultra-low latency and maximum motion clarity.
ASUS claims the monitor delivers input lag total 0,8 ms, What's on 56% lowerthan other 600Hz displays. Thanks to the technology ELMB 2 (Extreme Low Motion Blur 2) delivers a three-fold improvement in brightness and ghost suppression performance over the previous generation, making the XG248QSG one of the fastest and most responsive solutions on the market.
The monitor is equipped with 24,1-inch panel with a resolution of 1920×1080, support VRR (Variable Refresh Rate), certification Display HDR 400 and function low input lag, which is critical for eSports disciplines. The device also received tripod thread, allowing it to be installed in non-standard configurations, and the reduced footprint provides more freedom for the mouse and keyboard.
Other features include proprietary software DisplayWidget Center for managing OSD settings and profiles using the mouse, as well as a set of ROG Gaming AI AI Features, which automatically optimize the image and parameters depending on the game scenario.
The XG248QSG is aimed at gamers competing in high-level games where every millisecond can make all the difference. The monitor is expected to be available in Q2025 XNUMX.
At Computex 2025, NVIDIA CEO Jensen Huang confirmed that The company's share of the Chinese AI market has fallen from 95% to 50%, as a result of strict export restrictions imposed by the United States. According to him, The company is losing $50 billion in market share, and competition from Chinese manufacturers, especially Huawei, is intensifying every month.
The sanctions began during the Biden administration and were aimed at restricting the export of AI accelerators such as H100 and A100to prevent them from reaching "hostile countries." However, according to Huang, These measures turned out to be strategically wrong, because instead of isolating them, they encouraged the development of their own solutions inside China. NVIDIA's main rival is now Huawei with its accelerators Ascend 910B and 910C, which are already used in the Tencent, Baidu and ByteDance ecosystems for AI inference tasks.
Moreover, Huawei presented CloudMatrix 384, its first cluster system for AI training, which in China is already being called an analogue Blackwell GB200 NVL72 from NVIDIA. This suggests that the country is actively building own AI infrastructure, and if export restrictions continue, NVIDIA risks being completely forced out of the Chinese market.
Huang said that "The world cannot allow China to build its own AI ecosystem", otherwise the US will lose its global technological leadership. Despite attempts to adapt - for example, with the release of a limited version of the accelerator H20 for China — NVIDIA has already lost a significant portion of its revenue. Even though it plans to ship chips this year B40, they no longer compensate for the outflow of customers to Chinese brands.
New details have emerged about the upcoming portable xbox console, which will be based on an APU from AMD, and not on the ARM architecture, as previously assumed. Reputable insider KeplerL2 confirmed that the device will receive Power efficient custom APU AMD, specifically designed for the mobile format. This solution will ensure full compatibility with the current Xbox ecosystem and Windows.
Earlier, the network discussed the possibility of some future Microsoft devices switching to ARM, which would theoretically give energy consumption advantage, especially in a portable form factor. However, such a move would jeopardize support for the current Xbox game library, as the ARM architecture is incompatible with x86 games without additional emulation. The move away from ARM confirms Microsoft's strategy to expand the Xbox gaming presence on maximum number of platforms.
Microsoft is expected to test the market first through branded Windows-devices, developed in collaboration with OEM partners. The first such console, codenamed Project Kennan is being developed jointly with ASUS and will be released before the end of 2025. The device will be powered by Windows with the proprietary Xbox shell, reminiscent of the interface of a stationary console. In addition to AMD, the line of partner devices may use Intel Lunar/Panther Lake chips, Qualcomm X Elite 2 and even NVIDIA N1, but Microsoft's main console will remain AMD-powered.
So Microsoft is betting on wide compatibility and flexibility, ditching ARM in favor of proven AMD solutions, which is especially important for gamers who want full access to the Xbox Game Pass, cloud services and local games - even in portable format.
The exhibition COMPUTEX 2025 Kingston Technology presented a large-scale exhibition called Kingston Future City, focusing on memory and storage technologies for AI, industry and aeronautics. The concept is based on the motto Built on Commitment, highlighting Kingston's commitment to innovation, performance and reliability.
The new one was in the spotlight Kingston FURY Renegade G5 Solid State Drive, one of the fastest PCIe 5.0 NVMe SSDs on the market. It offers sequential read speeds of up to 14 MB/s and write speeds of up to 14 MB/s, while supporting capacities up to 000 TB. This solution has been demonstrated in action on ASUS PCs and is designed for gamers, creators, and AI workloads.
For industrial and server applications, Kingston introduced DC3000ME - NVMe SSD U.2 format, optimized for data centers. It maintains stable speeds of up to 14 MB/s when reading sequentially and up to 2 IOPS in random operations. In addition, RAM was introduced Server Premier DDR5, working in tandem with the DC3000ME in GIGABYTE's server AI solutions.
Kingston is actively promoting AI applications: together with Taiwan Intelligent Robotics Company, a system was demonstrated Multi-modal Inspection Robot, using DC3000ME drives and FURY Renegade DDR5 memory for real-time data analysis. Also in the area Innovative Creators Lab showed their collaboration with RTET, a student team that developed a rocket with a data storage system based on FURY Renegade G5, capable of collecting and transmitting telemetry directly in flight.
In addition, Kingston showed updated models DataTraveler Exodia S flash drives, memory cards Canvas Plus, as well as DDR5 memory modules with speeds up to 8800 MT/s and up to 96 GB of storage, underscoring the company's ambition to serve the most demanding industries, from manufacturing to aerospace.
The exhibition Computex 2025 company Phison demonstrated an impressive storage system for AI workstations - a configuration based on 32 SSD with Phison E28 controller, which combined devices into a single array via an expansion card Apex Storage X16 Gen5. The demonstration used workstation based on AMD Ryzen Threadripper PRO 7985WX, ASUS WRX90E-SAGE SE motherboard and three Apex PCIe devices.
The test in CrystalDiskMark recorded sequential read speed of 113 MB/s (601 GB/s) и write speed of 104 MB/s (628 GB/s), making this array one of the fastest solutions demonstrated in the consumer and professional segments. A Phison representative specified that Windows Kernel was limiting performance, and in other OS you can expect even better results.
Phison PS5028-E28 is the latest PCIe 5.0 controller, built on TSMC's 6nm process technology and using a quad-core architecture. One SSD based on it provides up to 14,8 GB/s read and up to 14,0 GB/s write, and the maximum capacity can reach 32 TB per drive. The demonstration used 32 M.2 2280 drives, evenly distributed across three X16 cards.
Each card Apex Storage X16 Gen5 supports up to 16 M.2 drives with full bandwidthusing 84 lines via Microchip PM584 switch. The addressable capacity is also declared to be up to 128 TB per card, which, when using three boards, theoretically allows us to achieve 384 TB in a single array - depending on the SSD configuration.
Phison positions the E28 as solution for enthusiasts, creators and workstations, capable of providing instant access to data and supporting AI algorithms in real time. Given the growing demands for input/output speed when working with LLM and 3D rendering, such arrays may become the standard in the high-end segment by 2026.
