An unusual engineering sample of an NVIDIA graphics card on a chip has been discovered in China. TU116, made in board format PG160 SKU18The special feature of this model is its support PCI Express 4.0 interface, although retail versions based on the TU116, such as the GeForce GTX 1660 and 1660 Ti, officially only support PCIe 3.0. This is a prototype, which, judging by the markings, is in the early stages of development.
According to the utility GPU-Z, the device was identified as NVIDIA Graphics Device with Device ID 10DE 2186, contains 1152 CUDA cores, 4 GB GDDR6 memory from Micron and runs on 128-bit bus with throughput 192 GB / s. Frequencies GPU reach 1005 MHz (base) и 2100 MHz (boost), which may indicate an experimental BIOS. The specified support DirectML, CUDA, OpenCL and Vulkan confirms the full-featured Turing architecture.
3DMark test showed throughput 12.23 GB / s via the PCIe 4.0 bus, which exceeds the limits of PCIe 3.0 x8, and thus confirms operation in PCIe 4.0 ×8 modeThe board itself does not have standard markings on the crystal, but other engineering samples of a similar model used markings D18Q — which corresponds to the TU116 chip revision.
This makes this sample extremely rare: of all the Turing chips, only the TU117STherefore, this card was either used for internal testing of the new bus on Turing or was an experimental attempt at unification with the Ampere platform. Such findings once again demonstrate the diversity of NVIDIA's internal engineering solutions that never made it to mass production.
Investment company Scion Asset Management, which is run by a well-known investor Michael Burry, bet on a fall Nvidia and Palantir sharesAs of September 30, the company acquired put options (that is, bet on the decline in value) of these companies on amounting to more than $1 billion.
Burry fears that the artificial intelligence market is highly overvalued and could soon collapse like the dot-coms did in 2000He had already warned about this through his social media account X (formerly Twitter), posting a cryptic message: "Sometimes the only winning move is not to play all."His statement sounded like a direct signal: he sees an impending crisis and acts in advance.
The cause of alarm was numerous complicated transactions, in which the largest AI companies participate. For example, Nvidia invests in startups, which then use the funds to buy its chips., creating a vicious circle. Such schemes raise suspicions of artificially inflating demand.
Against this background, Nvidia shares have already fallen by 2,2%, and Palantir has lost 6,7% of its value.Palantir CEO Alex Karp responded sharply, calling Burry's criticism "nonsense," but the market has begun to react. Many perceive Burry's actions as warning of a possible collapse, similar to his famous bet against the mortgage market in 2008.
NVIDIA has released GeForce Game Ready 581.80 graphics driver, aimed at laptops with Windows 11 and Windows 10. The update provides optimal support for new gamesIncluding Call of Duty: Black Ops 7 with technology DLSS-4, Anno 117: Pax Romana и Europa Universalis V with the support DLSS Super Resolution.
The new version also fixes bugs and improves performance. In particular, F1 25 were held optimizations when using DLSS Frame Generation. In addition, a serious crash on startup has been fixed. applications on Vulkan on old processors Core 2 Duo and Core 2 Quad, which caused applications to crash on startup.
Driver 581.80 is designed for a wide range of laptop video cards: RTX 50, RTX 40, RTX 30, RTX 20 series, GTX 16 and 10, and even MX and 900M seriesHowever, NVIDIA reminds that Laptop users are advised to check for certified updates on device manufacturer websites., especially with specific configurations or custom BIOS.
Additionally included in the installer new version of NVIDIA App, allowing users Automatically receive driver updates and manage settings.
RDSEED instruction error on AMD Zen 5 processors (specifically Strix Point and Granite Ridge) causes crashes on startup optimized Linux buildsSuch as cachyOS, using the compilation key -march=znver5A temporary fix in the Linux kernel that disables RDSEED support when a problem is detected has caused new crashes: systems cannot load the graphical shell correctly.
CachyOS developer Peter Jung stated that due to a recent kernel patch, "for several users the system no longer enters graphics mode." The reason is that GCC still uses RDSEED, if he sees support for CPUID, despite the fact that the instruction itself in Zen 5 may be faulty without fresh microcode.
AMD engineer Borislav Petkov suggested temporarily do not remove the RDSEED flag, but only issue a warning so as not to break the behavior of the assemblies. He noted: "The compiler should not simply use RDSEED - it should check its availability through CPUID". In turn, Intel expert Thomas Gleixner added: "You get what you compile - RDSEED does not work correctly, and There are no other workarounds other than updating the microcode or rewriting the entire code.".
The issue highlights the risks of using aggressive architectural optimizations sort of -march=znver5Although they provide high performance, microcode errors can completely disable the system, until a fix is released. Microcode updates for the Ryzen 9000 and Ryzen AI 300 are already in the works, but the exact release date has not yet been announced.
An unusual phenomenon has been spotted in China motherboard with LGA1700 socket, aimed not at gamers, but at tasks involving a large number of connected devices. The main difference is 22 USB ports, placed in a tight layout along the entire edge of the board, and one PCIe x16 slot, without reinforced fastening, but suitable for a video card or other expansion cards.
Judging by the configuration, the board supports 12th-14th generation Intel processors. Present four slots for DDR4 memory, two network ports, an audio codec, and a standard set of internal connectors. The main emphasis is on maximum number of USB ports, making the device suitable for simulators, laboratory benches, peripheral control or specific industrial tasks.
The board quickly went viral online thanks to a post on Reddit., where users jokingly refer to "500 USB hubs" and "rocket remotes." Despite the unconventional layout, the design looks neat and is clearly aimed at mass-market peripheral connectivity. There is support for video cards, but the emphasis is on USB, which completely distinguishes this model from classic gaming solutions.
In the third quarter of 2025 DRAM contract prices grew by 171,8% compared to last year, reaching a level that even outpaces the growth rate gold pricesThis explosive growth is due to a sharp increase in demand for memory from AI industries, which entails a shift in production priorities and a reduction in consumer product output. This was reported by CTEE, citing ADATA CEO Chen Libai, who stated that A major DRAM bull market will begin in Q4 2025., and in 2026 a sharp memory deficit.
Memory prices have been steadily rising throughout 2025. In September, they registered a 15-20% increase, and by October, they had reached 50% compared to previous expectations. Now, the effect has accumulated to an extreme level, and this is only the beginning: the retail market is lagging behind the contract market, but is already seeing a doubling of DDR5 prices. For example, the Corsair Vengeance RGB 2x16GB 6000MT/s kit is now $183 on Newegg, while the same kit was selling for $91 in July.
The greatest pressure falls on server RDIMM memory and HBM, but the price increase also applies to regular DDR5, as manufacturers increasingly prioritize deliveries to data centers. This leads to decline in consumer DRAM production and rising prices even on SSDs and hard drives.
Experts warn that this trend will continue for at least four years — this is the exact period for which companies like Samsung and SK Hynix have already signed contracts to supply DRAM to corporate clients.
Production high-performance NVIDIA Blackwell AI chips in the United States became possible not only thanks to the efforts of the company itself, but also due to close cooperation with a number of key Taiwanese suppliersNVIDIA CEO Jensen Huang openly stated this, noting the contribution of companies such as TSMC, Foxconn, Wistron, Amkor и SPILL, in the implementation of the "Made in USA" initiative.
According to Huang, in the nine months since the change of the US administration, a full-fledged production cycle has been established, including release, assembly and packaging advanced chips entirely within the country. The central element of this chain was TSMC factory in Arizona, which is capable of producing the most advanced chips in the world. Meanwhile, Foxconn and Wistron have announced multibillion-dollar investments in American production lines, and Amkor is building a plant advanced packaging, which closes one of the main bottlenecks in logistics.
Huang emphasized that without these partners, it would be impossible to create a sustainable supply chain in the US. He expressed his gratitude. engineers, workers, electricians and builders, which ensured the launch of production as quickly as possible. These efforts demonstrate that the United States can become a full-fledged center of global semiconductor production, but only if international industrial cooperation is maintained.
The record was set on an Intel platform. The build was based on a processor Intel Core Ultra 7 265K (Arrow Lake-S). For the tests, the energy-efficient processor cores were disabled, and the high-performance P-cores operated at a frequency of 2034 MHz.
The motherboard served as the basis for the system. Gigabyte Z890 Aorus Tachyon Ice on the Z890 chipset. The modules used were Corsair Vengeance DDR5.
To achieve this result, the overclocker used liquid nitrogen (LN2) coolingBoth the central processor and the RAM modules themselves were cooled.
Flagship smartphones in 2026 may feature distinct performance differences, even if they are based on the same chipset generation. According to the latest insider information, Qualcomm is preparing two versions of its upcoming top-end processor, the Snapdragon 8 Elite Gen 6. Plans are reported for a "Standard" and a "Pro" version.
Both versions of the chip will be a technological breakthrough, as they are rumored to be manufactured using advanced 2nm TSMC N2P process technologyThis transition from the 3nm process (which underpins the current Gen 5) should deliver an unprecedented leap in power efficiency and raw performance. Both chips are also expected to feature a custom third-generation processor architecture with a new 8-core layout. clusters "2+3+3", which is a departure from the 2+6 design of the previous generation.
The key differences between the models will lie in memory support and graphics capabilities. Sources claim that The Pro version will receive exclusive LPDDR6 support. — the latest RAM standard, which will deliver the highest bandwidth, critical for complex AI tasks and gaming. The standard model, in turn, will be limited to supporting all of the still-relevant, but slower, LPDDR5X standard.
Besides RAM, the two versions will have different specifications GPU. It is assumed that the “Pro” model will be equipped with a fully-fledged new graphics accelerator (conditionally, Adreno 850), while the “standard” version will receive either a cut-down version of this GPU, or an overclocked version of the previous generation (Adreno 840). Both chips will likely switch to a new flash memory standard. UFS 5.0, speeding up application launches and data transfers.
This strategy will allow Qualcomm and smartphone makers to more clearly segment the market. "Pro" devices will become uncompromising ultra-flagships, while "Standard" devices will form the basis of more affordable top-end models. However, insiders warn that cost of producing 2-nm chips will be significantly higher, which will almost certainly lead to another price increase for flagship Android smartphones in late 2026.
Company Sapphire confirmed, which she approved replacing the video card Nitro+ 9070 XT after the recent incident with connector 12VHPWR, which resulted in the device malfunctioning. The user who experienced the problem reported that they contacted customer support and received official confirmation that the card had been accepted for diagnostics and subsequent repair or exchange.
In its appeal, Sapphire specified that the graphics card and the damaged cable must be sent back for inspection. The user is responsible for the shipping costs. delivery и insurance, and all subsequent work is performed by the company. Despite the controversial nature of the breakdown, support responded promptly and agreed to consider the case under warranty, which generated a positive response in the community.
The Nitro+ 9070 XT issue sparked widespread discussion earlier, when a user posted a photo of a burnt 12VHPWR connector. Now, following Sapphire's official decision, the story has taken a positive turn: the manufacturer has not voided the warranty and has confirmed its willingness to fix the issue. The owner is awaiting the final inspection results, but has already noted the professional approach and prompt response from the support team.
