Seagate has officially unveiled Exos M and IronWolf Pro hard drives with 30TB and 28TB capacity, which became the first mass-produced models using HAMR thermal recording technologies, implemented on the platform Mosaic 3+. The new products are now available through Seagate's online store and partner channels, and are priced at $599.99 for 30 TB versions and $569.99 for 28 TB.
Seagate's HAMR drives use laser heating to heat the platter surface 450°C at the recording point, which allows for record-breaking storage density while saving CMR ComplianceEach platter in the 3.5-inch case holds 3 TB, and the rotation speed is standard 7200 / min. At the expense of 512 MB multi-segment cache Provides stable performance under peak loads, especially in real-world scenarios with mixed read and write operations.
In benchmarks, the Seagate Exos M 30 TB demonstrates 279 MB / s in sequential recording and up to 350 IOPS at 4K reading, and IronWolf Pro is even slightly higher. However, 4K writing in QD16 mode drops to 170 IOPS, which is explained by the nuances of HAMR recording. At the same time, the mixed load shows a noticeable advantage of the new models in latency and IOPS, especially when the write cache is disabled.
IronWolf Pro is designed for NAS and commercial use, and Exos M is for cloud data centers, including for AI loads. Both series are designed for 24/7 operation and are compatible with modern storage systems. A standard 48U rack can accommodate up to 36 PB of data, making these drives the most capacious in their class.
After a long hiatus caused by sanctions and billions in losses, NVIDIA announced the launch of new RTX 6000D graphics accelerator, created specifically for the Chinese market. The model is based on the architecture Blackwell, but has been adapted to comply with current U.S. restrictions. According to DigiTimes, deliveries will start in Q2025 XNUMX, and NVIDIA plans to sell between 1 million and 2 million units by the end of the year — to recoup more $10 billion in lost revenue.
The RTX 6000D unveiling came amid Jensen Huang's third visit to China in 2025, where he personally presented the card to key Chinese partners. The RTX 5090D's inclusion in the ban prompted the company to quickly adapt the lineup — RTX 6000D replaces previous solutions and complements H20 and RTX 5090D V2, which are also being prepared for sale in China.
RTX 6000D uses TSMC 4nm process technology, GDDR7 memory with a capacity of up to 1100 GB / s, and is intended to approach the performance of HBM cards. However, GDDR7 shortage in H2025 XNUMX may limit output. Despite this, the Chinese market is extremely important - it accounted for 13% of NVIDIA's revenue before the bans, and the company lost more than $ 7 billion during the period of restrictions.
With AMD trying to get out with the MI308 and PRO R9700, and internal competition from Huawei and Cambricon, NVIDIA maintains its edge thanks to CUDA and technological backlog. If the RTX 6000D launch is successful, it could be the biggest step toward regaining ground in the all-important AI hardware market.
A retro-computer enthusiast with the nickname konkretor has demonstrated one of the rarest x86 processors online — Rise mP6 266, working in conjunction with motherboard ASUS P5A-B on the ALi Aladdin V chipset. Despite its modest popularity, this chip is a unique example of the Socket 7 era, when dozens of CPU manufacturers existed on the market at the same time, competing with Intel and AMD.
The photo shows that the system successfully passes POST and detects Rise mP6-266 with 128 MB of memory. Inside Windows via CPU-Z it is clear that the processor is running at the frequency 200,55 MHz, although the name indicates the denomination "266". This is due to the typical 90s marketing "Pentium-rating" system, where manufacturers indicated the expected performance relative to Pentium, and not the actual frequency.
Rise mP6 was based on the Kirin architecture, manufactured using a 250 nm process technology at TSMC, and consumed less than 9 W. Despite the high voltage (2.83 V), it did not require active cooling. MMX support, Socket 7 compatibility, 512 KB L8 cache, and two 6 KB L2 cache lines made it a competitor to the Cyrix MII and AMD KXNUMX-XNUMX.
The BIOS shows a CMOS error and no keyboard, but the system continues to work, which confirms the functionality of the rare specimen. According to konkretor himself, the chip was purchased as "new warehouse stock from China" — the story is simple, but it has collectible value.
Research company Binarly has found Four dangerous vulnerabilities in UEFI firmware motherboards Gigabyte, which affect more 100 models devices. The problem is related to System Management Mode (SMM) - privileged operating mode CPU, in which the firmware performs system functions even before the operating system is loaded. Access to this memory is carried out through System Management RAM (SMRAM), and if the communication buffers do not pass validation, this allows malicious code to write itself into the protected area.
All four vulnerabilities have been rated as high severity - 8.2 CVSS score:
CVE-2025-7029 — allows arbitrary writing of data to SMRAM via unprotected OcHeader/OcData pointers
CVE-2025-7028 - gives the attacker control over flash memory operations
CVE-2025-7027 - allows using double pointer dereferencing for writing to SMRAM
CVE-2025-7026 - uses the uncontrolled RBX register for similar attacks via the CommandRcx0 function
All vulnerabilities can be exploited local administrator for installation invisible implants in the firmware, which will remain even after reinstalling the system. Attacks are possible both from a running OS and at the boot, sleep or recovery stage. It is important that SMM operates below the OS kernel level, which makes detecting and eliminating such attacks extremely difficult.
Gigabyte already started roll out BIOS/UEFI updates for motherboards on chipsets Intel 100, 200, 300, 400 and 500 series. Updates can be downloaded from the company's official website and installed via BIOS or utility. Q Flash. After flashing, you need to make sure that Secure Boot is enabled - this reduces the risk of unauthorized modification of the firmware in the future.
First tests completed on Geekbench platform AMD Ryzen Z2 Extreme — the most powerful AMD processor for portable gaming systems. It showed not only high performance CPUbut significant increase in graphics thanks to the integrated Radeon 890M graphics card on the architecture RDNA3.5.
Ryzen Z2 Extreme based on architecture Zen 5 + Zen 5C with 8 cores and 16 threads. The chip operates within TDP from 15 to 35 W, making it ideal for portable devices. Of the 8 cores, 3 are high-performance Zen 5 and 5 are energy-efficient Zen 5C. The maximum frequency is up to 5 GHz, with 16 MB L3 cache and support LPDDR5X-8000. GPU Radeon 890M has 16 computing units with frequency 2.9 GHz — this is the maximum for the Strix Point series, and it supports FSR, Fluid Motion Frames, Frame Generation and other functions.
In Geekbench tests on the console MSI Claw A8 with 24GB of LPDDR5X memory, the chip outperformed all other AMD Strix Point series APUs single core performance, and in multi-core tests it approached 10-core Ryzen AI 9 365. This is a huge step up from Ryzen Z1 Extreme, which the new Z2 surpasses by 27% in multi-thread и 26.5% in one stream.
Function Work Graphs, presented by Microsoft as part of DirectX 12 Agility SDK, promises to radically change the approach to completing tasks on GPU - providing asynchronous execution, minimizing the load on CPU and better scalability. However, despite the potential of this technology for next generation games, her Widespread adoption on new consoles from Sony and Microsoft is not yet expected.
This was stated by an AMD insider Kepler_L2, noting that although Next generation hardware and APIs (including GPU from AMD with Work Graphs support) and will be technically ready, most engines and development tools are still focused on the traditional graphics programming model. This means that Engines like Unreal Engine and studios' own technologies will not be able to quickly take advantage of Work Graphs in the coming years - especially during the transition period between generations.
Work Graphs allow GPU perform tasks (eg procedural generation, particle simulation, AI logic) without constant intervention CPU, what can significantly increase productivity и reduce delaysHowever, as the insider emphasizes, The mass use of this feature on PC will begin earlier than on consoles, where the implementation of new APIs depends on the policy of platform holders, support in devkits and adaptation at the middleware level.
For the time being There is little official information about the PS6 and the next Xbox, but rumors point to interesting features. It is expected that PlayStation 6 will come in two versions: home and portable. At the same time Portable Model May Be More Powerful Than Xbox Series S, and will also receive Backwards compatible with PS5However, as analysts note, the level of performance will depend on the adaptation of the games, and in the absence of updates from the developers, restrictions may occur.
Company AMD confirmed, which resumes the export of graphics accelerators MI308 to China, following NVIDIA, whose H20 also got the green light from the Trump administration. AMD officials said their export applications have already been submitted for review. US Department of Commerce и their approval is expected soon.
Ранее restrictions on export of AI chips, introduced under the Biden administration and tightened by the current government, have threatened the supply of high-performance solutions to China. AMD has been assessing possible $308 million in losses from MI800 ban, while NVIDIA was forced to write off up to $ 5,5 billion due to the H20 ban. Both companies have previously criticized the restrictions, saying they harm US global leadership in AI.
AMD said it "welcomes progress in trade talks" and believes Washington's new moves confirm A course for US leadership in artificial intelligence. Against the background of the news AMD shares jump nearly 5%, as does NVIDIA, whose H20 is also returning to the Chinese market. Recall that the MI308 is one of AMD's flagship chips for AI processing, competing with NVIDIA's Hopper and Blackwell in the data center and cloud computing space.
The abrupt change of course in Washington was a consequence Jensen Huang's meeting with President Trump, where the need to simplify export regulations was discussed. NVIDIA's CEO has previously said that "export controls are a failure" and called for global distribution American tech stack, comparing it with dollar hegemony.
Now both companies can count on recovery key source of income, especially in light of the large-scale AI projects being rolled out in China. It may also temporarily stop the growing interest in local Chinese solutions, such as Huawei Ascend or chips from Biren, giving the US a chance to regain lost ground.
Indian Billionaire Mukesh Ambani, the head of Jio Platforms and the country's richest man, has announced the launch of an ambitious project JioPC, the purpose of which is turn almost every television in india into a personal computerThe project is being implemented through the existing Jio ecosystem and has the potential to radically change access to digital technologies for hundreds of millions of users.
JioPC is cloud service, available through Jio Set-Top Boxes. These devices are either already installed by JioFiber subscribers or are available for purchase for $64. The service is located in in testing mode, works on invites, and only requires connection mouse, keyboard and account. After activation, users gain access to pre-installed LibreOffice, and can also launch Microsoft Office via browser.
While JioPC doesn't currently support cameras and printers, Jio Platforms has big plans: with its current subscriber base of over 488 million users, even partial coverage can mean tens of millions of "new PCs" without buying new hardware. Considering that in India only 15% of households have a PCWhereas 70% have a TV, the potential reach is impressive.
According to Counterpoint Research, there are currently about 57 million active users of TV set-top boxes, and although the traditional TV market is shrinking, the popularity of Internet services (OTT, IPTV) is growing. At the same time, the PC market in the country is showing Steady growth for seven consecutive quarters, but still lags behind the US and China due to low incomes and the dominance of smartphones.
Researchers from the University of Toronto have presented a new attack method called GPUHammer, which can change individual bits in GDDR6 video memory on NVIDIA graphics cards without physically accessing the data. Experiments have shown that one corrupted bit is enough to reduce the accuracy of a machine learning model from 80% to less than 1%.
Атака GPUHammer is GPU-analogous to the previously known Rowhammer vulnerability and affects memory cells, causing a flip of adjacent bits. Unlike conventional vulnerabilities, this does not require access to the model or its data - it is enough to divide the same GPU-environment, for example, in a cloud service. Testing was carried out on an NVIDIA RTX A6000 graphics card, but models of the Ampere, Ada, Hopper and Turing architectures are also at risk.
During testing of five neural networks (AlexNet, VGG16, ResNet50, DenseNet161, InceptionV3) a catastrophic decrease in accuracy was observed: the average drop from 71.26% to 0.08%. RAD (Rate of Agreement Degradation) was 0.99 - that is, almost every bit flip is critical.
NVIDIA has already published recommendations for protection: first of all, ECC (Error Correction Code) enableWhile this can reduce performance by up to 10% and reduce available VRAM by 6-6.5%, such measures are critical when working with AI in medicine, finance, or autopilots.
The latest models such as RTX 5090 и H100, are equipped with built-in ECC protection. For other video cards, it is recommended to enable ECC manually via nvidia-smi. GPUHammer is a wake-up call for the industry: GPU memory protection is coming to the forefront in the AI era.
In an unexpected and dramatic turn of events, NVIDIA CEO Jensen Huang after a personal meeting with US President Donald Trump achieved in the White House double relaxation regarding restrictions on the export of chips to China. According to NVIDIA itself, the company has already applied for Resumption of H20 deliveries GPU to China and is counting on fast approval from the American authorities.
This is about H20 - special edition GPU, designed for the Chinese market but hit with tough export restrictions on April 9. The Trump administration then demanded unspecified licensing of suppliesleading to NVIDIA asset write-offs worth up to $5,5 billion by the end of the first quarter of fiscal 2026. The company excluded revenue from China entirely from its forecast, promising to treat any future sales as an “unplanned bonus.”
Now, NVIDIA says the company has received confidence from American officialsthat the license will be granted. This will allow Restart H20 deliveries Chinese clients. In parallel Huang announced a new GPU-RTX PRO line for China - oriented towards artificial intelligence, digital twins, logistics and smart factoriesThe device complies with new regulatory requirements and is designed specifically for the needs of the Chinese market.
Despite problems with supplies to China, NVIDIA's business continues to grow rapidlyAccording to UBS, the company already sees potential for $1,5 trillion in revenue in the data center segment, and during its Q1 2026 report, NVIDIA itself said it was seeing “tens of gigawatts” of AI infrastructure projects around the world.
