Latest News - Page 372

NVIDIA has officially unveiled its beefed-up GB300 "Blackwell Ultra" NVL72 AI server, with its new GB300 AI GPU featuring 50% more performance over GB200, and a larger pool of 288GB of HBM3E memory.

NVIDIA's new GB300 "Blackwell Ultra" AI GPUs will continue the AI domination led by GB200, with memory capacity increased through using new 12-Hi HBM3E memory stacks, and even more compute power for AI workloads. NVIDIA is also coupling GB300 with the latest Spectrum Ultra X800 Ethernet switches (512-Radix).

The new NVIDIA GB300 NVL72 rack-scale solution has 1.5x more performance than an NVIDIA GB200 NVL72 AI server, as well as increasing Blackwell revenue opportunities by 50x for AI factories compared to AI servers built on the Hopper GPU architecture.

NVIDIA's GTC conference has officially kicked off, and Phison, one of the world's leading companies in controllers for NAND flash memory chips, has unveiled what it's describing as an affordable path to supporting a 1 trillion parameter AI model.

The race to the first 1 trillion parameter AI model is nearing completion, at least according to Phison, which predicts we will see the first 1 trillion parameter AI model before 2026. Phison has briefly outlined the progression of AI models, with a 69 billion parameter model unveiled in 2023 with Llama 2, then a big jump to the 405 billion parameter Llama 3.1 model in 2024, and then the 671 billion parameter DeepSeek R3 model in 2025. Keeping on this same trajectory, Phison expects the world's first 1 trillion parameter model will be unveiled before the end of 2025.

With the increase in the size of these AI models, the hardware needed to train and support them will also need to be increased, which is why Phison is looking at ways to reduce operational costs, and the company has a solution. Phison expects the operational costs of a 1 trillion parameter model will be approximately $3 million in raw GPU power, but the company has outlined a plan to alleviate the cost down to just $100,000. How? A combination of SSDs, Phison's aiDAPTIV+ software, and an NVIDIA GH200 Superchip.

The world's first Large Language Model (LLM) training laptop is being showcased at GTC 2025, and it's a creation between Phison and Maingear.

Phison Electronics, a world leader in NAND flash technology, has teamed up with Maingear to bring its aiDAPTIV+ software stack to Maingear's ML-series laptop, creating the first AI laptop PC capable of Large Language Model Operations. The new laptop, which is being demoed at GTC 2025 with an RTX 4090 laptop GPU, will feature an NVIDIA RTX 5090 laptop GPU when it is released for sale. As for its capabilities, Phison has announced it will be able to train and inference LLMs that are up to 8 billion parameters, while also having support for fine-tuning, which enables users to inject their own data into the AI model.

Why does this laptop exist? Phison is setting out to reduce the cost of running AI models by offloading the processing, which is currently on high-cost GPU memory, to low-cost NAND flash storage, and by doing this the operational cost of the AI model shrinks, enabling larger models to be trained on cheaper devices. By utilizing the power of the company's aiDAPTIV+ software, users will now be able to customize an AI model with their own data, all within a device that can fit in your backpack.

Phison is expanding its aiDAPTIV+ software to the NVIDIA Jetson platform, enabling enhanced generative AI inference at the edge of robotics development.

Phison has created a specialized SSD for the NVIDIA Jetson platform and has integrated its aiDAPTIVE+ software to enable Fine-Tune training, which allows users to take an AI model and add their own data into the model. The idea behind this integration is to reduce the performance cost of running the AI by offloading some of the processing from the GPU VRAM over to the SSD.

Phison explained to me that with these upgrades to the Jetson platform they are able to provide up to 14x improvement on Time to First Token (TTFT), which is the time it takes for an AI to provide a response to a question. Moreover, these changes also increased the token length by 8x, which is the length of the response generated by the AI. Ultimately, Phison's improvements made to Jetson have unlocked the platform in a way that enables faster responses with a greater length, enhancing the quality of the AI chat.

AMD's Ryzen 7 9800X3D now appears to be in plentiful stock at its MSRP at major retailers in the US, finally.

If you recall, when the Ryzen 9800X3D came onto the market last year, it received a positively glowing reception.

In our review, we called the processor the "undisputed gaming champion" and also noted that it's a "capable all-rounder across a variety of workloads." (Remember that its predecessor, AMD's Ryzen 7800X3D, while also a great gaming chip, didn't do so well with other workloads).

Another image of what's purportedly a reference RX 9070 board from AMD has surfaced online, though we should obviously be very careful about accepting the pictured GPU at face value.

VideoCardz noticed the image (via Olrak29 on X), which comes from a forum member over at Chiphell, a Chinese source of plenty of hardware leaks, but far from the most reliable one it must be noted.

The picture shows a pretty plain black graphics card with two fans, and it's presumably a prerelease sample of what would be a 'Made by AMD' version of the RX 9070.

AMD's RX 9070 GPU may witness more robust levels of supply after April, according to info that comes direct from a Chinese graphics card maker.

Yeston, who you've likely heard of, is a sizeable force in the GPU market over in Asia, and the firm made the revelation on X, as you can see above.

The wording leaves a slight question mark for us, though, as the arrival of a more stable level of supply could happen 'after' April begins or 'after' April is over. We assume it's the latter, as that'd make more sense, but to us, it could be read as the former, too.

PNY has announced the new CS2342 M.2 2230 NVMe, a new PCIe Gen4 storage solution with one of the fastest read and write speeds for a 2230 SSD - 7,300 and 6,000 MB/s, respectively. With fast transfer and read speeds, PNY says it's perfect for portable PC gaming handhelds like Valve's Steam Deck and ASUS's ROG Ally.

Available at retail in late March, in 1TB and 2TB capacities, it offers an upgrade option for gamers, creators, and enthusiasts with devices that feature an M.2 2230 slot for storage or expansion. The 2TB capacity option makes the most sense for gamers as game install sizes in 2025 aren't getting any smaller, and most affordable portable gaming handhelds are sold with 512Gb or 1TB capacities.

Of course, the big selling point here is speed. Having the capacity to store more games is one thing, but having titles load in a few seconds and not run into any throughput issues is key for a seamless experience. PNY notes that the CS2342 has been designed for seamless integration when installed in a Steam Deck, ROG Ally, or MSI Claw handheld.

NVIDIA GTC 2025, the AI conference for developers, is underway this week. The industry is gathering to explore and share what's next in AI, robotics, and accelerated computing. KIOXIA, the inventor of flash memory, a key technology that helped usher in the AI era of computing, is at GTC 2025, showcasing its latest SSD innovations and AI solutions.

This includes the recently announced KIOXIA LC9 Series NVMe 122.88 TB SSD, which is built for the AI era. The 2.5-inch form factor SSD is the first in KIOXIA's line-up built with eighth-generation BiCS FLASH technology, a 2Tb QLC die, and CBA (CMOS directly Bonded to Array) technology. This high-capacity PCIe 5.0 SSD is tailor-made for AI systems training large language models (LLMs) and cloud computing.

At GTC 2025, KIOXIA (located at booth #1811 on the show floor of the San Jose McEnery Convention Center) will also present a live demo of its groundbreaking KIOXIA All-in-Storage ANNS with Product Quantization (KIOXIA AiSAQ) technology. This technology leverages SSD storage instead of expensive DRAM to improve AI RAG (Retrieval Augmented Generation) performance dramatically.

Half-Life 2 RTX is a remaster and remake project from Orbifold Studios, a team of 100 modders, built with NVIDIA's impressive RTX Remix platform. In a nutshell, RTX Remix takes every scene and location from Valve's original Half-Life 2, transforms every material and surface into modern physically-based rendered (PBR) objects, and adds a suite of RTX technologies and AI tools to draw from, with the result being a fully ray-traced or path-traced version of a classic game that is now twenty years old.

Half-Life 2 RTX is a project we've been following closely for a few years, with early looks showcasing stunning and transformative results. With a fully playable demo now available for all GeForce RTX owners that includes two fully playable chapters from the game (Ravenholm and Nova Prospekt), we were given early access to the demo to get our first proper hands-on look at Half-Life 2 RTX.

Running on the GeForce RTX 5080, it's as impressive as we hoped it would be. Like all the best remasters and remakes that aim to stay faithful to the source material, Half-Life 2 RTX is like your fondest memories of the game brought to life with some of the most cutting-edge visuals ever seen in an interactive experience. It's like how you remember Half-Life 2 but with new assets and stunning ray-traced effects that only amplify and enhance what is still an all-timer.

Baldur's Gate 3 director Swen Vincke has taken to X, formerly Twitter, to share his thoughts on the claims that single-player titles are "dead". He doesn't think that's the case.

According to Vincke, single-player games aren't dying the slow death that many studio executives are declaring they are, and for a single-player game to be successful, it first needs to be good. If there is any example of single-player games not being dead, it is Baldur's Gate 3, which, following its release, went on to win Game of the Year and retain a solid, consistent player base that's still extremely alive today, two years after its release. At the time of reporting Baldur's Gate 3 has 36,000 concurrent players on Steam, with its 24-hour player peak on Valve's platform being 68,000.

It appears not only gamers are with Vincke's take on single-player games, but so are some industry developers, with David Goldfarb, a known developer behind Battlefield: Bad Company 2 and PayDay 2, commenting on Vincke's post saying, "Can I bring you to all my meetings, Swen? F**king exhausted." Other indicators that single-player games aren't dying are titles such as Black Myth: Wukong, God of War: Ragnarok, STALKER 2, and what will be Grand Theft Auto VI when it releases at the end of this year.

Assassin's Creed Shadows has been one of the most controversial launches for the franchise in recent memory, and now it appears reviews for the title are leaking ahead of when they should be posted.

Screenshots have been posted on subreddits showing review scores from OpenCritic and IGN Brasil, with both of the sources giving the contentious titles a respectable 8/10 and 9/10. Notably, reviews for the title are scheduled to drop on March 18, which at the time of reporting is today, but the specific time remains unclear. As for the reviews that leaked early, they have since been removed from their respective sources.

The ConsoleCreatures review for the title sourced on OpenCritic states the title gives players a lot of freedom and "takes you on a journey of exploration and history of one of Japan's most famous periods". As for the IGN Brasil, the short review was translated from Portuguese and states Shadows is "one of the most iconic titles in the franchise," with the reviewer going on to mention it successfully juggles combat and stealth mechanics against a Japense backdrop that's been previously untouched by Ubisoft.

Amazon has confirmed it's making two seasons of the God of War TV show, as the news has come from the showrunner Ronald D. Moore, who is overseeing the project that's currently in the works.

Moore appeared on The Sackhoff Show with Katee Sackhoff, where he was asked about what he's currently working on. The showrunner explained his current workplate consists of an "adaptation of this video game called God of War" and that his job is to be in the writer's room, adapting the video games to TV. Moore goes on to say the challenge is daunting with such a large and well-respected IP, and is especially daunting considering he doesn't consider himself a gamer.

Sackoff followed up by asking Moore if he has tried to play any of the God of War games, and the showrunner said he has taken a "stab at it" but gave up as he can't work out the controls. Moore goes on to explain that he grew up in '80s where video games were in arcades, making this whole business with what are traditional video game controllers nowadays extremely foreign to him. "The controllers now..." he laughs, "Press R1, which one's R1, oh I'm dead! I can't quite get a hold of that."

Mark Rober, a former NASA engineer and long-time YouTuber, has tested a Tesla's Autopilot visual imaging detection system against Light Detection and Ranging (LiDAR), which many other vehicle manufacturers use to detect their surroundings.

The goal of the tests was to demonstrate the strengths/weaknesses of both detection systems, with Tesla being criticized for using visual data-only Autopilot as specific physical obscurifications can result in the system failing to detect a potential danger, resulting in the correct time to break being missed. On the other hand, LiDAR has been criticized by Tesla CEO Elon Musk, who has said it's "fricking stupid, expensive and unnecessary," despite its ability to see through physical tests such as fog or heavy rain.

Rober tested various environments on both detection systems and found both the Autopilot system and the LiDAR system stopped the vehicle for a motionless child mannequin in the middle of the road. Both systems also passed the "child running out onto the road test." However, the systems differentiated when the child mannequin was engulfed in fog, as the LiDAR system was able to detect the mannequin through the fog, while Autopilot wasn't. The Tesla failed this test and slammed directly through the mannequin.

NASA has confirmed the James Webb Space Telescope (JWST), the world's most powerful space telescope, has captured direct images of multiple exoplanets within Earth's galactic neighborhood.

The space agency has taken to its blog to announce Webb has observed the planets within HR 8799, and through its imaging, it has gathered strong evidence that the system's four gas giant planets formed much like Jupiter and Saturn, which slowly built solid cores that eventually pulled in gas from a protoplanetary disk. The young star system is only located 130 light years away and has long been a target of planetary studies, given the age of HR8799 is only 30 million years old, which is just a fraction of our own solar system's 4.6 billion years old.

Webb's results have been published in The Astrophysical Journal. NASA explains in its blog post that gas giants can form in two ways. The first is by building solid cores with heavier elements that attract gas. The second is when particles of gas rapidly coalesce into massive objects from a young star's cooling disk. The first process is called core accretion, and the second is called disk instability. Webb's results indicate the planets within HR 8799 have taken the first route, and by understanding these processes, researchers can better distinguish between the two when hunting for more exoplanets while also contextualizing the evolution of our own solar system.

Bots in competitive or cooperative multiplayer games have been a problem for years, with a new study from Echelon Insights stating that 59% of gamers have encountered third-party bots in games. And with that, 71% agree that bots make it less fun to play certain games, with 77% agreeing that it's essential to verify that you are a human before jumping into the action. And these bots are getting increasingly sophisticated in the age of AI gaming.

'Razer ID verified by World ID' is one solution to this problem, and it's set to be part of the upcoming multiplayer game TOKYO BEAST. The game is set in Tokyo in the year 2124, a time when humans and "advanced autonomous androids" coexist. It's a competitive title that includes collectible unique NFTs alongside in-game purchases via Razer.

'Razer ID verified by World ID' is powered by secure, single sign-on (SSO) technology designed to verify "real human gamers" to create a more authentic and immersive gaming experience. World ID, a company co-founded by Sam Altman and Alex Blania, is all about confirming someone's unique online human identity with a focus on privacy. It can tell that someone is human without knowing who they are.

As announced in 2022, Ubisoft Toronto is leading development on a Splinter Cell remake, built from the ground up using Ubisoft's Snowdrop engine.

The original Splinter Cell (2002) calls back to a time when Ubisoft was positioned at the forefront of innovation in games - and as evidenced by the team's commitment to return to what made the game great, it's shaping to be a pleasant departure from the publisher's open-world formula.

As reported by Videogamer, former Ubisoft tools support specialist Wesley Chartrand shared some behind-the-scenes hints on LinkedIn, detailing his role as a scrum master and his focus on project efficiency. More notably, Chartrand spent two years "organizing and running retrospectives for the Splinter Cell DevOps team" - a sign that the studio is actively studying the DNA of classic Splinter Cell rather than simply modernizing it.

Intel XeSS 2 is the company's significant overhaul and update to its AI-powered upscaling solution, including XeSS Super Resolution, XeSS Frame Generation, and XeLL Low Latency. Providing feature parity with NVIDIA's DLSS suite of AI rendering tools, at GDC 2025, Intel has announced that the XeSS 2 SDK is now available to download.

The XeSS SDK is available on GitHub and is open to all developers. It gives them the tools to integrate the full range of XeSS 2 features into their games. Intel also offers developer guides for enabling XeSS Super Resolution, Frame Generation, and its Reflex-like XeLL Low Latency solution, alongside the XeSS Inspector Tool for "verifying, debugging, and tuning XeSS 2."

Plus, plugins for Unreal Engine and Unity, two of the most popular engines and platforms used by game developers today. Intel's XeSS 2 is optimized for Intel graphics hardware and covers a range of mobile and desktop products.

Texture compression has been used in PC gaming for decades, ever since the first texture was applied to a basic 3D character model or environment. In 2025, with the rise of 4K gaming across the PC and console space, these high-resolution textures often contain several layers. The dinosaur example below shows eight 4096 x 4096 PBR textures with 13 channels of surface detail covering roughness, ambient occlusion, and other properties.

Add this all up, and you've got 200 million texture values per texture that would need to be stored in memory (the VRAM on a GPU) for the single in-game asset. Regarding gaming in super-high resolution with ultra settings in 2025, this is one of the reasons why the general rule of thumb is - the more VRAM, the better. Even so, new AI-powered methods of texture compression will significantly improve performance.

At GDC 2025, Intel's Anis Benyoub is joining Microsoft on stage as it showcases 'Cooperative Vectors on DirectX,' a new feature that will lead to a 10X improvement in inference performance for Neural Block Texture Compression on Intel Arc GPUs. Joe Rozek from AMD and Alexey Panteleev from NVIDIA will also be there to discuss how Cooperative Vectors will be used with Radeon and GeForce RTX hardware.

CPU-Z's new v2.15 update delivers support for the latest silicon from AMD and NVIDIA, with support for Strix Halo and Krackan APUs, while added GPU support includes the Radeon RX 9070 series and GeForce RTX 5070 Ti.

Not only that, but CPU-Z now lets you check your GPU and ROP count, just in case you want to make sure your graphics card has the right ROP count, as that's now a thing unfortunately. AMD's new Ryzen AI Max 300 series "Strix Halo" APUs and Ryzen AI 340/350 "Krackan" APUs, as well as the new RDNA 4-based Radeon RX 9070 XT and RX 9070 are all supported in CPU-Z now.

NVIDIA's new Blackwell-based GeForce RTX 5070 Ti and RTX 5070 are now both supported by CPU-Z, as well as the entire fleet of new EPYC 9000 series CPUs from the monster 192-core, 384-thread EPYC 9965 down to the 8-core, 16-thread EPYC 9015.