The development of AI models has become increasingly costly as their size and complexity grow, requiring massive computational resources with GPUs playing a central role in handling the workload.
Phison, a key player in portable SSDs, has unveiled a new solution that aims to drastically reduce the cost of training a 1 trillion parameter model by shifting some of the processing load from GPUs to SSDs, bringing the estimated $3 million operational expense down to just $100,000.
Phison's strategy involves integrating its aiDAPTIV+ software with high-performance SSDs to handle some AI tool processing tasks traditionally managed by GPUs while also incorporating NVIDIA’s GH200 Superchip to enhance performance and keep costs manageable.
AI model growth and the trillion-parameter milestonePhison expects the AI industry to reach the 1 trillion parameter milestone before 2026.
According to the company, model sizes have expanded rapidly, moving from 69 billion parameters in Llama 2 (2023) to 405 billion with Llama 3.1 (2024), followed by DeepSeek R3’s 671 billion parameters (2025).
If this pattern continues, a trillion-parameter model could be unveiled before the end of 2025, marking a significant leap in AI capabilities.
In addition, it believes that its solution can significantly reduce the number of GPUs needed to run large-scale AI models by shifting some of the processing tasks away from GPUs to the largest SSDs and this approach could bring down training costs to just 3% of current projections (97% savings), or less than 1/25 of the usual operating expenses.
Phison has already collaborated with Maingear to launch AI workstations powered by Intel Xeon W7-3455 CPUs, signaling its commitment to reshaping AI hardware.
As companies seek cost-effective ways to train massive AI models, innovations in SSD technology could play a crucial role in driving efficiency gains while external HDD options remain relevant for long-term data storage.
The push for cheaper AI training solutions gained momentum after DeepSeek made headlines earlier this year when its DeepSeek R1 model demonstrated that cutting-edge AI could be developed at a fraction of the usual cost, with 95% fewer chips and reportedly requiring only $6 million for training.
Via Tweaktown
You may also likeHisense has announced its 2025 ULED mini-LED lineup, with all-new models across its U9, U8, U7, and U6 series TVs.
Hisense has been responsible for some of the best mini-LED TVs we’ve seen in recent years, and the company manages to keep its prices low, even for top-tier models. Last year’s Hisense U8N currently ranks as the mid-range pick in our best TVs guide, and the Hisense U7N and Hisense U6N are also among our best budget TV picks.
The new models announced today range in size from 55 to 100 inches, and are “powered by next-generation AI processing that works effortlessly behind the scenes to deliver smarter, more intuitive picture enhancements—without the need for manual adjustment,” according to the company.
The new top series, the U9, uses the company’s Hi-View AI Engine X processor, which is derived from its high-end ULED X lineup. Two other series, the U8 and U7, use a Hi-View AI Engine Pro processor.
According to Hisense, both processors "automatically fine-tune contrast, color accuracy, and motion clarity in real-time, ensuring lifelike images and fluid motion across all content types."
Google TV will remain the smart TV platform for the U9, U8, and U7 series, while, in a surprising change, the entry-level U6N series will use Amazon’s Fire TV platform.
The 2025 Hisense ULED TVs are scheduled for availability later this year. Pricing has not yet been announced.
U9 SeriesThe new U9N series expands upon 2024's Hisense U9N with a new 65-inch model to accompany the 75- and 85-inch screen sizes. Its advanced chipset is packed with AI picture-enhancing features, including AI 4K Upscaler, AI Super Resolution, AI Noise Reduction, AI LocalDimming, AI HDR Upscaler, and AI Depth Enhancer.
Other picture-improving features of the U9 series include an Ultra LR panel to reduce screen reflections and an ultra-wide viewing angle to retain contrast over a wider seating range. Dolby Vision IQ, HDR10+, IMAX Enhanced, and Filmmaker Mode are all supported.
For gaming, the U9 series has been upgraded to a 165Hz native refresh rate and supports FreeSync Premium Pro. A 4.1.2-channel built-in speaker array conveys Dolby Atmos soundtracks, with a 5.1.2-channel array used on the 75- and 85-inch models.
U8 Series (Image credit: Hisense )Available in 55- to 100-inch screen sizes, the U8 series uses the company’s Hi-View AI Engine Pro processor, which taps AI to automatically optimize picture settings in real-time.
According to Hisense, U8 series TVs can deliver brightness levels up to 5,000 nits and have expanded local dimming zones for 2025. Dolby Vision IQ and HDR10+ are supported here, plus the TVs are IMAX Enhanced certified.
Gaming support will get a boost on the new U8 series with a native 165Hz refresh rate. Audio, too, will get a boost from the U8N series’ 2.1.2-channel built-in Dolby Atmos speaker array up to 4.1.2 channels.
U7 Series (Image credit: Hisense)The Hisense U7N series was among our favorite budget TV options in 2024, and the new U7N will keep its budget status with models available for under $1,000.
Screen sizes for the U7N series will range from 55 to 100 inches, and the TVs will feature an AGLR-Antiglare low reflection panel that “ensures glare-free visuals from any seat in the house," according to Hisense.
Despite being budget TVs, the U7N series is packed with gaming features, including 165Hz, Game Booster 288Hz, Dolby Vision gaming, and FreeSync Premium Pro support. The U7N series TVs also feature a built-in 2.1.2-channel, 60W speaker array.
U6 Series (Image credit: Hisense)The entry-level series in the new mini-LED lineup is the Hisense U6N, which is available in 55- to 100-inch screen sizes. As mentioned earlier, this series uses Amazon’s Fire TV smart interface instead of Google TV, and the sets come with an Alexa Voice Remote.
Otherwise, the U6 series TVs have a 2.1-channel speaker array with a built-in subwoofer. Gaming support includes a 144Hz native refresh rate and FreeSync Premium.
You might also likeThe Ninkear MBOX 8 Pro is a modular mini PC with a detachable docking station for a 3.5-inch hard drive, offering expandable storage but raising concerns about its practicality, especially when compared to the best portable SSDs.
Per Androidpc (originally in Spanish), the MBOX 8 Pro design includes a magnetic external hard drive bay with four USB ports, integrating a traditional hard drive instead of relying solely on internal SSD storage that suits the best video editing computers.
However, hard drives generate heat and are more prone to mechanical failure than SSDs, so placing one in an enclosed, magnetically attached base may impact thermal management, making their integration into a mini PC debatable.
A bold storage experimentThe MBOX 8 Pro stands out for its ability to drive three 8K monitors via HDMI 2.1, DisplayPort 1.4, and USB4 Type-C, with Wi-Fi 6, Bluetooth 5.2, and dual Gigabit Ethernet.
It features a 4nm AMD Ryzen 7 8845HS processor, an octa-core APU with 38 TOPS AI acceleration, paired with up to 32GB DDR5 RAM and a 1TB NVMe SSD.
Combined with the Radeon 780M iGPU, this mini PC delivers performance that rivals some desktops, handling demanding applications and even 1080p gaming with ease.
Priced at around €586, the Ninkear MBOX 8 Pro provides a unique take on the mini PC market, though its unconventional design choices may not suit everyone, particularly those who rely on SSDs for speed and stability in video and photo editing PC workflows.
You may also likeWe’re just days away from the next Nintendo Direct – taking place on April 2, 2025 – which is highly anticipated as it will be entirely focused on the Nintendo Switch 2. Hopefully, it will shed light on price, the full feature set, and the all-important launch titles.
In the lead-up, Nintendo’s dropped some new tracks for its Nintendo Music app and, as first reported by NintendoLife, has updated a support page teasing a larger update to the standalone app arriving in October 2025.
Nintendo included tracks from classics Tetris and – a personal favorite – Dr. Mario in the March 25, 2025 drop. It also included some tracks from Kirby and the Forgotten Land, a Nintendo Switch title. It’s not the full soundtrack from Kirby, though, hinting that we could get another drop.
Even though it's not every song from Kirby, a more substantial drop for the service is appreciated. Nintendo Music is essentially an Apple Music- or Spotify-like music streaming service dedicated to music from the Nintendo universe.
All your favorite tracks from Mario Kart, Animal Crossing, and countless other titles are easily accessible for your listening pleasure via an app for Android and iOS – if you have a Nintendo Switch Online membership.
Now, on the updated support page, Nintendo clearly states an update is arriving in October 2025:
“Your data, such as your playlists, will be saved even after you cancel your Nintendo Switch Online membership. We are planning an update around October 2025 that will allow you to use some Nintendo Music functions, such as the ability to edit, share, or delete your playlists, even after you cancel your Nintendo Switch Online membership.”
Considering the update promises more social and management features even if you’re not a subscriber, Nintendo might be changing the requirement of a Switch Online Membership to access the music streaming service. It is nice to see that all your content – think playlists – will be saved if you discontinue your membership.
It might be setting the stage for some changes to accessing Nintendo Music or even a relocation of accessing and streaming the songs. Time will tell, but who knows, Nintendo might even better integrate the service with a larger update to its online offering or with the forthcoming Nintendo Switch 2.
Time will tell, but for now, you can enjoy some sweet tunes from Kirby and the Forgotten Land, Dr. Mario, or countless other titles via the Nintendo Music app if you’re a Switch Online subscriber.
If you’re curious about the Nintendo Music app, you can check out our initial thoughts here.
You might also likeWith the iPhone 17 Pro launch now likely about six months away, the rumor mill is in overdrive. But while recent speculation has focused on an unsubtle redesign for the flagship phone, the more interesting rumors are predicting camera upgrades – including 8K video recording.
As spotted by BGR, the Chinese leaker Fixed Focus Digital – who has a reasonably strong track record of Apple leaks on Weibo – has predicted that the next Pro iPhones will get the ability to shoot 8K video.
That isn't exactly an outlandish or unrealistic prediction. A number of Android phones, from the Samsung Galaxy S25 Ultra to the Xiaomi 15 Ultra, can already shoot in 8K/30fps. The ability to shoot in 8K also landed on Samsung Galaxy S20 way back in 2020.
The rumor also aligns with the latest iPhone 17 Pro camera rumors, which predict a new 48MP sensor for its telephoto camera. That would mean all three of its cameras would have the resolution needed to shoot 8K (which typically means 7680 x 4320 frames).
But even if the iPhone 17 Pro did get an 8K video upgrade, it would likely be a very niche feature. While that resolution may occasionally prove useful if you need to crop in on a scene during editing, other factors like bit-rate and lens quality (always a limitation on smartphones) will throttle the overall image quality – and likely make shooting in 4K modes the more sensible option.
Still, while iPhone are already impressive video cameras, there are a few new features that I think would be more useful than 8K on the iPhone 17 Pro.
iPhone 17 Pro: 3 video features I want to see 1. Open gate video recording (Image credit: Apple)While iPhones have historically had sensors with 4:3 aspect ratios, they've typically used a 16:9 crop for video recording. So-called 'open gate' video recording means being able to use the full width and height of the sensor, so you're not throwing away any detail –and it'd be great to see this as an option on the iPhone 17 Pro.
This would let you choose different aspect ratios – for example, shooting square video – while using the whole sensor. It would be more processor-intensive, but I'm sure an A19 Pro could handle it. Even relatively affordable cameras like the Fujifilm X-M5 are now offering open gate recording as an option for social content creators.
This is already possible on some iPhones with apps like Pearla, but it'd be nice to see it come to the default camera app or Final Cut Camera without a subscription.
2. A larger telephoto sensor (Image credit: Apple)Most iPhone video shooters stick to the main 24mm lens, because it's the only one with that's good enough for anything beyond cat or family videos. That's mostly due to its relatively large 1/1.29-inch sensor, so I hope the 17 Pro gets a much bigger telephoto sensor than its current (and aging) 1/3-inch offering.
Right now, the latest rumors are predicting that the 17 Pro's telephoto camera will get a boosted 48MP resolution, but the sensor size is less clear. It wouldn't need to match the main camera, but the Samsung Galaxy S25 Ultra's 50MP telephoto has a 1/2.52-inch sensor – so something in that ballpark would make a big difference for both video and stills. It could even make the telephoto a handy kind of b-cam for cut scenes.
3. A nano-texture screen option (Image credit: Future / Axel Metz)Okay, this one's not strictly a video feature, but it would be a nice option for video shooters. Apple now gives you a nano-texture screen option on everything from the iPad Pro M4 to the MacBook Pro M4 to help reduce glare and reflections – so why not offer a similar option for its Pro iPhones?
Given the iPhone's only viewfinder is its screen, and video is often shot outside in the glare of the sun, it would offer benefits for both recording and playback. Granted, you can already get halfway there with a matte screen protector, but Apple could go far beyond that with a new kind of display tech.
We've seen patents and rumors suggesting Apple's been working on display tech with an “ambient light rejection element” before, so perhaps it could be ready in time for the iPhone 17 Pro. If Apple wants to tout it as the ultimate phone for video creators, that would be a more useful quality-of-life upgrade than 8K resolution.
You might also likeTowards the end of January 2025, Jeffrey Emanuel, Founder and CEO of Pastel Network, wrote an interesting and insightful essay about Nvidia.
More specifically, he discussed what he saw as its overvaluation and the threat it faced from a small Chinese startup called DeepSeek. noting, “history shows that markets eventually find a way around artificial bottlenecks that generate super-normal profits,” and that, “Nvidia faces a much rockier path to maintaining its current growth trajectory and margins than its valuation implies.”
That article, which went viral, helped cause the largest-ever single-day drop in the stock market, wiping $2 trillion off global markets and slashing Nvidia’s market capitalization by $600 billion.
An overpriced turkeyThe latest target of Emanuel’s observations is Roseland, New Jersey-based CoreWeave, which offers cloud infrastructure optimized for Nvidia GPUs and is expected to go public in the coming weeks in what will be a closely watched IPO.
In his takedown of the company, Emanuel described it as the “WeWork of AI,” referring to the infamous flexible office space startup that was once valued at $47 billion but overexpanded, mismanaged funds, and collapsed after a failed IPO.
As Emanuel wrote two weeks ago, “If it [CoreWeave] really IPOs for $30b+ then it’s a much better short than NVDA ever was. They have absolutely no durable moat and will soon be structurally disadvantaged on the cost curve for inference for specific LLM models versus hyperscalers like AWS, which have their own proprietary silicon (because clearly CoreWeave doesn’t know the first thing about making custom silicon, and won’t anytime soon).”
MarketWatch picked up on this and cautioned investors. The site wrote, “CoreWeave set an estimated price range of $47 to $55 a share for its initial public offering, and an offering size of 49 million shares. Based on the midpoint of the estimated range, CoreWeave’s IPO would raise $2.5 billion for the company. That would make its IPO the fourth since 2022 to raise at least $2.5 billion, after Lineage, Arm Holdings, and Kenvue, according to Renaissance Capital.”
“CoreWeave would have a market cap of about $25 billion based on the midpoint of the estimated pricing range,” the site added. While that’s far less than the $35 billion many market watchers have been predicting, Emanuel says, “that’s still at least $20b too high for this turkey!”
Thanks to Steve Gelsi at @MarketWatch for helping me warn investors about the CoreWeave IPO. Luckily, they've already had to reduce the valuation by $10b from the initial price talk of $35b, but that's still at least $20b too high for this turkey! https://t.co/fnJorj87NkMarch 20, 2025
You might also like