Gigabyte, through its subsidiary Giga Computing, is offering qualified individuals and organizations the opportunity to test one of the world’s most advanced supercomputers, the Gigabyte G383-R80 server powered by AMD MI300A APUs for free.
There's a catch; after the seven-day trial period, the configured price of this high-performance system is $304,207.
The price isn't the only thing that might ward you off; in addition to it being a strict time-bound trial period, distributors aren't eligible to apply, and users must have a clear project in mind to qualify.
Claiming Gigabyte's G383-R80 offerTo participate, applicants must fill out a form on the Gigabyte Launchpad website. Giga Computing will then review the application based on the commercial value or innovative potential of the proposed project.
If approved, the company will contact the applicant within three business days to confirm details. The trial period lasts for seven days, though extensions of up to two weeks can be requested through a sales representative or via email. Access to the server will be granted within two weeks, and users must initiate their project within three days of receiving the access link.
The Gigabyte G383-R80 server is designed for demanding workloads such as AI training, AI inference, and high-performance computing (HPC). It features a 3U rack-mount chassis and supports up to four AMD Instinct MI300A APUs, which combine CPU and GPU cores for accelerated computing.
For storage, it has eight 2.5-inch NVMe Gen5/SATA/SAS hot-swap drive bays, 12 PCIe Gen5 x16 slots, and a variety of storage options, including M.2 NVMe SSDs and U.2/U.3 NVMe SSDs with capacities ranging from 400GB to 61.44TB.
Networking capabilities include onboard 10Gb/s Ethernet ports and support for PCIe expansion cards like RJ45, SFP+, QSFP28, and QSFP56.
You may also likeA team of researchers at the University of Hong Kong has designed and tested an advanced aerial robot capable of navigating complex environments at high speeds of up to 20 meters per second while maintaining precise control.
Named SUPER, the quadcopter drone uses cutting-edge LiDAR technology to detect and avoid obstacles, even thin wires that have posed challenges for traditional drones.
In research published on Science Robotics (via Techxplore), the team noted while SUPER has potential applications in search and rescue, its ability to operate autonomously in unknown environments suggests it could also be used for law enforcement and military reconnaissance.
The power of LiDAR for precision flightUnlike conventional aerial robots that rely on cameras and sensors, SUPER uses 3D light detection and ranging (LiDAR) to map its surroundings and process spatial data in real time, allowing it to function in low-light conditions.
With a detection range of up to 70 meters, the LiDAR system feeds data to an onboard computer that continuously analyzes the environment.
This system enables SUPER to chart two distinct flight paths, one prioritizing safety and another allowing for exploratory movement.
To demonstrate its capabilities, the research team tested SUPER against a commercial drone, the DJI Mavic 3.
While the DJI model avoided larger obstacles, it struggled to detect thin wires of smaller diameters. In contrast, SUPER successfully avoided all obstacles, including 2.5-mm thin wires, thanks to its high-resolution point cloud processing.
The test result also reveals the drone glided through dense forests, tracking moving targets without colliding with trees or branches.
You may also likeGoogle has dropped a major upgrade to the Gemini app with the release of the Gemini 2.0 Flash Thinking Experimental model, among others. This combines the speed of the original 2.0 model with improved reasoning abilities. So, it can think fast but will think things through before it speaks. For anyone who has ever wished their AI assistant could process more complex ideas without slowing its response time, this update is a promising step forward.
Gemini 2.0 Flash was originally designed as a high-efficiency workhorse for those who wanted rapid AI responses without sacrificing too much in terms of accuracy. Earlier this year, Google updated it in AI Studio to enhance its ability to reason through tougher problems, calling it the Thinking Experimental. Now, it’s being made widely available in the Gemini app for everyday users. Whether you’re brainstorming a project, tackling a math problem, or just trying to figure out what to cook with the three random ingredients left in your fridge, Flash Thinking Experimental is ready to help.
Beyond the Thinking Experimental, the Gemini app is getting additional models. The Gemini 2.0 Pro Experimental is an even more powerful one, albeit a somewhat more cumbersome version of Gemini. It's aimed at coding and handling complex prompts. It’s already been available in Google AI Studio and Vertex AI.
Now, you can get it in the Gemini app, too, but only if you subscribe to Gemini Advanced. With a context window of two million tokens, this model can simultaneously digest and process massive amounts of information, making it ideal for research, programming, or rather ridiculously complicated questions. The model can also utilize other Google tools like Search if necessary.
Lite speedGemini is also augmenting its app with a slimmer model called Gemini 2.0 Flash-Lite. This model is built to improve on its predecessor, 1.5 Flash. It retains the speed that made the original Flash models popular while performing better on quality benchmarks. In a real-world example, Google says it can generate relevant captions for around 40,000 unique photos for less than a dollar, making it a potentially fantastic resource for content creators on a budget.
Beyond just making AI faster or more affordable, Google is pushing for broader accessibility by ensuring all these models support multimodal input. Currently, the AI only produces text-based output, but additional capabilities are expected in the coming months. That means users will eventually be able to interact with Gemini in more ways, whether through voice, images, or other formats.
What makes all of this particularly significant is how AI models like Gemini 2.0 are shaping the way people interact with technology. AI is no longer just a tool that spits out basic answers; it’s evolving into something that can reason, assist in creative processes, and handle deeply complex requests.
How people use the Gemini 2.0 Flash Thinking Experimental model and other updates could show a glimpse into the future of AI-assisted thinking. It continues Google's dream of incorporating Gemini into every aspect of your life by offering streamlined access to a relatively powerful yet lightweight AI model.
Whether that means solving complex problems, generating code, or just having an AI that doesn’t freeze up when asked something a little tricky, it’s a step toward AI that feels less like a gimmick and more like a true assistant. With additional models catering to both high-performance and cost-conscious users, Google is likely hoping to have an answer for anyone's AI requests.
You might also likeiOS 18.3 is here – and contrary to rumors being spread on TikTok and elsewhere, it doesn’t install Elon Musk’s Starlink tech on your iPhone. So, there’s no reason not to get the latest iOS update on your iPhone and ensure that you’re up to date.
iOS 18.3 is a relatively minor update, which mostly impacts Apple Intelligence – enabling the AI features by default and rolling out some fixes for Notification Summaries – and fixing several bugs. It does, however, make a change for T-Mobile customers by allowing the iPhone 14, iPhone 15, and iPhone 16 lineups to potentially connect to the Starlink-powered terrestrial network of the carrier.
It is not, however, allowing that connection by default, and T-Mobile’s partnership with Starlink is still in beta for a select few customers who opt to join it and then get selected to participate. Apple doesn’t have a partnership with Starlink, but T-Mobile does, and you need to opt in a few ways. Let’s unpack this ahead.
The myth: iOS 18.3 installs Starlink on your iPhone (Image credit: Future/Jacob Krol)The concern in the now viral TikToks is that the latest version of iOS basically adds a direct connection to Starlink to your iPhone. The main point of concern is that ‘Starlink can now work with iPhone and access it’ without any formal announcement from Apple on if it’s a mandatory connection.
Apple initially launched its Emergency SOS via Satellite service alongside the iPhone 14 – with support for the iPhone 15 and 16 – so the smartphones could connect to a satellite. However, it’s not on by default and is only engaged when no LTE or Wi-Fi network is available.
Since then, some carriers have offered satellite networks alongside a typical phone network. T-Mobile is doing that, and it initially announced its partnership with Starlink in August 2022.
Apple also updated a support page detailing how to turn off carrier-powered satellite features. To do this, open Settings, navigate to Cellular, select your carrier, and turn off ‘Satellite.’
Simply, though, iOS 18.3 does not install Starlink on the iPhone. Essentially, it is packaged within iOS 18.3 as a carrier network settings update for T-Mobile that allows for the connection. It is not on by default, though, and you need to be selected to join the beta after requesting a spot.
The reality: T-Mobile has a partnership with Starlink that is currently in beta and iOS 18.3 is safe to install (Image credit: Apple)So no, iOS 18.3 does not add a direct line to the Starlink network – forced or unforced – as some viral TikToks claim. It makes network settings changes that allow T-Mobile-connected iPhone 14, 15, or 16 to connect to T-Mobile 1900MHz spectrum, accessed through antenna ‘band 25’ on the iPhone to access the Starlink network.
Even for that network connection to happen, you need to have an eligible T-Mobile plan, register for the beta, and be selected to participate in it. Then, you need to be in an area where that network is supported and when a typical cellular network or Wi-Fi is unavailable. You’ll know that is the case when you see “SAT” replace the standard cellular bars and “4G,” “5G,” or “5G UW” in the top right corner of your iPhone.
T-Mobile opened its Starlink network beta program in December 2024, and interested customers have been able to register for it. It was first available to Android smartphones, and then the capability for the iPhone rolled out with iOS 18.3.
The partnership and the ability for T-Mobile devices to connect to Starlink’s direct-to-cell technology aims to reduce dead zones and allow users to stay connected. T-Mobile is also the only cellular network in the United States to have this partnership with Starlink.
Apple’s satellite connectivity for its iPhone under the ‘Emergency SOS’ feature is not Starlink and is done through a partnership with Globalstar.
Furthermore, it’s also best practice to keep your iPhone and other devices up to date, as using older software can make them more susceptible to security and privacy issues. iOS 18.3, like most iOS updates, brings some new features but also, at times, critical bug fixes and important security patches.
So, long story short, iOS 18.3 does not add a direct connection to Starlink to your iPhone. It simply allows a T-Mobile-connected iPhone to use that network when you're outside of traditional coverage if you’ve opted into the beta and have been selected. It’s also a partnership between T-Mobile and Starlink – Apple isn’t involved there and it doesn't have any impact or change to Apple's Emergency SOS via Satellite functionality. That service has been using Globalstar satellites since its inception.
If you’re on T-Mobile and want to opt out of using Starlink, open Settings on your iPhone, click Cellular, select your Carrier (in this case, T-Mobile), and turn off Satellite.
You might also likeChipmaking giant AMD has confirmed it recently patched a high-severity vulnerability affecting its Zen 1 to Zen 4 CPUs.
The company published a new security advisory, detailing the bug and its potential for exploitation, noting, “Researchers from Google have provided AMD with information on a potential vulnerability that, if successfully exploited, could lead to the loss of SEV-based protection of a confidential guest."
SEV is short for Secure Encrypted Virtualization - a hardware-based security feature designed to enhance the confidentiality and integrity of virtual machines (VMs) running on AMD EPYC processors. It encrypts the memory of individual VMs using unique encryption keys, ensuring that neither the hypervisor nor other VMs can access their data.
Mitigations availableThe vulnerability is tracked as CVE-2024-56161, and has a severity score of 7.2/10 (high). It is described as an improper signature verification flaw in AMD CPU ROM microcode patch loader, which could allow threat actors with local admin privileges to load malicious CPU microcode. As a result, the confidentiality and integrity of a confidential guest running under AMD SEV-SNP would be lost.
“AMD has made available a mitigation for this issue which requires updating microcode on all impacted platforms to help prevent an attacker from loading malicious microcode,” the company concluded.
“Additionally, an SEV firmware update is required for some platforms to support SEV-SNP attestation. Updating the system BIOS image and rebooting the platform will enable attestation of the mitigation. A confidential guest can verify the mitigation has been enabled on the target platform through the SEV-SNP attestation report.”
The company only publicly disclosed the flaw recently, but the patch was actually released in mid-December 2024. AMD decided to delay the announcement to give its customers enough time to mitigate the problem.
You might also likeAt least two workstation specialists have put supercharged PCs with Nvidia RTX 5090 GPUs on sale over the past few days. The most impressive of them all is the Bizon ZX5500 which packs up to seven (yes, seven) water-cooled 32GB RTX 5090 GPUs in a tall tower casing. This is the best GPU ever built and buying it through system builders seems to be the only way to avoid months-long wait.
While BizonTech's solution will probably feature in our best workstation guide, it is not as expansive as Comino’s Grando server, which has eight RTX 5090 GPUs, but the latter has yet to get a launch date (I contacted Comino for more details).
The ZX5500 doesn't come cheap at just under $102,000 with the GPUs accounting the lion share (more than 83%) of the total cost. That’s almost 3x the price of MIFCOM’s Big Boss which has seven liquid-cooled RTX 4090 GPUs.
A beefier 6Kw power supply unit plus and the cards cost an extra $85,000 compared to the same system with a pair of RTX 5080 (with 16GB VRAM each). As a reminder, the suggested retail price of the RTX 5090 is ‘just’ $2000.
An RTX 5090 on its retail packaging on a desk (Image credit: Future)The ZX5500 can be updated to a 96-core ThreadRipper Pro CPU with 1TB of DDR5 RAM, almost 1PB of PCIe 4.0 SSDs (15 x 61.44TB SSD) and seven liquid-cooled Nvidia H200 AI GPU; such a configuration pushes the price above half a million US Dollars.
Where to find RTX 5090? Ask Pro system buildersBizontech is a niche boutique vendor that specializes in servers, workstations and clusters for AI, deep learning and HPC. The RTX 5090 is sold out pretty much everywhere and it seems that Nvidia is prioritizing business and creative outlets like Bizontech, Puget Systems and Punch Technology, with workstations seemingly ready to be shipped within days rather than week.
Jon Bach, President, Puget Systems told me, “Supply for the 5090 (and the 5080) is very limited, and we expect that to be the case for at least through March. Puget Systems has a good number of cards in hand at the moment because of our OEM relationships, but we appear to be somewhat unusual in that respect. Overall, we are filling orders, but expect our lead times to be affected until supply improves."
The creative crowd will love the RTX 5090 as it obliterates absolutely everything in its path but at a price. Puget Systems and Storagereview benchmarked it across a wide range of AI and creative tests and found that it performed significantly better than previous generations (and AMD’s finest cards) albeit with a much higher power station.
TechRadar’s John Loeffler published a review of the RTX 5090 recently, calling it the supercar of graphics cards and asking whether it was simply too powerful, suggesting that it is an absolute glutton for wattage. He continues, “It's overkill, especially if you only want it for gaming, since monitors that can truly handle the frames this GPU can put out are likely years away.”
This, of course, will be irrelevant to Nvidia’s plans to launch an even more powerful version of the RTX 5090, one with a rumored 96GB GDDR7 memory which will replace the RTX 6000 ADA in due time. If this card follows the same inflationary trajectory as its consumer version then I won’t be surprised if its ticket price reaches $15,000, making it the most expensive graphics card of all time.
You might also likeApple has delivered a new patch on Xprotect, its on-device malware removal tool, intended to block several variants of the macOS ‘Ferret’ family of threats.
As reported by AppleInsider, the new update will counter several issues, including Ferret variants FRIENDLYFERRET_SECD, FROSTYFERRET_UI, and MULTI_FROSTYFERRET_CMDCODES.
These malware variants are reportedly used by North Korean hackers in what has been dubbed the ‘Contagious Interview’ campaign, in which criminals would create fake job openings, primarily targeting software developers or high–profile industries like defense, government departments, or aerospace. The new updates to Xprotect will help block this family of malware from Mac devices, here’s everything we know so far.
The Ferret FamilyThese fresh Ferret family variants have been observed by researchers to be associated with the ‘Contagious Interview’ campaign. This attack prompts targets to communicate with an interviewer through a link which would show an error message - urging victims to install or update a communication software for virtual meetings.
These ‘updates’ would be disguised as Chrome or Zoom installers, like ChromeUpdate and CameraAccess persistence modules (really FROSTYFERRET_UI). These apps install a malicious persistence agent which runs in the background and steals sensitive data from the victim.
The latest Xprotect update will block most known variants which are disguised as macOS system files - including com.apple.secd (FRIENDLYFERRET). However, not all FlexibleFerret variants can be detected, as the malware landscape evolves so quickly.
The campaign has been observed as far back as 2023, and has been attributed to the well known Lazarus hacking group, which has been observed running several malicious job campaigns to trick jobseekers into downloading malware or trojanized remote access tools.
The data these attackers can access is dependent on the device they infect. Aaron Walton, Threat Intelligence Analyst at Expel points out anyone who falls victim to an attack using their work device, unwittingly puts their organization at risk.
"Though these bad actors typically target people through job offers, it’s fairly common that the individual will run the malware on a corporate device," he notes. "The attackers often know this and use it as a means to gain information from their target organization."
Malware protectionAt its origin, this is a social engineering campaign, so staying safe from these attacks is much easier if you can spot the signs. Social engineering attacks like phishing are often personalized, sometimes using information obtained from the dark web - obtained in a data breach, for example.
In this instance, the victims handed their information over as part of the ‘job application’ process, so thoroughly vetting any sites and companies you submit job applications to is really important.
Companies can't stop phishing attacks, and human error will always put organizations at risk, so to mitigate the risks every company, no matter what size, needs a robust cybersecurity strategy. Take a look at our SMB cybersecurity checklist to make sure you're covered.
"For organizations, it is important to have a strong defense-in-depth strategy—think of it as a multi-layered security fortress, where if one defense fails, another may stop the activity. That is, to defend the environment from many different angles. Employ endpoint detection, monitor networks, and empower employees to report suspicious activities", Walton comments.
As with most cyberattacks, vigilance is key. New malware threats are rising faster than ever, so being able to spot the signs can help limit the damage. If your device is suddenly much slower than normal, frequently crashes, or randomly reboots those are all signs that your device may be infected.
Another tell-tale sign is persistent pop-ups. These often bogus ads are pretty harmless themselves, but clicking on them might take you to a malicious site, and the ads are often a sign your device is infected. For a more detailed explanation of what to look for, check out our guide here.
For anyone who thinks this may apply to them, check out our list for the best antivirus software, which can be really helpful in locating and removing malware, as well as protecting against repeat infections.
If you do find malware on your device, make sure to remove the infected program immediately. Alongside this, it’s a good idea to disconnect from the internet to prevent the malware from spreading.
You might also likeChina is advancing its broadband infrastructure with its rollout of 50G-PON, a next-generation fiber technology capable of delivering speeds of up to 50Gbps (50,000 Mbps) downstream.
A newly published report by Dell’Oro Group, which gathers information from conversations with equipment vendors and publicly released tender award notifications, projects that PON equipment revenue will grow from $10.5 billion in 2024 to $12.1 billion by 2029.
While this growth will be driven largely by 10Gbps XGS-PON deployments in North America, EMEA, and CALA, China’s 50G-PON deployments place it ahead of the rest of the world. Last year, Omdia forecast that China will be the only commercial market for 50G-PON in 2024 and 2025, accounting for 93 percent of the global market and generating $1.55 billion in revenue by 2027.
Fiber to the RoomPON, or Passive Optical Network, is a fiber-optic technology that enables multiple users to share a single fiber connection using passive optical splitters. This design reduces the need for active electronic components between the provider and end users, lowering infrastructure costs, reducing power consumption, and improving network efficiency.
The 50G-PON ITU-T standard supports theoretical speeds of up to 50 Gbps downstream and up to 25 Gbps upstream, though current real-world deployments in China - led by China Telecom, its regional branch Shanghai Telecom, and ZTE - typically provide 10 Gbps all-optical access.
Beyond 50G-PON, China is also expanding Fiber to the Room (FTTR), which extends fiber-optic connectivity to individual rooms within homes and businesses. Unlike traditional fiber-to-the-home (FTTH) setups, which typically deliver fiber to a central modem and then rely on Ethernet or Wi-Fi for distribution, FTTR brings fiber-optic cables directly to each room, ensuring faster speeds, lower latency, and more stable connections.
Other highlights from Dell’Oro Group’s report include that cable distributed access equipment revenue will peak at $1.3 billion in 2028 as operators continue DOCSIS 4.0 and early fiber deployments.
Fixed wireless CPE is expected to reach its highest revenue in 2025 and 2026, driven by 5G sub-6 GHz and millimeter wave units, while Wi-Fi 7 residential routers and broadband CPE with WLAN are projected to generate $8.9 billion by 2029 as adoption grows among consumers and service providers.
“Quietly, broadband access networks are evolving into large-scale edge compute platforms, with the ability to enable service convergence far more quickly and easily than ever before,” said Jeff Heynen, Vice President at Dell’Oro Group.
“This evolution means that the revenue mix for broadband equipment is shifting over the next five years, with spending on traditional hardware and software now being supplemented by spending on AI and machine learning tools to facilitate convergence and service reliability.”
You might also like