I turned my Onn TV Box Into a VPN Router with Tailscale!

Tailscale has become a staple in my daily networking setup. As a personal VPN application that is free for personal use, it allows for the creation of a private network of devices that can be accessed from anywhere in the world. The primary utility is the ability to connect to home servers and hardware as if they were on a local network, without the security risks associated with exposing those devices directly to the internet.

In my latest Tailscale video, I explore the “subnet router” feature that allows any device on your local network to be accessible through Tailscale – even devices that can’t run the Tailscale client!

In the video I configured an Onn Google TV box to act as a subnet router. Most people use these devices strictly for streaming, but they can be deputized to serve as a gateway for an entire home network while still working as a TV box!

I opted for a wired Ethernet connection rather than Wi-Fi to ensure network reliability and to avoid congesting the wireless spectrum. While this specific hardware is limited to 100 megabit Ethernet, the technical process is identical for more powerful hardware like an Apple TV or a dedicated PC.

The configuration began within the Tailscale application on the Android TV box. Under the settings menu, I located the subnet routing option and entered my local network range, which in my case is 192.168.2.0/24. Once the route was added on the device, the next step required administrative approval through the Tailscale web-based admin console on a computer. After navigating to the “machines” section and selecting the TV box, I authorized the internal network route to allow traffic to flow through the device.

The results were immediate. To test the connection, I used a smartphone connected to a cellular network, entirely separate from my home Wi-Fi. By entering the local IP address of my HDHomeRun TV tuner into a browser on the phone, I was able to connect to the tuner through the TV box. It is worth noting that this method does not support network broadcasts, which means automatic discovery features for printers or tuners will not work. Users must know the specific local IP address of the device they wish to access.

To evaluate the throughput, I tested the connection from a location with a high-speed cellular signal. Through the TV box sitting in my basement, I was able to pull a sustained downstream speed of approximately 87 megabits per second. While the inbound speeds were notably slower—likely a limitation of the budget components in the Android box—the performance was more than adequate for streaming live high-definition TV from my HDHomerun tuner. Using VLC, I successfully tuned into a local broadcast signal with minimal latency and no need for transcoding.

This setup demonstrates that even inexpensive hardware can serve as a robust networking tool. Because TV boxes are typically left powered on and connected to the network, they make ideal candidates for this type of always-on routing. While those requiring higher bandwidth might prefer a more powerful host, this approach provides a functional and accessible way to manage a home network from afar.

Can This Unusual RTX 3060 GPU Justify Its $500 Price? Gaming and LLMs Tested!

I recently acquired the low-profile graphics card with an Nvidia RTX 3060 chip grafted on with 12GB of VRAM that’s been buzzed about in the tech press. The card is notable for its compact dimensions and the fact that it is entirely bus-powered with no additional power connector required. While this limits its maximum computational output, it offers a simplified installation process for small-form-factor systems. The package includes an additional mounting bracket for different case sizes and a basic manual. You can find one over at Newegg (compensated affiliate link).

But is it worth the $500 price tag? That’s what I explore in my latest video.

For my initial testing, I installed the card into a desktop featuring an Intel 10900K processor and 32GB of RAM. To my surprise, the standard Nvidia drivers installed without issue, recognizing the hardware immediately.

I began with a gaming test using Cyberpunk 2077. At 1080p resolution with the lowest settings, the card maintained roughly 70 frames per second. While playable, this performance falls short of what a standard, desktop 3060 typically delivers with similar settings. The card is fundamentally limited by the power it can draw from the system bus.

Synthetic benchmarks further illustrated these limitations. In the 3D Mark Time Spy test, the system returned a score of 4,846. This is significantly lower than a standard 3060 and even trails the performance of a much older GTX 1080. During a sustained stress test, the card achieved a stability score of only 85.8%, indicating significant thermal and power throttling.

But one potential use case for this card is local large language model (LLM) execution, as it features 12GB of video RAM. Using LM Studio, I tested the Gemma 4 12B-QAT model. While the model fit entirely within the VRAM, the generation speed was approximately 18 tokens per second. The GPU was fully utilized during this process, confirming that the power constraints also impact its effectiveness for artificial intelligence tasks. It is functional for this purpose, but notably slower than a standard version of the same GPU.

To see how the card performed in a multi-GPU setup, I moved it to a similar specced system equipped with an RTX 3080. The installation was seamless, and the two cards worked in tandem. When running a mixture-of-experts model like Gemma 4 35B-A4b, the system achieved 66 tokens per second because the primary compute tasks remained on the faster 3080 while using the 3060 for additional memory. However, when I switched to denser models like the 31-billion parameter Gemma 4 or the 27-billion parameter Qwen 3.8, performance dropped to between 6 and 8 tokens per second. In these scenarios, the 3080 was forced to wait for the slower 3060 to complete its portion of the workload, creating a significant bottleneck.

The build quality of the card, particularly the heatsink, is basic and likely contributes to the thermal issues observed during testing. While the additional 12GB of memory provides a way to load larger models that might not otherwise fit on a single consumer card, the performance trade-offs are substantial.

Given the high price point and the inherent power and thermal limitations, this card remains a difficult hardware choice to justify for most standard applications.

NextGen TV Update: 5G TV Backers Pledge No DRM — With a Catch..

I have been following the shifts in over-the-air television for some time, particularly the ongoing debate over the ATSC 3.0 standard. While major broadcasters continue to push for encryption and digital rights management (DRM) through NextGen TV, a different path is emerging for low-power television stations across the United States. A new proposal centered on 5G broadcast technology is currently before the FCC, offering a potential alternative that could change how signals are received on mobile devices and home tuners alike – and its backers are pledging not to encrypt the broadcast signal – but it’s important to read their fine print.

Learn more in my latest video!

Unlike traditional 5G cellular communication, which involves a two-way exchange between a phone and a tower, 5G broadcast operates as a one-to-many system. It sends data out from a single source to any device capable of receiving it, much like a traditional radio or television signal. Because it utilizes the same fundamental technology as modern smartphones, it could allow users to watch television on their devices without a cellular subscription or a data plan, provided the hardware is equipped with a compatible antenna.

One of the primary arguments for 5G broadcast is the promise of open access. HC2 Broadcasting Holdings, which owns a significant number of low-power stations, recently informed the FCC that their implementation would remain free-to-air without the use of DRM or paywalls. This stands in contrast to the current rollout of NextGen TV, where encryption has complicated the manufacturing of set-top boxes and slowed consumer adoption. Supporters of the 5G standard argue that it would be less expensive for manufacturers to integrate this technology into smartphones and boxes than it would be to comply with the complex encryption standards currently required by the major broadcasters.

The proposal includes specific limitations that warrant closer examination. While HC2 has pledged to keep signals unencrypted, they are seeking permission to broadcast at the bare minimum – only 480i—the standard definition used in the analog era. This would allow them to satisfy the legal requirement for broadcasting while leaving the majority of their spectrum available for other more profitable data services. In earlier filings, the group even suggested they might prefer to use their spectrum entirely for data casting rather than television, a move that the National Association of Broadcasters (NAB) has rightly criticized as failing to serve the public interest.

The tension between large and small broadcasters is evident in recent FCC filings. Sinclair Broadcast Group and the NAB have both voiced opposition to the 5G proposal, arguing that ATSC 3.0 should remain the single national standard. This creates a rift within the industry, as many low-power stations are members of the NAB yet find their interests aligned with the 5G alternative.

Meanwhile, Qualcomm has emerged as a significant ally for 5G broadcast, suggesting that the hardware required for smartphones to receive these signals is attainable and could be implemented without significant added costs to the consumer.

Beyond daily entertainment, there are potential applications for emergency services. Instead of receiving a simple text message during a crisis, a 5G broadcast signal could deliver rich media, such as live video feeds or detailed maps, to millions of devices simultaneously without straining existing cellular networks. This capability could satisfy the public benefit requirements that all broadcasters must meet in exchange for using the public airwaves.

The ATSC organization has taken a relatively neutral stance, urging the FCC to be cautious but not outright opposing the development of additional standards. As the FCC considers whether to allow further experimentation or a formal rulemaking process for 5G broadcast, the focus remains on balancing technological innovation with the public’s right to access free television.

Whether the market can support multiple standards or if the current friction will lead to a single unified path depends on how these technical and regulatory hurdles are resolved in the coming months.

The Cheapest 32GB Nvidia GPU You Can Buy for Local AI (Nvidia Tesla V100)

For the past couple of months, I have been exploring the practicalities of locally hosted AI models and applications. My primary objective to see what we can do with things we already own, or find cost-effective alternatives to a $5,000 Nvidia GPU.

In my latest video, I take a look at a relatively affordable 32GB GPU solution : an Nvidia V100 GPU pulled from a datacenter!

In an earlier video, I looked at a new Intel GPU with 32GB that retailed for $1,300. One of these pulled GPUs costs about half that. I got mine from Server Parts Deals who is also selling these cards on Amazon (compensated affiliate link).

Although the V100 lacks some of the specialized features found in modern RTX cards that newer models will begin to depend on, for now it provides ample performance from popular large language models (LLMs) like Qwen 3.8 27B.

Integrating a data center card into a consumer environment requires several specific modifications. These cards are designed for server racks where external fans force air through the chassis, meaning the card itself lacks active cooling. To address this, I sourced a specialized blower fan (compensated affiliate link) from an online marketplace to attach to the rear of the unit. Because I do not currently have a dedicated desktop for this project, I opted to use a small form factor PC with the GPU connected via an OcuLink interface.

The power requirements added a layer of complexity to the assembly. Data center cards use different power pin configurations than standard consumer hardware, necessitating a specific adapter cable (compensated affiliate link). Furthermore, because the external enclosure needs to power both the GPU and the high-draw blower fan, multiple power leads and adapters were required to ensure the cooling system remained operational at a constant speed. While some enthusiasts have developed variable-speed cooling solutions, I focused on the minimum requirements necessary to maintain stable operating temperatures under load.

The software installation process presented its own set of challenges. Initial attempts to configure the V100 on Windows 11 were unsuccessful; despite installing the correct data center drivers, the system consistently returned a hardware error, likely due to the OCuLink connection or BIOS issues. Consequently, I shifted the project to a Linux environment running Ubuntu 26.04. To streamline the configuration and driver installation, I utilized ChatGPT’s command-line Codex application to manage the technical hurdles and document the installation for future reference. This approach allowed for a more efficient setup of the necessary dependencies and libraries.

Once operational, I tested the system using the llama.cpp web interface, starting with the Qwen 3.8 27B dense model. When tasked with summarizing a few PDF documents, the V100 achieved a pre-fill speed of 50 tokens per second and a generation speed of 45 tokens per second. The 32GB of VRAM allowed for a large context window of over 96,000 tokens, enabling the model to process a 70-page document without exceeding the hardware’s limits. Even as the context window filled to 80% capacity, the generation speed remained a functional 24 to 30 tokens per second while drawing approximately 270 watts of power.

Performance shifted significantly when testing a “mixture of experts” model, such as Gemma 4 26B-A4B. Because these models only activate a fraction of their parameters for any given task, the processing speeds were much higher. The system reached pre-fill speeds of 1,100 tokens per second and output generation at over 90 tokens per second. These results indicate that for text-based local AI tasks, this older data center hardware remains competitive with much more expensive modern alternatives.

Beyond text, I tested the card’s ability to generate images and video using ComfyUI. While the V100 can run modern models like Flux for images LTX 2.5 for video, the process is considerably slower than on contemporary hardware. Generating a single 1024×1024 image took approximately one minute, while a five-second video clip at 720p required three and a half minutes of processing. By comparison my Intel card could do the same job in under a minute and a half. Long videos were much slower vs. the newer Intel card in my testing.

Newer cards from Intel and Nvidia have specific optimizations for these workflows that the V100 lacks, as it was originally released in 2017. For example, the exceptional MiniMax H3 model was too slow to be usable on the V100.

There are clear trade-offs when using a decade-old data center GPU. Many emerging AI features expect hardware capabilities that simply did not exist when this card was engineered. While it handles current language models with ease, it may not support the next generation of multimodal AI tools as effectively. However, for those specifically seeking a large context window and high VRAM capacity for local text processing on a budget, these secondary-market cards offer a viable path forward.

My Theory on Why YouTube Changed View Counts..

YouTube recently updated its platform to change the way video views are calculated. Under the new system, a view is recorded as soon as a single frame of a video is watched. Previously Youtube measured a level of genuine engagement before increasing the view count. My theory? YouTube is doing this to try and capture more of the revenue flowing to creators.

Check out my latest video where I present the evidence!

On my own channel, this change has resulted in a noticeable increase in reported views. Typically, my daily traffic ranges between 15,000 and 30,000 views, driven largely by search-focused, evergreen content such as printer reviews. Now I’m seeing 45,000 or more.

While the higher numbers might initially appear positive, they have rendered traditional performance metrics less reliable. For instance, a recent video I produced regarding local AI ranked highly in my internal analytics based on view count. However, a closer look at the advanced analytics—which now requires several additional steps to access—revealed that the “engaged views” were actually on the lower end of my typical performance. The public-facing view count no longer reflects actual audience interest.

YouTube says that this change was implemented to align metrics between long-form videos and Shorts. However, the result is that the most meaningful data is now obscured within advanced analytics tools. This may be an attempt to reclaim a cut of revenue that currently flows directly to creators.

Many creators work with outside agencies to secure direct advertising deals. These agreements are often predicated on public metrics, specifically the average view count per upload. With the recent change, these public figures have become less indicative of a channel’s actual influence. This creates a challenge for agencies trying to distinguish between legitimate audience engagement and accidental frame clicks. It also lowers the practical value of the RPM (rate per thousand views) that has served as a benchmark for the industry. From YouTube’s perspective, having an engaged view be the metric was too valuable a measurement to share publicly.

There is evidence to suggest this timing is intentional. Shortly after the view count change, the platform rebranded its internal agency as YouTube Creator Partnerships, signaling a renewed effort to connect brands with creators directly. By positioning themselves as the sole gatekeepers of accurate engagement data, they provide a compelling reason for brands to bypass outside agencies in favor of their own proprietary systems.

The recent integration of Amazon into the YouTube Shopping affiliate program further illustrates this shift toward a controlled revenue pipeline. For years, creators have used direct Amazon affiliate links in their descriptions because the system was efficient and did not require sharing a commission with the platform. I’ve also noticed that viewers are far more likely to click a link that I provided vs. one that appears through YouTube Shopping.

But just recently, I observed an experiment where external affiliate links in video descriptions were rendered unclickable, while links generated through the official YouTube affiliate program remained functional. This suggests a move away from the open web toward a walled garden where the platform can facilitate—and take a cut of—every transaction.

By devaluing public data and testing restrictions on external linking, the platform is exercising its position as a dominant media entity to maximize its own revenue. While these changes may not prevent creators from seeking independent partnerships, they certainly make those deals more difficult to quantify and execute. Brands may feel compelled to funnel their creator sponsorships through YouTube to have a better window on performance.

Local AI On an Old 8 GB GPU?

The current landscape of artificial intelligence often suggests that high-end, increasingly expensive hardware is a prerequisite for meaningful performance. However, much can be accomplished using hardware that many people already own. I recently spent time working with a five-year-old gaming laptop equipped with 16 gigabytes of system memory and an Nvidia 3070 GPU with 8 gigabytes of video RAM. While this is a far cry from a modern workstation, it provides a capable environment for running local language models if the settings are adjusted correctly.

Check out what it can do in my latest local AI video!

The primary challenge with an 8-gigabyte GPU is balancing the size of the model with the context length, which is essentially the memory the model uses to track a conversation or analyze a document. For my testing, I utilized the Qwen-3-8B model, an 8-billion parameter model that occupies about 5.76 gigabytes of memory. This leaves a small but workable margin for context.

In the LM Studio application, I set the context limit to 25,600 tokens (compared to 1 million tokens on frontier models like ChatGPT and Gemini). To keep the performance steady at around 35 tokens per second, I employed several technical tweaks, including unified KV cache, flash attention, and 8-bit KV cache quantization. This quantization is particularly important because it compresses the context memory, allowing more data to fit onto the GPU.

To test the practical utility of this setup, I tasked the model with analyzing a term paper I wrote in college nearly 30 years ago. Although the text file was small, the process of analyzing it for patterns can consume significant memory. I inserted several “traps” into the text—phrases like “it’s a trap” and “Han shot first”—to see if the model would notice them while summarizing the content. The model processed the prompt in a few seconds, provided an accurate summary without hallucinations, and correctly identified the out-of-context phrases.

Moving beyond simple text analysis, I attempted to use the laptop as an AI server for more complex tasks like coding. By connecting a secondary machine to the laptop using an open-source coding harness called OpenCode, I tried to extract a list of public officials from a website and format the data into a CSV file. This is where the limitations of an 8-gigabyte system become more apparent. The model struggled to process the raw HTML of the webpage due to the limited context window. Once I simplified the input to plain text, the model successfully formatted the data, including identifying vacancies in a specific district. It required more manual intervention than a larger 70-billion parameter model might, but it eventually achieved the desired result.

Coding tasks showed similar limitations. When I asked the model to write a Space Invaders game in HTML, it produced a functional interface where a player character could move, but it failed to include enemies or game logic on the first attempt. Even after a second prompt to fix the errors, the logic remained incomplete. Smaller models often lack the reasoning capabilities required for complex, one-shot coding successes, frequently requiring a more iterative approach and careful management of the conversation history.

Image generation is also possible on this hardware. Using the SDXL model through a templated interface, I was able to generate local images in approximately 30 to 32 seconds each. While this setup cannot handle the video generation tasks possible on high-end cards, the static image quality is respectable. Achieving the right output required some trial and error with prompts; for instance, the model initially struggled to generate a dog in a cockpit until I removed gender-specific language that seemed to be steering it toward human subjects.

The efficiency of these local models has improved significantly over the last year. For those who may have tried local AI in the past and found it lacking, the current generation of small-parameter models offers much more utility on aging hardware.

While 8 gigabytes of video memory requires careful optimization and realistic expectations regarding context and complexity, it remains a viable entry point for those looking to keep their data local and make use of the machines they already have. If you’re willing to give up performance, AI applications like LM Studio can also use system memory which can allow for larger context memory and AI models.

GMKTEC M3 Pro Review – DDR4 and an Older Intel Chip Make it Somewhat Affordable

I have been searching for lower-cost options in the mini PC market recently and found another candidate that meets the criteria : the GMKTec M3 Pro. This device is powered by an Intel i5-13500H processor and DDR 4 RAM. This is older technology, but at this point older is the only way to reach affordability in the current environment.

Check it out in my latest Mini PC review!

The unit is priced at $329 for a barebones configuration, which excludes RAM and storage. For those preferring a pre-configured system, Amazon has some options preconfigured with RAM and storage (compensated affiliate links).

The M3 Pro features a compact design with a build that utilizes metal on the sides and plastic for the top and bottom panels. On the front of the device, there are two 10 gigabit per second USB-A ports. The rear houses two additional USB-A ports, though these are limited to 5 gigabits per second. For display connectivity, there are two HDMI outputs capable of 4K resolution at 60Hz. A USB Type-C port is also present and supports a third 4K 60Hz display. However, it is important to note that this is not a Thunderbolt or USB4 port. Due to the older chipset, the Type-C port is limited to USB 3.2 specifications, meaning data transfers are capped at 10 gigabits per second.

The M3 Pro includes a 2.5 GB Ethernet port that performed to its rated specifications during my testing, maintaining full potential on both downstream and upstream traffic. The Wi-Fi 6 performance was less consistent, reaching about half a gigabit per second on the downstream and approximately 350 megabits per second on the upstream. Internally, the system is upgradeable. Removing the bottom panel reveals two RAM slots supporting up to 64 GB of DDR4 memory. Because these Intel chipsets favor dual-channel memory, it is advisable to install RAM in pairs. Storage is handled by two NVMe slots: one 2280 slot and a secondary 2242 slot for shorter drives.

In terms of performance, the M3 Pro handles specific workflows with varying degrees of efficiency. When testing local AI using the Gemma 26B-A4B mixture of experts model, the system generated approximately 12 tokens per second. This is usable but noticeably slower than modern AMD-based systems equipped with faster DDR5 memory, which can exceed 20 tokens per second. Video editing in DaVinci Resolve is manageable for basic 4K 60Hz tasks, such as simple transitions or stringing clips together. However, more complex projects involving heavy effects cause the system to slow significantly. The lack of a Thunderbolt port also precludes the use of an external GPU to bolster this performance.

General computing is where the machine feels most responsive. Web browsing is nearly instantaneous, and the device handled high-bitrate 4K 60Hz YouTube playback without dropping frames. It achieved a score of 29.6 on the Speedometer benchmark, which is consistent with other systems in this category with current generation chipsets.

Gaming performance is comparable to a Steam Deck. Running Cyberpunk 2077 at 1080p on the lowest settings resulted in about 20-25 frames per second, though lowering the resolution to 720p would likely push the frame rate above 30. It serves as a capable box for 8-bit and 16-bit emulation and can handle some GameCube titles. In the 3DMark Time Spy test, it scored 1,644, which aligns with expectations for this processor generation.

Thermal management can be adjusted through a high-performance mode in the BIOS. In this mode, the fan is more aggressive, and the system passed a 3DMark stress test without significant thermal throttling, maintaining a temperature of roughly 69 degrees Celsius under sustained load. While the fan is audible during heavy tasks, it is not particularly loud compared to other mini PCs, and it remains largely silent performing basic computing tasks. Power consumption sits at 10 to 11 watts at idle and reaches about 75 watts under a full load.

Linux compatibility is a strong point for this hardware, likely due to the maturity of the older chipset. Running Ubuntu 26.04, the experience was snappy and all hardware components—including video, audio, Ethernet, Wi-Fi, and Bluetooth—were detected without manual configuration.

While the M3 Pro lacks the hardware required for demanding local AI or high-end video production, it remains a functional choice for general office work, home server duties, or use as a dedicated Linux desktop. The trade-offs in port speeds and memory technology are reflected in the lower entry price, making it a utilitarian option for users with standard computing needs.

Mini Haul! A Surplus AI GPU and New Media Server Project Guts

I am currently beginning two new hardware projects and recently acquired several components to get them underway. While I have the core components ready, I am still sourcing specific parts and looking for technical feedback on the final configurations.

In my latest haul video, I unbox the first two pieces of these projects!

My first project involves a necessary upgrade to my home media server. My current setup utilizes two USB drive arrays containing four 4TB drives each. This configuration creates too many potential points of failure, so I am moving toward a proper self-constructed NAS device with drives connected to the motherboard via SATA. I have settled on a Topton motherboard featuring an integrated Intel N150 processor (compensated affiliate link). This board requires DDR5 RAM and supports eight SATA drives through the use of two breakout cables. Because my Unraid setup requires 10 total drives to accommodate two parity drives, I will need to find a reliable SATA PCIe expansion card to support the additional two units.

The motherboard is equipped with 10-gigabit LAN and two 2.5-gigabit connectors, providing significant networking capacity. It also features two NVMe slots, which I intend to use for cache and application storage for Docker containers. My primary challenge now is identifying a case that can comfortably house and cool 10 drives. I am also seeking a SATA expansion card that is proven to be stable for 24/7 operation. Transitioning the software should be straightforward, as Unraid allows for moving a boot drive between different hardware sets with minimal reconfiguration.

The second project focuses on local AI workflows using repurposed data center hardware. I purchased a pulled Nvidia Tesla V100 with 32GB of memory (compensated affiliate link). This card was originally designed for high-density server racks and lacks any built-in active cooling. In its original environment, external high-pressure fans move air through the card’s heatsink. To make this functional for my purposes, I have sourced an aftermarket cooling fan and a dual eight-pin adapter cable to connect the card to a mini PC via Oculink.

While the Tesla V100 is older technology, its 900 GB/s of memory bandwidth makes it a potential candidate for running large language models like Qwen 3.8 locally. For those considering similar hardware, the 32GB PCIe version is the most versatile option, though it is important to be aware of the cooling and power requirements. It may struggle with modern image or video generation tasks compared to newer consumer-grade cards, but its high VRAM capacity at this price point remains a significant advantage for text-based models.

I am interested in any insights regarding the cooling of enterprise GPUs in a home setting or recommendations for a 10-bay chassis that maintains small-form-factor efficiency. I will be moving forward with the builds as soon as the remaining cooling components and expansion cards arrive.

Disclosure: I paid for the parts featured in this video with my own funds.

Is This How Microsoft Ends Physical Xbox Games?

Following the recent news regarding the shift away from physical media in the gaming industry, Microsoft has introduced a disc-to-digital program that allows users to insert a physical game into an Xbox with an optical drive and register that title for digital use. While welcomed by gamers, does this signal Microsoft’s intention to ditch physical games in the near future?

In my latest video, we take a look at how this feature works and what the future might be.

This system utilizes the unique serial numbers embedded in the DRM of Blu-ray discs. When a disc is inserted, that specific serial number is assigned to a Microsoft account, enabling any console associated with that account to play the game. If the disc is loaned to a friend, the license shifts to their account until the original owner re-inserts the disc or acquires another copy. In many ways, this mirrors the licensing system used by Nintendo Switch cartridges, where the hardware serves primarily as a validation key for the software license.

The faint barcode is the unique identifier of this Xbox Disc

However, this solution does not extend to legacy hardware. Xbox 360 and original Xbox games were manufactured on DVDs, a format that did not include the serialization found on Blu-rays. Consequently, there is currently no clear path for maintaining backward compatibility for these older titles should Microsoft fully transition away from internal optical drives.

Share of Physical Sales

Based on current market trends, that transition appears increasingly likely. Recent data shows that Xbox accounts for just over 4% of the total physical game market. In a market that has seen a steady decline for years, such a small share provides little incentive for manufacturers to continue including expensive optical drives in their hardware.

Consumer behavior also supports this shift toward digital-only ecosystems. Leaked sales data from Microsoft’s legal proceedings with the FTC revealed that the Xbox Series S, which lacks a disc drive, has significantly outsold the more expensive Series X. While Microsoft initially forecasted that the Series S would capture about 65% of its console sales, the actual figure was closer to 75%.

Further leaked documents from 2022 detailed a product roadmap featuring a mid-generation refresh for the Series X, code-named “Brooklin.” This proposed device was designed without an optical drive, signaling an intent to move the entire Xbox lineup to an all-digital format. While economic factors and shifting memory costs have since altered these specific release plans, the underlying strategy remains centered on a digital delivery model.

The growth of Game Pass has further diminished the necessity of physical media for many players. With tens of millions of subscribers paying a monthly fee to access a library of titles—including new releases—the need to purchase individual physical games has reached an all-time low. When combined with the high adoption rate of the Series S, a significant portion of the Xbox user base has already moved away from physical discs. This is reflected in the broader industry data, which shows physical game sales dropping from a peak of 292 million units in 2009 to approximately 37 million units today.

While the mass market for physical games is fading, a dedicated subset of collectors and enthusiasts remains. Independent publishers like Limited Run and Strictly Limited continue to find success by licensing titles for physical release. To support this community, Microsoft could consider allowing the use of external Blu-ray drives via USB. Since the company already manages the database of serialized game keys, providing a path for external hardware would satisfy the needs of those who value physical ownership without requiring Microsoft to include a drive in every console. Maintaining this level of flexibility would serve as a gesture of goodwill toward long-time fans who wish to keep their physical libraries viable in an increasingly digital landscape.

Caldera Music for Plexamp – TV App, Multi-room Sync, and Amazing Visuals! (sponsored post)

PlexAmp serves as a functional gateway for those who maintain personal music libraries on a Plex server, particularly for listeners who prioritize lossless audio. Recently, a new suite of applications called Caldera Music has emerged to expand how this music is distributed throughout a home. Developed by Plex co-founder Elan Feingold, this software suite focuses on multi-room audio synchronization and dedicated interfaces for television-based devices like Apple TV and Android TV.

Check it out in my latest video!

The Caldera Music suite consists of three primary components, starting with a headless player designed for Linux-based devices. This allows for a low-impact installation on hardware such as a Raspberry Pi or a mini PC. The setup process involves executing a series of commands in a terminal to connect the device to a Plex server and establish the default audio output. Once configured, the headless player operates in the background without a user interface. Instead, it functions as a casting destination. Using a standard Plex Amp client on a mobile device, a listener can redirect the audio stream to the headless player, effectively turning any Linux device into a networked audio endpoint. The player consumes only about 34 megabytes of RAM and is very low impact.

For more complex setups, the suite includes a desktop player for Mac, Windows, and Linux. Unlike typical media players, this application remains minimal, residing in the system tray or menu bar rather than the taskbar or dock. Its primary utility lies in its ability to route audio to multiple destinations simultaneously. A user can cast music from Plex Amp to the desktop application, which then synchronizes the signal across various outputs. This enables a multi-room audio scenario where the same track plays in unison across different parts of a house, such as a desktop computer and a remote headless player.

Visual presentation is another focal point of the suite, particularly within the big-screen application for Apple TV and Android TV. The TV apps function as standalone clients, allowing users to browse their entire library, access artist radios, and view metadata directly on their television. It incorporates a library of over 600 visualizations that sync with the music. These visual elements are customizable; users can link specific visualizations to certain tracks, mark favorites, or disable ones they do not wish to see. For those who prefer a more understated experience, these features can be turned off entirely to focus on the playback data and progress meters.

The technical specifications within the desktop and TV applications cater to those interested in high-fidelity audio. The software supports adjustable sample rates to match the source material, which is a necessary feature for high-res lossless playback. It also includes various filters and supports multi-channel audio, such as 5.1 or 7.1 configurations, ensuring that complex audio files are transmitted correctly from the server to the final output. Users can monitor the audio path through a debug interface that tracks the signal as it moves from its original format, such as FLAC, through the device’s audio architecture to the HDMI or analog output.

The existence of Caldera Music, which is available for free highlights the capabilities of the open Plex API. By allowing external developers to hook into the server architecture, the platform enables the creation of specialized tools that address specific user needs, such as synchronized multi-room playback and advanced television interfaces.

Caldera Amp is a great example of the growing ecosystem where the central server acts as a foundation for a variety of third-party enhancements and niche media applications. As more talented developers explore these integration possibilities, the ways in which personal media libraries are accessed and distributed are likely to become increasingly flexible.

Disclosure: Plex sponsored this video as part of my monthly series, however they did not review or approve this content prior to uploading.

What Playstation Boycotters Should Be Asking For..

Sony’s plan to end the manufacturing of physical video games in the coming years has triggered a boycott among PlayStation enthusiasts who want to retain the option of owning their media. Despite the negative press surrounding this decision, Sony has signaled to the industry that it intends to proceed with this transition. While a boycott may face challenges given the current market trajectory, there are practical avenues where consumer energy might be directed to preserve physical media in a different form.

In my latest video, I lay out a practical solution that might help maintain physical ownership for the small subset of the market that wants it.

Current data from Circana regarding physical media market share in the United States shows that Nintendo currently holds 63% of the market, while PlayStation accounts for 32%. While PlayStation’s share is notable, it represents a portion of a market that is shrinking rapidly.

At the peak of physical media in 2009, approximately 292 million physical game units were sold in the U.S. Projections suggest that by June 2026, that number will drop to 37 million units across all platforms. Within the PlayStation ecosystem, the decline is even more visible; only seven games have sold more than 100,000 physical copies so far this year. In contrast, 100 different games reached that same milestone in the first half of 2008.

Despite the move toward digital-only consoles, technical solutions exist that would allow for continued physical ownership. For instance, the Xbox Series S features a high-speed expansion slot capable of supporting modern game data. If flash memory prices decrease, it is possible for publishers to release games on cartridges that could be inserted directly into these slots, allowing for traditional ownership and resale.

Another option involves the use of external hardware. In the past, Microsoft produced an external optical drive for the Xbox 360 to support the HD DVD format, which connected simply via USB. Modern console manufacturers could allow consumers to connect external optical drives to digital-only machines. This would enable gamers to buy physical discs, copy the data to their hard drives, and validate ownership without requiring an internal drive to be built into every console. This approach would be a cost-effective manufacturing compromise that maintains a path for physical collectors.

The infrastructure for physical releases is also being sustained by a specialized cottage industry. Independent publishers like Limited Run, Strictly Limited Games, and Iam8bit have built viable businesses by producing physical versions of games that were originally digital-only. This market has grown to include larger titles and elaborate collector’s editions.

While Sony currently mandates that it must manufacture the discs for its own platform, the underlying technology remains the Blu-ray disc. Since Sony will likely continue producing Blu-ray discs for the film industry, the capacity to press game discs remains, provided the platform holders continue to grant permission to these independent publishers.

The survival of physical media in other industries offers a potential roadmap for gaming. In the music industry, vinyl records have seen a significant resurgence, even outselling CDs, because a dedicated segment of the population values the physical object and the specific quality of the medium. Similarly, the market for 4K Ultra HD movies remains stable because the quality often exceeds what is available via streaming. In both cases, the entire consumer base does not need to participate to make the market sustainable; a dedicated minority is sufficient.

The business case for returning to standard internal optical drives in every console may no longer exist, but there is a clear case for maintaining the software support that allows physical media to function. By encouraging console manufacturers to support external hardware and independent physical publishers, the community can preserve the ability to hold a game in their hands. Success in this area will likely depend on finding a middle ground that acknowledges the shift toward digital while protecting the rights of those who prefer to own their software.

Local AI Video and Image Generation on the Intel’s Low Cost 32GB GPU!

The Intel B70 GPU with 32 gigabytes of video RAM remains one of the least expensive cards on the market (affiliate link)for its capacity, and it has served as the centerpiece for my recent explorations into local AI. My current setup utilizes this card alongside an Oculink dock (affiliate link) and a Mini PC running Ubuntu 26.04.

My latest video takes a look at running Comfy UI on the Intel B70 to generate video and images – the results are staggering for local hardware!

In one test using the Minimax H3 model, a five-second video clip took approximately two minutes and 36 seconds to generate. This particular output was rendered at 480p resolution. While the hardware can handle up to 1080p, performance decreases as the resolution increases.

In my practical applications, the model has shown varying degrees of success with complex prompts. I experimented with making a spoof of a McDonald’s commercial and several Star Trek-themed clips. The system effectively handles motion and visual consistency, but it faces challenges with dialogue. In clips involving two characters, the model often struggles to keep voices distinct, sometimes mixing them together. It also requires precise prompting to get exactly what you’re looking for. But that’s no different than the frontier video generators.

Beyond narrative clips, the hardware handles utility-focused tasks like creating looping backgrounds for motion graphics. It also supports image-to-video generation, where a single still image is animated, or two images are used as “before and after” anchors for the model to bridge with video. The node-based interface of Comfy UI provides a level of control that allows for complex orders of operation, which can help refine these outputs.

Static image generation is notably faster. Using the Flux 2 Klein 9B model, I generated a 1024×1024 image in 35 seconds. I also tested a more complex request: a three-panel storyboard featuring a consistent character across different frames. The model maintained the character’s appearance and clothing throughout the sequence, which is a significant requirement for practical design work. This specific model requires an API key for access (likely for watermarking output), though it runs locally.

Configuring these tools on Linux requires some technical effort. I have relied on frontier models like ChatGPT’s Codex command line to assist with the configurations necessary to bridge the gap between the software and the Intel hardware.

One of the benefits for doing image and video generation locally is that it provides a sandbox for iterating on prompts without consuming paid cloud tokens. A user can refine their ideas on local hardware and then take those perfected prompts to a high-end cloud model for a final, high-resolution render.

The current state of local AI feels increasingly more accessible compared to just a year ago. The quality of these open-weight models is approaching what industry-leading frontier models were producing in the recent past. As these models become more efficient, the entry barrier will likely continue to drop. My next objective is to borrow an AMD-based GPU rig to observe how it handles these same workloads and whether it offers any specific performance advantages over the Intel setup as the AMD AI Pro 9700 is now closer in price (affiliate link) to the Intel board.

Contraband Camera? Banned DJI Cameras Available on the Grey Market

The federal government’s efforts to restrict specific consumer electronics are beginning to manifest in tangible ways for American consumers. While earlier regulatory actions primarily targeted telecommunications infrastructure, the scope has expanded to include portable consumer devices.

In my latest video, I talk about how I acquired a DJI Osmo Pocket 4P, a compact vlogging camera that currently sits on the list of prohibited items for commercial sale in the United States. I also take a look at the status of DJI’s legal efforts to get themselves off the list.

The Osmo 4P features a one-inch sensor capable of 4K resolution at 60 frames per second and an integrated mechanical gimbal, offering a more streamlined experience than the mobile phone rigs I have used previously.

Because the Federal Communications Commission (FCC) ban prevents new DJI products from being officially approved for the U.S. market, obtaining the camera required an unconventional approach. A friend purchased the device at retail price while on a work trip in Shanghai and brought it back into the United States in its original packaging.

This highlights a significant gap in the current enforcement; while commercial import for resale is restricted, personal possession and individual transport across the border have, thus far, continued without issue. On the secondary market, this same package is often marked up significantly, with prices reaching nearly $800 on some sites due to tariffs and the scarcity of domestic supply.

There is a specific technical hurdle for those attempting to use these devices domestically. The camera must be activated through a mobile application before it can be used. If the software detects a United States location during this initial setup, the device may refuse to activate. To bypass this, my unit was activated in China before being transported. Once that process was complete, I was able to pair it with my local app and set the interface to English.

For those who prefer to avoid these complications and maintain access to warranty support, companies like Insta360 offer hardware with similar specifications (compensated affilite link) that remains legal for sale in the U.S., though often at a slightly higher price point.

The legal tension surrounding DJI stems from the 2021 National Defense Authorization Act, which tasked the Department of Defense with identifying companies linked to the Chinese military. DJI was added to this list, a designation the company has challenged in court. A recent appeals court ruling noted that the government had not provided a public rationale for this classification, as the relevant evidence was classified and never presented to the originating court. The appeals court has now signaled that the government must present documentation to support its claims if the designation is to remain.

However, a more recent legislative development presents an even larger obstacle. The 2025 National Defense Authorization Act, signed into law in December 2024, specifically named DJI and Autel Robotics in the legislation. The law dictated that if a required report on these companies was not completed by the Department of Defense within a year of enactment, both companies’ products would automatically be placed on the FCC’s covered list. Because that report was not finished on time, the law forced the FCC to ban both Autel and DJI from the US market. This mandate includes not only drones but also “surveillance equipment.” Because this is a statutory requirement passed by Congress, the FCC does not have the authority to remove these companies from the list on its own.

This leaves DJI products relegated to the grey market for the foreseeable future. While many users are currently bringing these devices across the border from places like Canada or Mexico, or receiving them via international shipping, there is an inherent risk. The government possesses the legal authority to confiscate items on the covered list at the border. While enforcement has not yet reached the point of individual seizures for personal items, the long-term status of these electronics remains uncertain as the legal and political landscape continues to shift.

A 32GB GPU for a Third of the Price? Intel’s B70 for Local AI

Local AI has made tremendous strides over the last few months and I’ve been looking at the most affordable ways to implement it, like using an old Mac, or pairing two GPUs together to maximize memory. In my latest video, I sought out the lowest cost 32GB GPU solution I could find and landed on the Intel B70 (compensated affiliate link).

See it in action here!

While prices have fluctuated recently—this GPU cost a few hundred dollars less just a few weeks ago—it remains a practical alternative to higher-end Nvidia options for those operating on a more modest budget. It won’t be as fast but it’s fast enough for most workflows.

The primary advantage of the 32-gigabyte capacity is the ability to run dense models, such as the 27-billion-parameter Qwen 3.8, entirely on the GPU. This prevents the system from having to offload data to slower system RAM or the CPU. In my testing, the setup generates approximately 25 tokens per second, rarely dropping below 20 even when the context window is heavily loaded. For example, I used the model to summarize a complex appeals court ruling regarding DJI litigation and the result was every bit as good as what you might get from ChatGPT or Gemini. The system processed the legal text and generated a structured summary and a chart without relying on cloud resources.

The hardware configuration I demoed involves an Oculink connection to a MinisForum Mini PC equipped with an AMD Strix processor. While a mini PC of this caliber might be more power than necessary when the GPU is doing the heavy lifting, it served as a convenient platform for the project. The operating system is Ubuntu 26.04, which has proven stable for these applications.

Normally I’d do a how-to in a video like this to get things working. But these days, it’s easier to just let AI do most of the heavy lifting especially given how frequently things are changing in this space. For this project I had ChatGPT’s CODEX command line interface verify the hardware install and find the appropriate drivers. This approach has allowed me to resolve configuration issues that might otherwise be time-consuming, resulting in a more efficient setup.

Practical applications for this local setup extend into coding and data management. I developed a tool called “The Gadget Tracker,” which filters RSS feeds from technology publications. The Qwen model analyzes article text to determine if they pertain to hardware gadgets rather than general industry news or software algorithms. Valid entries are then stored in a local SQLite database for review in a to-do list format. I also demoed Qwen 2.8 modifying the application’s user interface, adding social media images to the display cards and adjusting UI elements. The model’s reasoning process is perhaps a little more methodical than a frontier model with more resources; it often takes time to contemplate its plan before executing code, which helps minimize errors in the final output.

Another test involved extracting data that is difficult to capture through traditional copy-and-paste methods. I tasked the model with gathering names and email addresses for the New York City Council from their official website. Because the emails were embedded within the site’s source code rather than displayed directly on the page, a simple scrape wouldn’t work. The local model used a curl command to pull the HTML, identified the relevant strings, and formatted the data into a CSV file in under two minutes.

There are trade-offs to this local approach, particularly regarding power and acoustics. The GPU fan is audible even when the system is idle, and the card consumes about 55 watts just sitting with a model loaded. Under a full processing load, that consumption rises to over 320 watts.

Despite the power demands, the ability to process 96,000 tokens of context locally provides a level of privacy and unlimited usage that cloud services cannot match. The technology in this space is moving quickly, with software updates frequently unlocking better performance from existing hardware. This specific configuration has reached a point where it handles the majority of the tasks I previously delegated to proprietary models, proving that local hardware is increasingly capable of meeting professional demands. My focus now shifts to optimizing the power draw while continuing to test how these dense models handle more complex, long-term projects.

I added 16TB to my Unraid Server From a Storage Bin!

The current configuration of my media server is running out of space and with storage prices at an all time high, I needed to find a more creative solution. So, I went rummaging through my basement and dug up 16TB spread across a number of older, unused hard drives.

In my latest video, I demonstrate just how flexible the Unraid NAS OS can be when adding a bunch of random drives to the mix!

For the past several months, the system has operated on a Beelink ME Mini connected to a four-bay USB drive enclosure. This setup utilized four 4TB drives, with three dedicated to data storage and one acting as a parity drive for data protection. With only 650GB of space remaining and a collection of new media files to move over, an expansion was necessary.

To address the shortage, I began a migration to Beelink’s ME Pro (which has two built in SATA slots along with faster 5 gigabit ethernet) while repurposing storage hardware I already owned. I plan to move my Unraid’s parity drive used for data protection from the USB array to one of the SATA slots. This will free up another slot in the USB array for storage. I also plan to add a second parity drive to the ME Pro’s other SATA slot to add additional protection.

I pulled several 4TB drives that had been sitting idle in a storage container and integrated them into a second four-bay drive array. This is where Unraid is very useful, as it allows for storage expansion just by adding drives without the need for complex RAID calculations and rebuilds.

Find the storage array and Mini PC on Amazon (compensated affiliate link)

While using multiple USB enclosures introduces potential points of failure compared to a traditional server tower with internal SATA connections, the previous USB-based configuration functioned without interruption for over nine months. The only thing it went out of service for was Unraid OS updates.

The migration process began with identifying the serial number of the existing parity drive. Because all the drives in my collection are the same make and model, tracking the serial numbers ensures that the parity drive is not accidentally assigned as a data drive, which would result in data loss. I also created a backup of the Unraid boot drive configuration to a zip file as a precaution.

After preparing the software, I moved my existing NVMe drives from the old mini PC to the new hardware. One NVMe drive serves as a cache for the array, while the other is dedicated to running Docker containers, such as Plex, to ensure higher performance for applications. Once these were installed, I moved the Unraid USB boot drive to the new machine and connected my existing USB arrays.

Once I booted up on the new system, Unraid spun up without issue. This is what I love about Unraid – hardware migration is usually a simple swap of drives from one PC to the next.

To expand the storage, I connected the second drive array and assigned the four additional 4TB drives to the pool. Unraid requires new drives to undergo a clearing process, which involves writing zeros to the entire disk to maintain parity integrity. This process took approximately seven hours for the 4TB units. During this time, the server remained functional, and existing files were still accessible. Once the clearing was finished, I formatted the drives and updated the media share settings to include the new disks.

One last note on Unraid parity : parity drives need to be the same size or larger than your largest storage disk. So in my case 4TB drives are as large as I can go at the moment. If I wanted to add a 6TB drive, I’d need to ensure my parity drives are 6TB or more.

Along with relocating my parity drive, this hardware transition has increased the available storage to 20TB. While the current setup relies on USB enclosures, it serves as a practical way to utilize existing hardware. The next logical step in the evolution of this server will likely involve moving away from USB entirely in favor of a motherboard with enough native SATA ports to house all drives within a single chassis. But for the time being, the system is back online with the capacity required for my current storage needs.

Disclosures: Beelink sent the Mini PCs and Terramaster sent the new array to the channel free of charge. The Wavlink array came in free of charge through the Amazon Vine program. I paid for the Unraid license with my own funds. This review was not sponsored nor has any brand reviewed or approved the video prior to uploading. All opinions are my own.

ModRetro’s M64 is a Nintendo 64 FPGA Console for the Millennial Masses

This has been a good year for fans of the Nintendo 64 with two new consoles replicating the original console in 4k. First we had the Analogue 3D which continues to get improvements via firmware updates, and now we have the new ModRetro M64 which costs a bit less.

In my latest video we take ModRetro’s new console out for a spin.

This new device, like Analogue’s console, utilizes a field programmable gate array (FPGA) processor, allowing it to replicate original hardware behavior with a high degree of accuracy.

For many, the primary appeal of a device like this is the ability to connect to a modern television without the signal degradation or connectivity issues common with original hardware. While the output is technically 4K, the image remains faithful to the original source. It is not doing much to enhance the graphics beyond a simple upscale, so the visual limitations of the 1990s remain apparent.

The console is priced at $229 and is available in several colors, with transparent plastic that mirrors the aesthetic of the late nineties. While it competes with other FPGA devices like the Analogue 3D, it is significantly less expensive – approximately $85 less when factoring shipping costs.

The M64 does not ship with a controller, but it is compatible with original peripherals or new options. Mod Retro offers a $90 “pro” controller (compensated affiliate link) featuring a metal backing and a rechargeable battery. This controller feels remarkably sturdy with its metal construction, though some early units like mine have minor friction issues near the edges of the analog stick gate. Users can also opt for more affordable wired alternatives from 8bitdo, Retrofighters and Hyperkin. There’s a list of compatible controllers on ModRetro’s website.

The console design includes a few thoughtful touches. There is a dedicated eject button that allows for one-handed cartridge removal. When powered on, an internal light illuminates the installed cartridge, and front-facing LEDs indicate which controller ports are active. A physical dial on the console allows users to navigate an on-screen control panel even if a controller is not immediately reachable.

On the back of the unit, there is a USB-C port for power, an HDMI output, and two currently unused USB-C ports. A micro SD slot with an included 16GB card handles the operating system and updates, though it does not currently support running ROM files directly. For those who still prefer a tube television, an optional $90 adapter is planned that will provide high-quality analog output, including RGB and S-video, which provides a superior experience vs. what the stock hardware can do without modification.

Testing various titles reveals a clean and crisp audio-visual experience. Classic titles like Mario 64 run as expected, though the limitations of the original 240p resolution are very visible on a 4K screen. The system handles 480i interlaced content, such as ReVolt, with reasonable success, though finer details in some games can still appear jagged. Currently, the built-in scanline filters and CRT modes tend to darken the image significantly and obscure detail, suggesting that further firmware refinement is needed in that area. I prefer the clean mode which looks pretty good.

Compatibility is high but not universal. Currently, about 16 games, including heavy hitters like Donkey Kong 64 and Starcraft 64, have known issues. Flash cartridge support is also mixed; while the Summer Kart 64 functions well for standard titles, older Everdrive models don’t seem to work, and virtual 64DD support through the Summercart is not yet functional. Because the system uses an open-source core derived from the MiSTer project, there is an expectation that these compatibility gaps will close as the community and the manufacturer continue to issue updates.

The software interface offers several customization options, including integer scaling to fill the screen and the ability to overclock the system to smooth out games that originally suffered from frame rate drops. Updates can be performed over Wi-Fi, though the system does collect some telemetry data for debugging when connected to WiFi. Some features, such as wireless rumble and the built-in controller save pack, are still awaiting activation in future firmware releases.

While the Analogue 3D may currently offer more flexibility and graphical quality for enthusiasts, the Mod Retro M64 is a less expensive and capable entry point that will likely be easier to find in stock in the long run. Ultimately though, Nintendo 64 games were designed for the specific limitations of analog televisions, which masked many graphical imperfections. While playing in 4K is convenient, the true potential of this hardware may be best realized when the analog adapter arrives, allowing the system to interface with the CRT monitors these games were originally built for.

Disclosure: I paid for the system and controller with my own funds.

Part One of my Media Center Upgrade.. Repurposing Retired 4TB Drives

My home media server is running short on storage space and unfortunately the cost of storage (even for mechanical drives) has increased noticeably. A 4-terabyte WD Red drive that sold for just over $100 a few months ago now retails near $200. To work around these market conditions, I decided to repurpose four older 4-terabyte NAS drives—two WD Reds and two Seagate IronWolfs—that I had previously pulled from other projects.

In my latest Gadget Picks channel video, I test out the new enclosure I’ll be placing them in – a TerraMaster D4-320 (compensated affiliate link). While the D4-320 visually resembles a network-attached storage server, it functions strictly as a direct-attached USB-C 3.2 enclosure. It accommodates up to four SATA mechanical or solid state hard drives. The single USB-C port on the rear supports theoretical transfer speeds up to 10 gigabits per second.

Installation is a tool-less process. The front trays pull out, and plastic side rails secure the drives in place, though screws are provided for those who prefer a more permanent mounting method. The unit supports hot-swapping, allowing drives to be exchanged while the system is running, provided the host operating system supports the action.

During my initial testing on a Mac, I monitored the unit’s power consumption and data transfer rates. With all four mechanical drives installed, the enclosure briefly drew nearly 90 watts as the disks spun up, before settling to an idle draw of roughly 27 watts. When formatted individually, a single drive wrote data at approximately 160ish megabytes per second while consuming 12 watts, and read data at about the same rate.

Because the D4-320 does not feature built-in hardware RAID controller capabilities, it presents the disks as a simple group of independent drives. To test the enclosure’s maximum throughput with mechanical disks, I utilized macOS software to configure the four drives into a striped RAID 0 array.

This configuration is not advisable for critical data due to the risk of total data loss if a single drive fails. But it serves as a practical benchmark as RAID 0 can make use of the aggregate total amount of storage space and get the performance benefit of all four drives as data is striped across each.

Operating as a single 16-terabyte volume in RAID 0, the enclosure achieved read and write speeds between 550 and 560 megabytes per second. Power consumption during these sustained read and write cycles measured approximately 33.7 watts.

With the enclosure functioning properly and the repurposed drives verified, the next phase of this project involves connecting the hardware to my existing Unraid server. This will add 16 terabytes of capacity to the current storage pool and provide a demonstration of how the Unraid handles capacity expansions.

Stay tuned!

Disclosure: Terramaster provided the enclosure free of charge. No other compensation was received and they have not reviewed or approve this content prior to publication.

Lenovo’s $1,849 Panther Lake Mini PC: Yoga Mini i Review

My latest review takes a look at one of the most expensive Mini PCs I have ever tested: The Lenovo Yoga Mini i. It enters the market with a starting price of $749, though the high-end configuration I have been testing retails for $1,869 (compensated affiliate links). This premium version features the new Intel Panther Lake architecture, specifically the Core Ultra X7 358H processor with Intel Arc B390 graphics. While the price point for this specific model is high, the performance of the new processor provides a clear look at where integrated graphics and efficiency are heading.

Check out my video review here!

While the lower end device will be more than adequate for most basic PC tasks, more demanding tasks like video editing, gaming and local AI will perform significantly better on the high end model tested here.

The internal hardware of this higher-end model includes 32 GB of soldered LPDDR5X 8533 RAM. Because the memory is soldered to the motherboard, it cannot be upgraded later. The base model offers 16 GB of non-upgradeable RAM, which is sufficient for standard office tasks or web browsing. Users can upgrade the storage via a relatively accessible NVMe SSD slot at the top of the unit, though the physical process of removing the top lid is difficult and requires a strong suction cup device like this one (compensated affiliate link).

The Yoga Mini serves as a Copilot Plus PC, equipped with a fingerprint reader, built-in speakers, and dual microphones for voice interaction. The system is powered via USB-C and includes a 100 watt USB-C adapter, but any USB power source that can deliver at least 60 watts will do (including docking stations).

The port selection is comprehensive for a small form factor. It includes a headphone jack, two Thunderbolt 4 ports, and two additional USB Type-C ports that all support video output. It can support up to five independent displays utilizing the Thunderbolt and USB-C ports along with the HDMI. Networking is handled by a 2.5 gigabit Ethernet port and a Wi-Fi 7 radio, both of which reached their expected speeds during my testing.

The aesthetics of the device can be managed through the Lenovo “AI Turbo Engine” software, which controls a light ring on the bottom of the unit. This lighting can be customized by color or set to react to system notifications and audio. There is also a capacitive feature on the top of the lid intended to toggle lighting or launch applications with a tap, though its responsiveness was inconsistent in my experience.

General performance is consistent with modern high-end processors. For basic computing, such as web browsing and 4K video playback, the Yoga Mini is responsive. In the Speedometer benchmark, it achieved a score of 37.3, which is comparable to the current Mac Mini M4. The integrated GPU is notably capable in creative applications. In DaVinci Resolve, the system rendered 4K 60 video projects and transitions in real-time and was noticeably faster rendering in more demanding video effects. While it won’t rival a more powerful external GPU for video editing, it’s performance was significantly better than other recent high-end Mini PCs I’ve reviewed.

For those interested in local AI, the unit ran mixture-of-experts models at approximately 24 tokens per second, on par with other machines in its class. Dense models ran at around 3.5 tokens per second.

Gaming performance represents a significant step forward for integrated graphics. In Cyberpunk 2077, running at 1080p with low settings, the system maintained frame rates between 70 and 80 frames per second. Synthetic benchmarks like 3DMark Time Spy suggest that this integrated GPU performs on par with older dedicated desktop cards, such as the GTX 1080, while utilizing a fraction of the power. The system remained stable under sustained load, passing stress tests with a 99.6% rating while keeping temperatures around 73°C and maintaining a quiet fan profile.

Power efficiency is a highlight of the Panther Lake chip. The unit draws between 9 and 12 watts at idle and peaks at approximately 60 watts under full load. For users who prefer alternative operating systems, I found that the latest version of Ubuntu Linux worked without issue. Every hardware component, from the Wi-Fi to the audio, was detected immediately, and the system idled at roughly two watts less than it did under Windows.

The Yoga Mini demonstrates that integrated mobile processors are now capable of handling tasks that previously required dedicated graphics cards or large desktop towers. While component costs and current market constraints keep the price of this high-end configuration at a premium, the efficiency and graphical capability of the hardware indicate a maturing of the mini PC category. For those who require this level of performance in a small footprint, the technology is now available to support it.

If only the price was more reasonable!

Disclosure: The Mini PC was provided on loan by Lenovo. They did not review or approve this content prior to publication nor was any compensation received in exchange. All opinions are my own.

US Bans More Tech.. Robot Vacuums, Solar Inverters & DJI Shadow Companies

The FCC’s “covered list” of banned foreign technology continues to grow. What began with drones and network routers has now widened to include power inverters and robotic devices, categories broad enough to include products people use at home, including portable battery systems and robot vacuums.

Learn more in my latest video.

Like their router declaration, the FCC’s order extends to any new foreign made product irrespective of where it was made but does not pull existing devices off the market. New products, however, face a different path. Unless a company secures a waiver which includes a commitment to onshore its manufacturing, those devices cannot legally enter the U.S. market.

The power inverter order appears aimed in part at grid-connected equipment. The rationale cited in the FCC order refers to a report alleging that a Chinese inverter manufacturer remotely disabled their products in several countries during a distributor dispute.

But the language does not stop at utility-scale or commercial gear. It will also apply to consumer backup power products and solar-capable battery units that many households use during outages. If a device converts DC to AC or AC to DC and includes some form of remote connectivity like a Bluetooth connection to an app, it will fall under the order. That means a product intended for emergency use in a home could end up in the same regulatory framework as larger infrastructure equipment.

The robotics order is similarly expansive. The FCC’s definition includes mobile robots with obstacle avoidance, navigation, or autonomous movement, as long as they exceed a modest weight threshold and include sensors, connectivity, and software that governs movement or data collection. In practice, that could capture a large number of consumer devices, including robot vacuums and robotic floor cleaners, not just industrial machines or security robots.

The waiver process appears to follow the same model used recently for routers. Companies seeking approval for new products must provide detailed supply-chain disclosures, including components, software, and firmware origins. They also need to present a U.S. manufacturing or onshoring plan. Without both, the FCC is unlikely to allow those products to continue entering the market.

Some companies are already getting waivers in other categories. Netgear, Adtran, Eero, and Nokia all have received permission for their routing products. Also noteworthy is that Flock, the controversial surveillance technology company known for its license plate tracking camera systems, received a drone waiver.

In the case of DJI, Congress specifically banned the company’s “surveillance products” along with those from Autel in legislation last year. The law goes beyond just drones and includes other products like their popular Osmo cameras.

DJI attempted to launch two shadow brands to skirt the ban: Skyrover for drones and Xtra for cameras. The FCC has now added those companies to the list as well, cutting off that route. An Osmo Pocket 4 Pro camera sold under the Xtra name was expected this fall, with deposits had reportedly been taken. But refunds were later issued after it became clear the product would not be allowed into the United States.

For consumers, the immediate effect may be limited. Existing products can still be bought. Gray-market imports may still circulate. Individuals may even be able to bring some products back from overseas for personal use. But over time, these restrictions could narrow choices, delay new releases, and push manufacturers to restructure supply chains around U.S. compliance demands which will undoubtedly drive up prices.

The bigger question is: what consumer products will be targeted next?

Stress Test Your Media Server ! SpaceInvaderOne’s Transcoding Benchmark

I recently came across a benchmark tool designed to measure how many simultaneous live transcodes a server can handle. While this is particularly relevant for those running media applications like Plex or Jellyfin, the tool, developed by a member of the Unraid community known as SpaceInvaderOne, is a Docker container that functions on any system supporting that environment.

Check it out in my latest video!

For those using Unraid (compensated affiliate link), the installation is generally a one-click process, though systems with Nvidia GPUs require some specific configuration. I began by installing it on my Intel-based server. After ensuring the default port assignment 8088 was free of conflicts, the container downloaded the necessary components. Before running a benchmark, the tool requires downloading several test clips to local storage.

The interface offers choices between hardware and CPU-based transcoding. Given that my machine used an Intel N150 processor, I focused on hardware transcoding. I initially ran a test involving 4K HDR10 source files being converted to H.265 with tone mapping to SDR. The benchmark adds streams incrementally until the hardware can no longer maintain a viewable playback speed. My first machine, a Beelink ME Mini, managed only a single stream under these specific settings. When I switched the test to a 4K to 1080p H.264 conversion, the result was two simultaneous streams.

These results were lower than expected for an Intel N150. However, the benchmark helped identify a hardware bottleneck. The Beelink machine I was using lacks fast memory, which appears to limit its transcoding capacity. To verify this, I ran the same H.264 test on another N150-based machine, a GMKtec G3 Plus, which has faster RAM. That system achieved three streams. This comparison highlights how secondary components like memory speed can impact the performance of a processor’s integrated graphics.

I also tested higher-end hardware using an Nvidia RTX 3080. On Unraid, this required installing the Nvidia driver and adding a specific “runtime” parameter to the Docker container’s advanced settings. In the H.264 to H.264 test, the Nvidia GPU managed seven concurrent streams. While the performance is higher than the mini PCs, the tool also indicated that this setup consumes significantly more power – especially when it’s not under load.

Beyond live transcoding, the tool includes a benchmark for file conversion speeds. This is intended for users who prefer to pre-encode their media libraries to reduce the workload on their server during playback. There is also a community leaderboard available, which allows for a comparison of hardware performance against data submitted by other users.

These benchmarks function as a stress test. In a standard home environment, performance might be slightly higher than these results suggest because media servers typically buffer ahead to account for temporary performance dips. Having this data provides a baseline for what a server can realistically handle. I plan to integrate this tool into my regular hardware evaluation process, especially as I prepare to transition to a new server in the coming week. Stay tuned!