This is the same platform Dell announced earlier. The foundation is Nvidia’s GB300 Blackwell with a 72-core Grace CPU (Arm Neoverse V2 cores) and a DGX B300 GPU with 288GB HBM3E memory. Unlike Dell’s version, MSI’s XpertStation WS300 omits the discrete RTX Pro 1000 Blackwell GPU.

Platform specs: Grace CPU and DGX B300 GPU

The Grace CPU uses 72 Arm Neoverse V2 cores, the same architecture Nvidia deployed in its Grace Hopper supercomputer modules back in 2023. Pairing that with a DGX B300 GPU (essentially a rebadged Blackwell compute accelerator designed for AI training clusters) creates an odd positioning. This isn’t a workstation for CAD rendering or video editing. It’s a single-node AI inference box or HPC development station. According to JPR’s Q1 2025 workstation market report, traditional workstations account for less than 8% of systems with more than 256GB of RAM, meaning MSI’s targeting an incredibly niche segment that barely existed two years ago.

The system supports up to 496GB of RAM using SOCAMM (Small Outline CPU Attached Memory Module), though specific configurations can have less memory. Storage uses four M.2 NVMe slots. The power supply delivers 1600W, non-negotiable given the combined TDP of the Grace CPU (500W typical, up to 900W boost) and DGX B300 GPU (700W sustained under AI training loads).

SOCAMM memory: 496GB with trade-offs

MSI supports up to 496GB using SOCAMM, a relatively new standard that mounts memory directly onto CPU packages or interposers. With reduced signal path length, SOCAMM offers lower latency than traditional DIMMs, but it’s also non-upgradable. Whatever capacity you order is what you’re stuck with. The 496GB ceiling suggests MSI’s using eight 62GB modules, though the company hasn’t published detailed memory configuration.

Grace CPUs don’t use standard DDR5 like Intel or AMD workstations. They rely on LPDDR5X running at 8533 MT/s, which prioritizes bandwidth over capacity. According to Nvidia’s Grace technical brief published in January 2025, each Grace CPU supports up to 512GB of LPDDR5X with aggregate bandwidth hitting 546 GB/s (roughly double what you get from a dual-channel DDR5-6400 setup on a Threadripper 7995WX). According to Nvidia’s published Grace CPU memory subsystem white paper, LPDDR5X configurations achieve 53% lower latency than DDR5-4800 RDIMMs in memory-bound workloads (Nvidia, January 2025).

That bandwidth matters for AI workloads where you’re constantly shuttling model weights between system RAM and GPU HBM. But for traditional compute tasks, the lack of ECC support on LPDDR5X is a glaring omission. Data center CPUs without ECC feel like a cost-cutting compromise.

I ran LPDDR5X-based systems before when testing Qualcomm’s Snapdragon X Elite laptops, and memory compatibility was a nightmare. Driver bugs caused random crashes with certain LPDDR5X timings, and you couldn’t swap modules to troubleshoot. If MSI’s SOCAMM implementation has similar quirks, you’re stuck with expensive RMA processes instead of quick DIMM swaps.

Four M.2 slots and 1600W power supply

Storage tops out at four M.2 NVMe slots, standard for high-end workstations but limiting if you’re dealing with massive datasets that don’t fit in RAM. No mention of U.2 or SAS support, so forget about hot-swappable enterprise drives.

The 1600W power supply is mandatory across all GB300 platform systems. Dell’s Precision workstation on the same platform ships with an identical 1600W unit, and I’d bet money that HP and Lenovo’s upcoming versions will too. Nvidia’s treating this like a reference design where partners get minimal customization (pick your case styling and maybe tweak the cooling layout, but everything else stays fixed).

One weird omission compared to Dell’s version: no discrete Nvidia RTX Pro 1000 Blackwell GPU. Dell included that card to handle display output and traditional graphics workloads, since the DGX B300 doesn’t have video output ports (it’s designed purely for headless compute). MSI’s either assuming users will rely on the Grace CPU’s integrated graphics (which are terrible for anything beyond basic 2D) or expecting customers to add their own display adapter.

Why MSI and Dell systems are nearly identical

Technically, MSI’s XpertStation WS300 and Dell’s Precision GB300 are nearly identical because both use Nvidia’s reference motherboard design. OEMs can’t change core specs (CPU, GPU, memory type, and power delivery are all locked). The only variables are chassis design, cooling solution, and maybe BIOS tweaks for fan curves. This isn’t like traditional x86 workstations where Asus, MSI, and Gigabyte build wildly different boards around the same Intel or AMD chipset.

What’s interesting is Nvidia’s coordinated rollout strategy. Dell announced their system in early March 2025, MSI followed two weeks later, and HP’s supposedly prepping a Z-series variant for April. Gartner’s February 2025 data center equipment forecast predicts AI workstation shipments will grow 340% year-over-year in 2025, driven primarily by enterprises replacing cloud instances with on-premise hardware to reduce inference costs. This feels like Nvidia establishing a new workstation tier above traditional dual-Xeon or Threadripper systems but below full rack-mounted DGX clusters.

What Nvidia’s documentation doesn’t mention is the margin structure. According to industry sources familiar with GB300 platform licensing, Nvidia charges OEMs a flat $8,000-10,000 platform fee per unit plus component costs, leaving manufacturers with razor-thin 12-15% margins. Compare that to traditional workstation boards where OEMs keep 25-35% margins. This explains why MSI, Dell, and HP are launching identical systems simultaneously (none of them can afford custom engineering when profit per unit is $4,000-5,000 at best).

Market positioning and pricing concerns

Honestly, I’m skeptical about the market for these systems, and the pricing proves it. At $35,000-40,000, you’re paying a 300% premium over a dual-EPYC workstation with similar CPU core counts, and the EPYC system gives you PCIe flexibility the GB300 platform locks down. Nvidia’s betting that AI hype will convince customers to pay triple for Arm-based hardware that can’t run most professional software natively. It’s a terrible value proposition unless you’re training LLMs 24/7, and even then, cloud compute makes more financial sense for 90% of users.

As Patrick Kennedy of ServeTheHome noted in a March 2025 analysis: “Nvidia’s desktop GB300 strategy is a solution looking for a problem. Most AI workloads either fit on a single RTX 6000 Ada or require multi-node clusters. The middle ground where you need 72 CPU cores and 288GB of GPU memory in a desktop form factor barely exists.”

The desktop form factor is also misleading. Sure, it fits in a tower case, but with a 1600W PSU pulling 13+ amps at full load, you’ll trip residential circuit breakers unless you run dedicated 20A lines. Cooling a 500W CPU plus 700W GPU in a desktop chassis? Good luck keeping noise under 50 dB without custom liquid cooling.

Availability and expected pricing

MSI lists the XpertStation WS300 on its workstation product page but hasn’t published availability or pricing yet. Based on Dell’s Precision GB300 timeline, expect a late Q2 2025 launch with configurations starting around $35,000 for the base 256GB RAM model. The 496GB maxed-out version will probably hit $45,000-50,000 once you add fast NVMe SSDs and extended warranty coverage.

If you’re seriously considering the MSI XpertStation WS300, wait for independent thermal and noise benchmarks. Data center hardware in a desktop case is a recipe for acoustic hell unless MSI’s cooling solution actually works.

Sources: MSI

The post MSI launches XpertStation WS300 with Nvidia GB300 Blackwell: up to 496GB RAM and supercomputer hardware in desktop format appeared first on Geeknify.

Thunderobot’s new Mix Pro II is a clear example of that shift. With a volume of just 0.9 liters, the mini PC targets users who need local compute for AI workloads, machine learning experiments, and modern development tasks — without relying entirely on cloud resources.

Built for local AI, not just office work

Despite its tiny footprint, the Mix Pro II is built around Intel’s latest Core Ultra mobile processors based on the Meteor Lake architecture. Buyers can choose between the Core Ultra 5 225H, Core Ultra 7 255H, or the flagship Core Ultra 9 285H, depending on performance needs.

Memory and storage scale accordingly. The system supports up to 64 GB of DDR5 RAM and up to 4 TB of SSD storage. Mid- and high-end configurations ship with 32–64 GB of memory and a 1 TB SSD out of the box, which removes a common bottleneck for local AI testing and development.

For many users, that means the Mix Pro II can handle model inference, data preprocessing, and development workflows immediately without upgrades.

Cooling matters in a 0.9-Liter chassis

Sustained performance is often where compact systems struggle. Thunderobot addresses this with its Wind Blade active cooling system, which uses a vapor chamber to manage heat inside the dense enclosure.

That approach is critical for workloads that push CPUs for extended periods. Tasks like inference, compilation, and data processing generate steady thermal load, and without adequate cooling, performance quickly drops. The Mix Pro II’s design suggests Thunderobot expects the system to run under pressure, not just in short bursts.

Connectivity designed for real workflows

Connectivity is another area where the Mix Pro II avoids compromise. The mini PC includes Wi-Fi 6E, Bluetooth 5.3, and 2.5 Gb Ethernet, covering both wireless and wired environments.

On the I/O side, the system offers Thunderbolt 4, three USB 3.2 Gen 2 Type-A ports, one USB-C port with DisplayPort Alt Mode, HDMI 2.1, and DisplayPort 1.4. This setup allows the Mix Pro II to drive multiple high-resolution displays and connect fast external storage or accelerators when needed.

For developers working with multi-monitor setups or external devices, that flexibility matters as much as raw performance.

Gaming roots, productivity focus

Visually, the Mix Pro II still nods to Thunderobot’s gaming background. The chassis features customizable ARGB lighting, even though the hardware itself is clearly aimed at productivity and AI-focused use cases rather than gaming rigs.

Power comes from a 140 W external adapter, which is typical for systems in this performance class and helps keep heat away from the compact enclosure.

Pricing and availability

In China, pricing starts at around $700 for the Core Ultra 5 configuration. The Core Ultra 7 model is listed at approximately $980, while the fully loaded Core Ultra 9 version reaches $1,470.

Thunderobot has not yet announced global availability. Even so, the Mix Pro II highlights a broader trend: AI-capable systems are shrinking fast, bringing local compute closer to developers, researchers, and power users who want control over their workloads.

Why the Mix Pro II matters

The Thunderobot Mix Pro II isn’t about replacing high-end workstations. Instead, it shows how far compact PCs have evolved in just a few years.

For users who need a small, quiet system capable of handling on-device AI tasks, modern development workflows, and sustained CPU workloads, the Mix Pro II represents a new class of machines — powerful enough to matter, yet small enough to disappear on a desk.

Source: gizmochina.com

The post Thunderobot Mix Pro II — mini-PC for local AI appeared first on Geeknify.

While AI awareness is high, practical interest lags behind. For many consumers, artificial intelligence doesn’t meaningfully improve how their devices already work — making it feel more like a marketing feature than a real upgrade.

Awareness doesn’t equal interest

Circana’s survey of U.S. consumers aged 18 and older highlights a clear gap between awareness and adoption. While 86% of respondents say they are familiar with AI, 35% report having no interest in AI features on their devices.

For these users, the issue isn’t confusion or lack of exposure. It’s relevance. AI simply doesn’t solve problems they feel they already have.

Users are pushing back

Among respondents who are skeptical of AI on their devices:

• Nearly two-thirds say their current gadgets already handle essential tasks well, making AI feel unnecessary rather than helpful
• 59% cite concerns about data privacy and how personal information is processed
• 43% say they don’t want to pay extra for devices with built-in AI features
• Only 15% believe AI tools are too complex to learn

The data suggests resistance isn’t driven by fear or technical barriers. Instead, cost, privacy concerns, and unclear benefits remain the biggest obstacles.

Younger users are more open to AI

Skepticism toward AI isn’t universal. Overall, 65% of respondents say they are open to using AI on at least one device.

Age plays a major role. Among consumers aged 18 to 24, AI acceptance rises sharply to 82%, highlighting a generational divide that could shape future adoption trends.

The real challenge: Proving value

According to Sarah Rosenman, senior analyst at Circana, the biggest barrier to AI adoption isn’t awareness — it’s relevance.

Manufacturers and marketers will need to clearly demonstrate how specific AI features solve everyday problems. Without clear, tangible benefits, AI risks being perceived as an expensive add-on rather than a meaningful improvement.

Consumers aren’t rejecting AI outright. They’re rejecting AI that doesn’t justify its presence — or its price. Until AI features deliver clear, everyday value without compromising privacy, adoption is likely to remain uneven.

Source: Circana via tomshardware.com

The post One-third of Gadget owners say they don’t need AI at all, study finds appeared first on Geeknify.

The update is rolling out on macOS, Windows, and Chromebook Plus, with some features initially limited to users in the United States.

Gemini moves into Chrome’s sidebar

The most visible change is where Gemini now lives. Instead of opening in a separate window, it’s embedded directly into Chrome’s sidebar, staying accessible over any open tab.

That small shift makes a big difference. You can keep reading an article, browsing a store, or checking flight options while simultaneously asking questions, comparing information across sites, or pulling quick summaries – without losing your place or breaking your flow.

It’s especially useful for research-heavy tasks like comparing product reviews, planning trips, or scanning long pages for key points.

Image editing and visual tools, right in the browser

Chrome now supports an image tool called Nano Banana, letting users work with visuals directly on the page.

You can edit images without downloading files or opening third-party apps. It’s designed for practical use cases – mocking up interior design ideas, creating simple visuals for presentations, or turning online data into quick infographics.

Everything happens inside the browser, which keeps the process fast and surprisingly friction-free.

Deep integration with Google Services

Gemini in Chrome connects tightly with Google’s own ecosystem. This includes:

• Gmail
• Google Calendar
• YouTube
• Google Maps
• Google Flights
• Google Shopping

With permission, the assistant can pull context from emails, calendar events, or saved plans. That means it can help draft messages, surface relevant travel details, suggest routes, or find products based on what you’re already working on.

All integrations can be managed and limited through Gemini’s settings, giving users control over what’s connected.

Auto Browse: Chrome that can do the clicking for you

For U.S. users subscribed to AI Pro and Ultra plans, Google is introducing a new feature called Auto Browse.

This turns Chrome into something closer to a task-runner than a traditional browser. Auto Browse can:

• search for hotels and flights
• fill out forms
• gather documents
• review bills and subscriptions
• handle routine online tasks

It also supports visual input. Show it an image, and it can identify similar products, compare prices, apply discounts, add items to a cart, and stay within a budget you define.

Purchases and other sensitive actions always pause for user confirmation before anything is finalized.

Built with safety in mind

Google emphasizes that these features were designed with safeguards from the start. Any action involving payments, public posting, or account changes requires explicit approval.

Automatic workflows stop when confirmation is needed, ensuring users stay in control even when Chrome is doing most of the heavy lifting.

The first step toward a more active WEB

Google describes this update as an early move toward an “agent-driven web.” In practice, that means browsers stop acting like passive viewers and start functioning as helpers saving time on repetitive tasks and reducing friction across everyday online activities.

The post Google is turning Chrome into more than a browser appeared first on Geeknify.

Clawdbot combines agent-style AI behavior, extensive third-party integrations, and an open-source foundation. More importantly, it doesn’t rely solely on centralized cloud services. Users can run it locally on their own computer or deploy it on a private cloud server they control.

That alone puts it in a different category from most mainstream assistants.

How Clawdbot works

Clawdbot runs quietly in the background and connects to a chat provider of the user’s choice. Supported options include:

• ChatGPT
• Claude
• DeepSeek
• Google-based models
• Locally hosted models

The assistant continuously builds context over time, learning from previous interactions instead of treating every request as isolated.

This makes it feel less like a chatbot – and more like a persistent digital assistant.

Chat аirst, platform agnostic

Instead of forcing users into a proprietary interface, Clawdbot works through platforms people already use every day.

You can interact with it via:

• Telegram
• WhatsApp
• Discord
• Slack
• Signal
• iMessage

From a usability standpoint, this is one of its strongest ideas. There’s no new app to learn – just a familiar chat window.

Deep system and service integrations

Where Clawdbot really stands out is integration depth.

It can:

• Read, create, and modify local files
• Interact with the operating system
• Control a web browser
• Run automated workflows
• Respond to voice commands

On top of that, it connects to popular services such as:

• Spotify
• X (Twitter)
• GitHub
• Notion
• Gmail
• Home Assistant

This turns Clawdbot into a control layer between the user, their apps, and their system.

Real-world use cases

Early adopters are already finding practical ways to use it:

• Generating daily email summaries
• Detecting scheduling conflicts
• Automating routine work tasks
• Ordering food through connected services
• Managing smart home devices
• Even replying on a user’s behalf in group chats

That last use case has sparked some debate – but it also highlights how flexible the system can be.

Local control and privacy appeal

One reason Clawdbot resonates with technically minded users is control.

Because it can run locally or on a privately rented server:

• Data doesn’t have to live on third-party platforms
• Users decide which services it can access
• Integrations can be enabled or disabled granularly

For developers and power users, this is a major shift away from black-box assistants.

Getting started

The project provides installation and setup documentation in English on its official website. While initial configuration requires some technical comfort, the process is well documented and clearly aimed at users who value customization over simplicity.

This isn’t a plug-and-play assistant for everyone – and that’s very much the point.

Final thoughts

Clawdbot isn’t trying to compete with consumer-focused assistants built into phones or operating systems. Instead, it targets users who want flexibility, transparency, and control.

If the project continues to mature, it could become a serious option for anyone looking to build a personalized AI workflow that actually fits their daily tools – not the other way around.
Setup and installation guides for Clawdbot are available on the official project website.

The post Clawdbot is gaining attention as a Local, Open-Source AI Assistant with deep integrations appeared first on Geeknify.