Stock Availability Updates
| Model | Estimated Shipping | Status |
|---|---|---|
| NVIDIA DGX Spark | Shipping now | |
| Dell Pro Max GB10 | Shipping now | |
| ASUS Ascent GX10 | Shipping now |
The New Wave of AI Supercomputers Powered by Blackwell
- Now on Your Desk
Imagine having the power of a supercomputer sitting next to your monitor.
That’s exactly what the new wave of AI workstations powered by Blackwell delivers.
These systems are small but mighty - no bigger than a lunchbox, yet capable of running the kind of models that once required a data centre.
With up to 1 petaflop of compute and enough memory to handle 200 billion-parameter models, they turn long waits and cloud bottlenecks into instant results.
→ For researchers, it means no more queues.
→ For startups, no runaway cloud bills.
→ For enterprises, a secure, on-prem path to innovation.
NVIDIA DGX Spark
And because they share the same NVIDIA software stack used in full-scale AI factories, you can start at your desk and scale seamlessly when you’re ready.
Top 10 Reasons These Small But Mighty Systems are Changing the Game
Until now, building serious AI meant waiting for shared clusters or watching cloud costs spiral.
Blackwell-powered workstations change that - putting data centre-class performance right at your desk.
Here’s why it matters:
Freedom from the Cloud
Run AI locally, with predictable costs and no usage-based fees. No queues, no surprise bills.
Built for AI
Up to 1 petaflop of compute and 128 GB of unified memory let you train, fine-tune, and inference models with up to 200 billion parameters.
Compact and Flexible
About the size of a lunchbox, these systems fit where you work - at a desk, in a lab, or at the edge. And with ConnectX networking, two units can be linked to handle even larger models.
Scalable Investment
Start with one system, expand to two for double the memory and 400B+ parameter models. It grows with your needs.
Secure and Compliant
Keep sensitive data on-premise to meet privacy, regulatory, and legal requirements.
For sectors like healthcare, finance, government, and research, this means easier compliance - without sending data outside your walls.
Ready on Day One
Preloaded with NVIDIA DGX OS and the full AI software stack, so you can get started immediately with familiar tools like PyTorch, TensorFlow, Jupyter, and more.
Developer Ecosystem Access
Comes with NVIDIA Developer Program benefits - SDKs, blueprints, pre-trained models, and community support - helping you get results faster.
Your Entry Point to the World’s Most Powerful AI Architecture
DGX Spark and its OEM counterparts are built on the same NVIDIA platform that powers enterprise data centres and AI factories worldwide. Start small, then scale to racks or the cloud without rewriting code - your AI journey stays seamless at every step.
No Specialist Facilities Needed
Plug into a standard wall socket - no need for special cooling, racks, or a dedicated data centre space. Quiet and compact, these systems are designed to sit comfortably in an office, lab, or classroom without disrupting the workspace.
Energy Efficient by Design
Huge performance in a tiny footprint, with lower power draw compared to traditional server racks or heavy cloud usage.
You’ve got Options
The NVIDIA DGX Spark is the first of a new class of desktop AI supercomputers - but it isn’t the only one.
Dell and ASUS have already announced their own versions, each built on the same powerful Blackwell architecture but tailored for different needs. These systems are expected to ship just a few weeks after NVIDIA’s Founders Edition.
Dell Pro Max with GB10
Designed with enterprises in mind. Dell’s version puts a strong emphasis on security, compliance, and seamless scaling into the wider Dell AI Factory ecosystem. It’s the safe choice for organisations that need peace of mind around regulations and long-term support.
ASUS Ascent GX10
Built with developers and labs in mind. ASUS has engineered its box for quiet, efficient cooling and durability, making it ideal for continuous workloads and environments where sustained performance matters most.
And that’s just the start. Other manufacturers will follow with their own takes - giving you more choice in design, features, and support.
The price points are likely to vary between manufacturers. That means you’ll be able to pick the version that fits your budget as well as your technical needs, without compromising on performance.
No matter which brand you choose, you’re getting the same breakthrough:
Blackwell-powered AI performance at your desk, ready to grow with you.
What’s Inside
These systems aren’t just smaller versions of traditional servers - they’re built on a new kind of architecture designed from the ground up for AI.
At the core is the NVIDIA GB10 Grace Blackwell Superchip, which combines:
- A powerful Blackwell GPU with fifth-generation Tensor Cores (for training, fine-tuning, and inference at scale).
- A 20-core Grace CPU, tightly linked to the GPU with NVIDIA’s NVLink-C2C interconnect - delivering up to 5× the bandwidth of PCIe Gen 5.
- 128 GB of unified memory shared across CPU and GPU, so even massive models (up to 200B parameters) can fit without bottlenecks.
That means you can prototype, fine-tune, and deploy with the exact same tools and frameworks - from PyTorch and TensorFlow to NVIDIA NIM microservices - and scale seamlessly from your desk to the data centre or cloud.
For the full technical breakdown, you can download the official NVIDIA DGX Spark datasheet (PDF).
The Secret Sauce: NVIDIA AI Software
The hardware is only half the story.
What makes these systems truly powerful is that they arrive ready to work - preloaded with the same AI software stack used in the world’s largest AI factories.
From the moment you switch it on, you have direct access to:
- CUDA and NVIDIA developer SDKs for building and optimising models.
- PyTorch, TensorFlow, and Jupyter for everyday research and development.
- NVIDIA AI Enterprise for production-grade reliability.
- NVIDIA NIM microservices to deploy generative AI and LLMs at scale.
Instead of spending weeks configuring drivers and dependencies, you can focus on what matters: training, fine-tuning, and deploying models - with the confidence that your workflows can scale seamlessly from your desk to the cloud or a data centre.
What You Can Do With It
Prototype and Fine-Tune Models
Train and adapt large language models locally. What once meant waiting days in a cluster queue can now be done overnight on your own system. Handle models with up to 200B parameters on a single unit.
Run Inference at Scale
Deploy chatbots, AI assistants, or vision models with real-time responsiveness - no waiting for cloud GPUs to spin up. With up to 1 petaflop FP4 AI performance, workloads stay fast and predictable.
Build with Generative AI
Explore text-to-image, video generation, and design automation. Iteration cycles shrink from “maybe tomorrow” to “try it again right now.”
Advance Data Science Pipelines
Process, train, and analyse large datasets without shuffling between shared servers. More experiments, fewer bottlenecks.
Develop Agentic AI
Build AI systems that can reason, plan, and act across multiple tasks - from enterprise productivity to robotics - and test them locally without downtime.
Push AI to the Edge
Compact enough for labs, factories, and clinics. Deploy where data is created, reducing latency and keeping information secure.
Performance in Practice
Independent third-party benchmarks are still emerging - but vendor specifications give us a clear picture of what these systems can do:
- Model size capacity: up to 200B parameters on one system, 400B+ when two are linked with ConnectX networking.
- Unified memory: 128 GB shared between CPU and GPU, avoiding bottlenecks in large-scale workloads.
- Energy efficiency: runs off a wall socket, drawing less power than a typical server node.
We’ll update this section as independent testing data becomes available.
Specifications
Curious about the numbers? Here’s how the core specs line up across NVIDIA’s Founders Edition and early OEM versions. All are powered by the same Grace Blackwell architecture - differences are mostly in design, features, and support
| NVIDIA DGX Spark | Dell Pro Max GB10 | ASUS Ascent GX10 | |
| Architecture | NVIDIA Grace Blackwell | ||
| Compute | Up to 1 PFLOP FP4 | Same | Same |
| CPU | 20 core Arm, | Same | Same |
| GPU | NVIDIA Blackwell Architecture | Same | Same |
| OS | NVIDIA DGX™ OS | ||
| System Memory | 128 GB LPDDR5x, | Same | Same |
| Storage | 1 or 4 TB NVME.M2 | ||
| Model Size | Up to 200B parameters (single unit) | Same | Same |
| Networking | 1× ConnectX-7 (200GbE/InfiniBand) | Same | Same |
| Scalability | Link 2 systems → 400B+ params | Same | Same |
| System Dimensions | 150 mm L x 150 mm W x 50.5 mm H | ||
| System Weight | 1.2 kg | ||
| Design | Compact reference design | Enterprise cube, Dell AI Factory ecosystem | QuietFlow cooling, ruggedised chassis |
| Focus | First to ship | Compliance, ProSupport, regulated industries | Cooling, dev-lab durability, multi-display support |
| Estimated Ship Date | Shipping Now | November, TBC | November, TBC |
| More Info | Download NVIDIA DGX Spark Datasheet | Dell Pro Max with GB10 Blog Post |
No matter which option you choose, you’re getting the same Grace Blackwell performance and NVIDIA AI software stack.
The differences come down to design, features, and ecosystem - and we’ll help you pick the right fit.
| NVIDIA DGX Spark | Dell Pro Max GB10 | ASUS Ascent GX10 | |
|---|---|---|---|
| Architecture | NVIDIA Grace Blackwell | Same | Same |
| Compute | Up to 1 PFLOP FP4 | Same | Same |
| Sample #3 | Row 3, Content 1 | Row 3, Content 2 | Row 3, Content 2 |
Why Work With AlloComp
Buying one of these systems is just the first step. The real value comes from having the right partner by your side. That’s where we come in:
- Vendor-agnostic advice - NVIDIA Founders Edition, Dell, ASUS, Gigabyte, PNY and more - we help you choose the system that actually fits your needs.
- Early access - we work closely with our partners, so we can often secure allocations before stock goes public.
- Support beyond the box - setup, integration, optimisation, and scaling when you’re ready.
- A local partner, with global reach - based in Ireland, connected across EMEA, plugged into the global AI ecosystem.
At AlloComp, we don’t just ship hardware. We help you find the right home for your AI - whether that’s at your desk, in your data centre, or in the cloud.
AI supercomputing is no longer locked away in data centres or behind cloud queues. With Blackwell-powered systems now at your desk, you can move faster, work smarter, and stay in control.
Mighty future.
