Simon Willison received a preview unit of the NVIDIA DGX Spark, a desktop "AI supercomputer" retailing around $4,000. He details his experience setting it up and navigating the ecosystem, highlighting both the hardware's impressive specs (ARM64, 128GB RAM, Blackwell GPU) and the initial software challenges.
Key takeaways:
* **Hardware:** The DGX Spark is a compact, powerful machine aimed at AI researchers.
* **Software Hurdles:** Initial setup was complicated by the need for ARM64-compatible software and CUDA configurations, though NVIDIA has significantly improved documentation recently.
* **Tools & Ecosystem:** Claude Code was invaluable for troubleshooting. Ollama, `llama.cpp`, LM Studio, and vLLM are already gaining support for the Spark, indicating a growing ecosystem.
* **Networking:** Tailscale simplifies remote access.
* **Early Verdict:** It's too early to definitively recommend the device, but recent ecosystem improvements are promising.
Key takeaways:
* **Hardware:** The DGX Spark is a compact, powerful machine aimed at AI researchers.
* **Software Hurdles:** Initial setup was complicated by the need for ARM64-compatible software and CUDA configurations, though NVIDIA has significantly improved documentation recently.
* **Tools & Ecosystem:** Claude Code was invaluable for troubleshooting. Ollama, `llama.cpp`, LM Studio, and vLLM are already gaining support for the Spark, indicating a growing ecosystem.
* **Networking:** Tailscale simplifies remote access.
* **Early Verdict:** It's too early to definitively recommend the device, but recent ecosystem improvements are promising.
Nvidia's DGX Spark is a relatively affordable AI workstation that prioritizes capacity over raw speed, enabling it to run models that consumer GPUs cannot. It features 128GB of memory and is based on the Blackwell architecture.
Nvidia introduces the Rubin CPX GPU, designed to accelerate AI inference by decoupling the context and generation phases. It utilizes GDDR7 memory for lower cost and power consumption, aiming to redefine AI infrastructure.
Distiller is a pocket Linux box that runs Claude Code 24/7, offering remote access via QR code. It provides a full Claude Code enabled VS Code environment and terminal session, along with hardware I/O access for developers, firmware engineers, and indie hackers.
A user shares their experience running the GPT-OSS 120b model on Ollama with an i7 6700, 64GB DDR4 RAM, RTX 3090, and a 1TB SSD. They note slow initial token generation but acceptable performance overall, highlighting it's possible on a relatively modest setup. The discussion includes comparisons to other hardware configurations, optimization techniques (llama.cpp), and the model's quality.
>I have a 3090 with 64gb ddr4 3200 RAM and am getting around 50 t/s prompt processing speed and 15 t/s generation speed using the following:
>
>`llama-server -m <path to gpt-oss-120b> --ctx-size 32768 --temp 1.0 --top-p 1.0 --jinja -ub 2048 -b 2048 -ngl 99 -fa 'on' --n-cpu-moe 24`
> This about fills up my VRAM and RAM almost entirely. For more wiggle room for other applications use `--n-cpu-moe 26`.
>I have a 3090 with 64gb ddr4 3200 RAM and am getting around 50 t/s prompt processing speed and 15 t/s generation speed using the following:
>
>`llama-server -m <path to gpt-oss-120b> --ctx-size 32768 --temp 1.0 --top-p 1.0 --jinja -ub 2048 -b 2048 -ngl 99 -fa 'on' --n-cpu-moe 24`
> This about fills up my VRAM and RAM almost entirely. For more wiggle room for other applications use `--n-cpu-moe 26`.
The article discusses the growing trend of running Large Language Models (LLMs) locally on personal machines, exploring the motivations behind this shift – including privacy concerns, cost savings, and a desire for technological sovereignty – as well as the hardware and software advancements making it increasingly feasible.
Bee is an AI wearable that aims to solve memory retention issues by summarizing conversations and generating daily diary entries. However, the author's experience with the device was marred by inaccuracies and privacy concerns.
The NVIDIA Jetson Orin Nano Super is highlighted as a compact, powerful computing solution for edge AI applications. It enables sophisticated AI capabilities at the edge, supporting large-scale inference tasks with the help of high-capacity storage solutions like the Solidigm 122.88TB SSD. This review explores its use in various applications including wildlife conservation, surveillance, and AI model distribution, emphasizing its potential in real-world deployments.
A USB stick equipped with a Raspberry Pi Zero W runs a large language model using llama.cpp. The project involves porting the model to an ARMv6 architecture and setting up the device as a composite that presents a filesystem to the host, allowing users to interact with the LLM by creating text files that are automatically filled with generated content.
The Hat uPCIty Lite is a PCI Express evaluation board with an open-ended PCIe X4 slot, designed for the Raspberry Pi 5. It supports external power, isolates PCIe Express power delivery to protect the Pi, and is compatible with PCIe x1 interface in Gen2 and Gen3 standards. The board includes all necessary accessories and is built with high-quality components.
