Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11You can run local LLMs on a DGX Spark with NVIDIA’s guided vLLM serving recipe or by building llama.cpp with CUDA and serving a GGUF model. Choose based on the model format and workload, then check the exact model-and-software configuration: the Spark’s 128 GB of unified memory and NVIDIA’s “up to 200 billion parameters” capability do not guarantee that every model, context length, or inference stack will fit or perform well.
Choose an inference route
Both routes can expose an OpenAI-compatible API, but they use different model and setup workflows. NVIDIA positions vLLM for serving workloads that benefit from throughput and continuous batching. Its llama.cpp walkthrough builds a CUDA-enabled server for GGUF checkpoints. Neither source establishes a universal performance winner.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Dell NVIDIA Tesla V100 GPU SXM2 32GB NWWWX by DELL | $854.96 | Buy on Amazon |
| 2 |
|
Gigabyte NVIDIA GeForce RTX 3060 Gaming OC V2 Graphics Card - 12GB GDDR6, 192-bit, PCI-E 4.0,... | $695.00 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
| Route | Best fit | What you set up | Memory and storage planning |
|---|---|---|---|
| vLLM | Serving workloads where throughput, continuous batching, and NVIDIA’s guided configuration matter. | Use a recipe matched to the Spark count, model variant, and precision. The recipe supplies the model, container, environment, and serving configuration. | Check the selected recipe’s hardware and memory configuration; compatibility is not determined by parameter count alone. |
| llama.cpp | GGUF checkpoints and a build-from-source workflow without a separate Python inference framework. | Build llama.cpp with CUDA, download a GGUF checkpoint, and launch llama-server. |
For NVIDIA’s illustrated example, plan for about 30 GB of free RAM and about 40 GB of free disk; these are example-specific estimates, not universal minimums. |
How large an LLM fits on a DGX Spark?
NVIDIA specifies 128 GB of LPDDR5x unified memory and describes the platform as capable of supporting models up to 200 billion parameters on one system. These are vendor specifications and a capability statement, not a promise that every model at that size will run with every quantization, context length, or serving stack. Memory must also accommodate the operating system, model runtime, and the KV cache used to handle context.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsThe Spark’s other NVIDIA-published specifications include a 20-core Arm processor, Blackwell GPU, 273 GB/s memory bandwidth, and 1 TB or 4 TB NVMe M.2 storage. NVIDIA lists the included 240 W power supply for optimal performance. These specifications describe the hardware; they are not independent performance measurements. See the DGX Spark hardware overview.
#1 Best Overall
- GPU Chipset: NVIDIA
- Memory: HBM2
- Programming Interface: CUDA
- Memory Capacity: 32GB
- Slot Compatibility: SXM2
For a single Spark, NVIDIA’s current vLLM recipe selector recommends Qwen3.8-27B NVFP4 and describes the quantized model as fitting one Spark with its hardware-specific configuration. That is a concrete starting point, not evidence that any similarly sized model or different precision will work with the same settings. For a larger or different model, assess the recipe and available memory rather than extrapolating from the 200-billion-parameter headline.
Set up the Spark before installing a model
First boot requires a reliable internet connection because the setup wizard downloads and installs the full software image. You can complete setup at the machine with a display, keyboard, and mouse, or over the local network from another computer. The initial choice does not lock you into that access method: afterward you can use it locally, over NVIDIA Sync, SSH, remote desktop, or a combination.
- Attach peripherals before connecting power. The system starts as soon as power is connected. Connect wired Ethernet before installation if you plan to use it.
- Prepare a stable internet connection. NVIDIA does not recommend using a captive portal or an unstable phone hotspot for setup updates.
- Connect power and complete the on-screen setup wizard. The wizard handles account creation, network settings, and downloading and installing the software image. Do not shut down or reboot while updates are installing.
- If there is no display, try another video connection. NVIDIA suggests trying HDMI if a display connected over USB-C/DisplayPort does not appear.
Follow the official first-boot guide for the setup process and consult the system overview for available access methods.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Run a model with NVIDIA’s vLLM recipe
The recipe selector asks you to choose one Spark, one Station, or two Sparks before selecting a model. For one Spark, its recommended Qwen3.8-27B NVFP4 recipe is configured for that single-device setup. NVIDIA’s playbook emphasizes that “Model size is not the only compatibility requirement”: container architecture, vLLM version, quantization, parsers, and parallel configuration must match the chosen model and hardware.
- Open NVIDIA’s vLLM recipe selector. Select the hardware configuration that matches your system, such as one Spark.
- Choose the exact model variant and precision you intend to serve. Do not assume that settings for one quantization or model variant carry over to another.
- Enable only the capabilities you need. Options such as tool calling or reasoning may require corresponding configuration.
- Use the complete launch configuration from that generated recipe. Take the model ID, container, environment, and full serve command from the same recipe’s launch instructions; do not combine values from different recipes.
The recipe’s launch tabs provide the configuration and instructions for its selected model. If you choose a different model, use its matching recipe: NVIDIA warns that download steps, container, environment, memory settings, parser, or parallelism may differ. This guide does not reproduce a command because the correct one depends on the selected recipe.
Rank #2
- NVIDIA Ampere Streaming Multiprocessors: Building blocks for the world's fastest, most efficient GPUs, the all-new Ampere SM brings twice the FP32 throughput and improved energy efficiency
- 2nd Generation RT Cores - Experience 2x the 1st Generation RT Cores throughput, plus competitive RT and shading for a whole new level of ray-tracing performance
- 【3rd Generation Tensor Cores】Get up to 2X the throughput with structural sparsity and advanced AI algorithms such as DLSS
- Core Clock: 1837MHz
- WINDFORCE 3X Cooler
Run a GGUF model with llama.cpp
NVIDIA’s llama.cpp path builds the project with CUDA so it can use the Spark’s GB10 GPU, downloads a GGUF checkpoint, and starts llama-server. The resulting server provides an OpenAI-compatible /v1/chat/completions endpoint.
NVIDIA illustrates the workflow with Qwen3.6-35B-A3B with MTP support. Treat that as a walkthrough example, not a general recommendation for every user or workload. The playbook lists DGX OS, Git, CMake 3.14 or later, CUDA Toolkit, and network access to GitHub and Hugging Face as prerequisites. It estimates roughly 30 minutes to build and run, in addition to downloading the model; the estimate is not a guarantee for every network or machine.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Disk: NVIDIA gives about 35 GB as an order-of-magnitude download size for the example’s default quantized GGUF and estimates about 40 GB of free disk for the example download and build artifacts.
- Memory: NVIDIA estimates about 30 GB of free RAM for the example model and KV cache. The checkpoint still must fit alongside the KV cache in the machine’s available unified memory.
These figures apply to NVIDIA’s illustrated setup, not every GGUF model or context length. Follow the llama.cpp walkthrough for DGX Spark for its CUDA build and launch instructions.
Check software versions on your system
NVIDIA’s release notes list DGX OS 7.5.0, GPU driver 580.159.03, CUDA Toolkit 13.0.2, and kernel 6.17 for the Founders Edition. The release notes explicitly scope those versions to that edition; GB10-based partner systems may receive updates on a different schedule. Check the installed versions on your actual machine and follow the update instructions for your system before applying commands or environment assumptions from a recipe. Consult the DGX Spark release notes.
Quick Recap
Troubleshoot model-fit and launch problems
- The model does not fit or runs out of memory: Recheck the selected quantization, context length, and KV-cache requirements. Leave room for the operating system and runtime; parameter count alone is not a memory budget.
- A vLLM recipe fails to launch: Verify that the model variant, precision, container, vLLM version, parser, environment, and parallel configuration all come from the same matching recipe.
- A llama.cpp build or model download fails: Confirm that DGX OS, Git, CMake 3.14 or later, and CUDA Toolkit are available, and that the machine can reach GitHub and Hugging Face. Check free disk space against the example’s download and build needs.
- Instructions do not match installed software: Confirm your system edition and current versions. Founders Edition release-note values should not automatically be assumed for a GB10 partner system.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
