Can NVIDIA DGX Spark (GB10) · 64 GB run Llama 3.1 8B Instruct?

64 GB unified · 273 GB/s memory bandwidth · Llama 3.1 8B Instruct has 8B parameters.

Short answer

Yes, excellent for chat; borderline for coding in a repository.

Best quantization for chat: Q6_K (6.3 GB of weights) on Ollama. Generation: 30–40 tok/s, leaving about 49 GB free for your other apps.

How it rates for each workload

Same model and machine, different jobs. Agents and long documents send far bigger prompts than chat, so they need much more speed and memory.

Workload	Verdict	Quant	Generation	First reply	Context
Chatwith Open WebUI	Excellent	Q6_K	30–40 tok/s	under 1 s	8K	Details
Coding questionswith Continue	Acceptable	Q6_K	25–35 tok/s	≈2 s	16K	Details
Coding in a repositorywith Aider	Borderline	Q8_0	18–25 tok/s	≈7 s	32K	Details
Agentic codingwith OpenCode	Borderline	Q8_0	17–25 tok/s	≈10 s	32K	Details
Long documents (64K)with Open WebUI	Acceptable	Q8_0	13–17 tok/s	≈35 s	64K	Details

Speeds are estimates calibrated against measured benchmarks on similar hardware. “First reply” is the time to read the workload’s opening prompt. How we calculate

Get it running

Copy-paste commands for the Q8_0 build with a 64K context. Ollama and llama.cpp download the file straight from Hugging Face.

ollama run hf.co/unsloth/Llama-3.1-8B-Instruct-GGUF:UD-Q8_K_XL

Downloads 10.58 GB (UD-Q8_K_XL). Ollama's default context is small: set OLLAMA_CONTEXT_LENGTH=65536 before `ollama serve` for long prompts.

Other models for NVIDIA DGX Spark (GB10) · 64 GB

The best-rated models for chat on this machine.

Everything NVIDIA DGX Spark (GB10) · 64 GB can run

Cheapest machines for Llama 3.1 8B Instruct

Hardware rated comfortable or better for chat, lowest price first.

Same chip, other memory: 128 GB

Different tool, context size or workload?

These pages use typical settings. The full check models your exact setup.

Open weights (restricted)Check my exact setup