NVIDIA · unified memory

NVIDIA DGX Spark

The NVIDIA DGX Spark has 128GB of unified memory, roughly 115GB of it GPU-addressable, with 273 GB/s of bandwidth. The largest model that loads cleanly at a quality quant is DeepSeek-V4-Flash (llama.cpp, 8K context).

Unified memory: 128GB shared between CPU and GPU; we count ~90% as GPU-addressable for model weights. This class fits huge models but is bandwidth-bound: a box that fits a 120B-class model can still decode slowly, while prompt processing may land a higher tier — one headline number would mislead in both directions, so speed here is a tier, labeled beta until calibrated.

Unified memory: 128 GB
GPU-addressable: ~115 GB (90%)
Memory bandwidth: 273 GB/s
Backend: cuda
Price (as configured): $4,699

Specs as of 2026-07-11 · source

Check price on Amazon

Affiliate link — we may earn a commission. Verdicts are computed before any link is attached.

What an NVIDIA DGX Spark can run

Every model we track, graded at its best clean-fit quant.

Model	Best-quant verdict · llama.cpp · 8K context
Qwen3-8B	Fits · A · Q8_0interactivebeta
gemma-4-26B-A4B-it	Fits · A · Q8_0interactivebeta
Qwen2.5-7B-Instruct	Fits · A · Q8_0interactivebeta
Qwen2.5-1.5B-Instruct	Fits · A · Q8_0interactivebeta
Llama-3.2-1B-Instruct	Fits · A · Q8_0interactivebeta
Llama-3.1-8B-Instruct	Fits · A · Q8_0interactivebeta
DeepSeek-R1	Won't fitbeta
Qwen3.5-9B	Fits · A · Q8_0usablebeta
gpt-oss-20b	Fits · A · MXFP4interactivebeta
Qwen3.6-35B-A3B	Fits · A · Q8_0interactivebeta
Qwen2.5-3B-Instruct	Fits · A · Q8_0interactivebeta
Qwen3-32B	Fits · A · Q8_0painfulbeta
Qwen3.6-27B	Fits · A · Q8_0painfulbeta
Qwen2.5-0.5B-Instruct	Fits · A · Q8_0interactivebeta
Mistral-7B-Instruct-v0.3	Fits · A · Q8_0interactivebeta
gpt-oss-120b	Fits · A · MXFP4interactivebeta
Qwen3-14B	Fits · A · Q8_0usablebeta
Qwen3-30B-A3B	Fits · A · Q8_0interactivebeta
Qwen2.5-32B-Instruct	Fits · A · Q8_0painfulbeta
DeepSeek-V4-Flash	Fits · B · UD-Q2_K_XLinteractivebeta
Qwen2.5-14B-Instruct	Fits · A · Q8_0usablebeta
Llama-3.2-3B-Instruct	Fits · A · Q8_0interactivebeta
gemma-3-4b-it	Fits · A · Q8_0interactivebeta
gemma-3-12b-it	Fits · A · Q8_0usablebeta
Llama-3.1-70B-Instruct	Fits · A · Q8_0painfulbeta
Phi-3.5-mini-instruct	Fits · A · Q8_0interactivebeta
Phi-4	Fits · A · Q8_0usablebeta
DeepSeek-R1-Distill-Qwen-32B	Fits · A · Q8_0painfulbeta
gemma-3-27b-it	Fits · A · Q8_0painfulbeta
Llama-3.3-70B-Instruct	Fits · A · Q8_0painfulbeta
Qwen2.5-72B-Instruct	Fits · B · Q8_0painfulbeta
gemma-2-2b-it	Fits · A · Q8_0interactivebeta
DeepSeek-R1-Distill-Qwen-14B	Fits · A · Q8_0usablebeta
DeepSeek-R1-Distill-Llama-8B	Fits · A · Q8_0interactivebeta
DeepSeek-R1-Distill-Qwen-7B	Fits · A · Q8_0interactivebeta
gemma-2-9b-it	Fits · A · Q8_0usablebeta
Mistral-Nemo-Instruct-2407	Fits · A · Q8_0usablebeta
QwQ-32B	Fits · A · Q8_0painfulbeta
Mistral-Small-24B-Instruct-2501	Fits · A · Q8_0painfulbeta
gemma-2-27b-it	Fits · A · Q8_0painfulbeta

Compare NVIDIA DGX Spark

Have a specific model in mind? Check it against the NVIDIA DGX Spark at your own quant and context length.

Check with the NVIDIA DGX Spark preselected →

Verdicts computed from published specs and measured GGUF sizes — how the math works.