What AI models can NVIDIA RTX Spark (128GB Unified) run?

The NVIDIA RTX Spark (128GB Unified) can run 0 AI models. See the full compatibility table above for speeds and quality ratings.

Is NVIDIA RTX Spark (128GB Unified) good for AI coding?

Yes. With 128 GB, the NVIDIA RTX Spark (128GB Unified) supports the Full AI Builder tier: concurrent coding + reasoning + embeddings.

How much memory does NVIDIA RTX Spark (128GB Unified) have?

The NVIDIA RTX Spark (128GB Unified) has 128 GB of unified memory with 273 GB/s bandwidth.

Can NVIDIA RTX Spark (128GB Unified) run 70B models?

70B models typically require 24-48 GB of inference memory. The NVIDIA RTX Spark (128GB Unified)'s 128 GB may not be sufficient. Consider a higher-tier device.

Is NVIDIA RTX Spark (128GB Unified) worth it for AI?

Pricing for NVIDIA RTX Spark (128GB Unified) has not been announced. It offers 128 GB unified memory, but OwnRig should treat recommendations as provisional until pricing and benchmarks are available.

Unified SoCupcoming

Unified SoC

NVIDIA RTX Spark (128GB Unified)

128 GB Unified · 273 GB/s

From

TBA

Pricing not announced

Unified Memory

128 GB

Bandwidth

273 GB/s

TDP

100W

Models

Tier

Full

The NVIDIA RTX Spark (128GB Unified) with 128 GB unified memory. Performance data is being compiled.

Source: OwnRig methodology

Unified Memory

128 GB

Bandwidth

273 GB/s

Memory Type

Unified

TDP

100W

GPU Cores

6144

CPU Cores

FP4 AI

1000 TOPS

Host Systems

Surface Laptop Ultra, Surface RTX Spark Dev Box, ASUS ProArt P16, Dell XPS 16, HP OmniBook X 14, Lenovo Yoga Pro 9n, MSI Prestige N16 Flip AI+

Builder Capability: Full AI Builder

Supports concurrent coding + reasoning + embeddings. Can run 70B models at quantized precision.

Software

Inference Backends

The software stacks that matter most for real-world inference on this device.

CUDA

beta

CUDA is part of the RTX Spark platform, but final Windows Arm runtime behavior for local LLM stacks should be verified on shipping systems.

TensorRT-LLM

beta

Primary NVIDIA optimization path for local agent workloads; OwnRig has no RTX Spark tok/s benchmarks yet.

Windows ML

experimental

Relevant for Windows-native local models and Windows Copilot Runtime; do not treat as validated for arbitrary GGUF workloads.

Ready to Buy

Available in these Machines

MicrosoftLaptop

Microsoft Surface Laptop Ultra (RTX Spark, 128GB)

TBA

MicrosoftMini

Microsoft Surface RTX Spark Dev Box (128GB)

TBA

FAQ

Frequently Asked Questions

What AI models can NVIDIA RTX Spark (128GB Unified) run?: The NVIDIA RTX Spark (128GB Unified) can run 0 AI models. See the full compatibility table above for speeds and quality ratings.
Is NVIDIA RTX Spark (128GB Unified) good for AI coding?: Yes. With 128 GB, the NVIDIA RTX Spark (128GB Unified) supports the Full AI Builder tier: concurrent coding + reasoning + embeddings.
How much memory does NVIDIA RTX Spark (128GB Unified) have?: The NVIDIA RTX Spark (128GB Unified) has 128 GB of unified memory with 273 GB/s bandwidth.
Can NVIDIA RTX Spark (128GB Unified) run 70B models?: 70B models typically require 24-48 GB of inference memory. The NVIDIA RTX Spark (128GB Unified)'s 128 GB may not be sufficient. Consider a higher-tier device.
Is NVIDIA RTX Spark (128GB Unified) worth it for AI?: Pricing for NVIDIA RTX Spark (128GB Unified) has not been announced. It offers 128 GB unified memory, but OwnRig should treat recommendations as provisional until pricing and benchmarks are available.

Own this GPU?

See every AI model it supports, expected performance, and how to build around it.

Check my rig

All GPUs