NVIDIA RTX Spark Pairs Up to 128 GB Unified Memory With DLSS 5 and Local Nemotron 3.5 Bolt

NVIDIA RTX Spark, a Windows PC platform for AI development, creation and gaming, combines an NVIDIA Blackwell RTX GPU and NVIDIA Grace CPU with up to 128 GB of unified memory, according to an NVIDIA developer blog post published October 27, 2026. It targets slim laptops and compact desktops and carries the full NVIDIA CUDA stack plus RTX technologies including ray tracing and DLSS 5.

DLSS 5 and the DLSS suite

DLSS 5 with 3D-Guided Neural Rendering is supported on RTX Spark, starting with NBA 2K27. NVIDIA says an October model update runs 2x faster than the September version and halves GPU overhead while maintaining comparable visual fidelity. It also reduces VRAM use at 4K by 20%. The model will ship by default in future DLSS 5 games.

NVIDIA describes DLSS 5 as a final stage in the graphics pipeline that enhances lighting and material detail. The engine-rendered frame remains the foundation, and developer controls determine what can change. Avalanche Software used the DLSS 5 SDK controls to tune its integration to preserve the artistic intent and style of its game. NVIDIA's comparison figure uses Hogwarts Legacy.

RTX Spark also supports DLSS Super Resolution, Ray Reconstruction and Dynamic Multi Frame Generation, which can generate up to 5 additional frames per rendered frame. The latest NVIDIA app beta adds a faster Performance Mode (Preset C) for Frame Generation, with minor image-quality trade-offs.

The DLSS 4.5 Unreal Engine plugin, version 8.9.0, adds official Windows on Arm support for Unreal Engine 5.8, including native ARM64 binaries and support for building and packaging UE 5.8 applications. On GeForce RTX 50 Series and RTX PRO Blackwell GPUs, 6x Multi Frame Generation produces up to five additional frames per rendered frame.

Nemotron 3.5 Bolt E4B 4GB for AI characters

NVIDIA ACE introduces Nemotron 3.5 Bolt E4B 4GB for on-device AI characters on GeForce RTX and RTX Spark. NVIDIA claims about 40% higher average quality than the prior model on relevant benchmarks and up to about 80% fewer output tokens than peers. Developers can access it through the In-Game Inferencing (IGI) SDK.

Tooling

• RTX Kit: eight neural rendering SDKs on RTX Spark, including RTX Neural Texture Compression, RTX Mega Geometry, and RTX Character Rendering for strand-based hair and skin with path tracing.
• Nsight: support for native Arm64, emulated x64 and Arm64EC applications from day one. Nsight Graphics 2026.3 adds a Windows on Arm installer. Nsight Systems 2026.5 provides a single installer for all Windows versions.
• Unreal Engine: the NVIDIA RTX Branch of Unreal Engine 5.8 provides an updated base for ray tracing and neural graphics in UE 5.8 projects.
• MotionBricks: new SOMA human-motion models, downloadable checkpoints, ONNX exports, and an interactive reference implementation built with Newton.

What this means

The post positions RTX Spark as a single target for graphics, neural rendering and local language-model inference, with the same CUDA and RTX stack developers already use. The Windows on Arm pieces (UE 5.8 plugin support, Nsight installers, support for native and emulated applications) address the porting path for existing projects. The Nemotron 3.5 Bolt figures are NVIDIA's own claims against unnamed benchmarks and peers, so developers should test the model on their own character workloads before committing to it. The DLSS 5 gains (2x speed, halved GPU overhead, 20% lower VRAM at 4K) are likewise NVIDIA's comparison of its October model against its September one.