Specs guide · Last checked August 1, 2026
NVIDIA RTX Spark Specs: What Is Confirmed and What Is Still Unknown
A cautious RTX Spark spec table covering up to 6,144 CUDA cores, 20-core CPU, 1 petaflop FP4 AI performance, 128GB unified memory, media, gaming, and launch gaps.
Fast answer
The safest specs are NVIDIA's own up-to numbers: 6,144 CUDA cores, 20-core CPU, 1 petaflop FP4 AI performance, and 128GB unified memory.
Important caveat
Exact laptop-by-laptop configurations, power limits, clocks, storage, and pricing are still OEM-specific unknowns.
What to watch next
Do not compare FP4 petaflops directly with NPU TOPS without explaining workload, precision, and memory differences.
Specification table
| Spec | Current value | Status | How to read it | Source |
|---|---|---|---|---|
| Blackwell RTX GPU | Up to 6,144 CUDA cores | Confirmed | Official product-page maximum. Exact device clocks and power limits remain OEM-specific. | NVIDIA RTX Spark product page |
| CPU | Up to 20-core ultra-efficient CPU | Confirmed | NVIDIA describes the CPU as part of the RTX Spark Superchip; final laptop SKUs still need product pages. | NVIDIA RTX Spark product page |
| AI performance | Up to 1 petaflop FP4 AI performance | Confirmed | Useful for local AI positioning, but not directly comparable with NPU TOPS without workload context. | NVIDIA RTX Spark product page |
| Unified memory | Up to 128GB unified memory | Confirmed | The memory pool is the key local AI differentiator for larger models and agent workflows. | NVIDIA RTX Spark product page |
| Local agents and LLMs | 120B-parameter LLMs and up to 1M token context stated by NVIDIA | Announced | NVIDIA's release states this workload target; real-world throughput and usability need launch testing. | NVIDIA Newsroom release |
| Creator workloads | 12K 4:2:2 video, 4K AI video, and large 3D scenes stated by NVIDIA | Announced | Workload support depends on final apps, drivers, memory configuration, and thermals. | NVIDIA Newsroom release |
| Gaming | RTX graphics, DLSS suite, Reflex, G-SYNC; NVIDIA 1440p/100 FPS statement | Announced | Treat as launch positioning until independent laptop and desktop benchmarks exist. | NVIDIA RTX Spark product page |
| Price | Not published | Unknown | No safe pricing answer until OEM pages or retail listings appear. | NVIDIA Newsroom release |
How to read up-to numbers
Treat every headline number as a platform maximum until OEMs publish final SKUs. A slim laptop can have different power, memory, and thermal limits than a compact desktop.
Related RTX Spark guides
Compare Windows PC use with a dedicated desktop AI development system.
Start with the overview, confirmed facts, and recommended next guides.
Track announced RTX Spark devices and fall 2026 availability.
Fall 2026 timing, OEM partners, and proof needed.
Agents, LLMs, coding, creative AI, and 128GB unified memory.
RTX, DLSS, Reflex, 1440p statements, and review checklist.
Compare Windows RTX/CUDA AI PCs with Mac silicon.
Compare CUDA/RTX AI with efficient Windows NPU PCs.
Frequently asked questions
What are the confirmed RTX Spark specs?
NVIDIA lists up to 6,144 CUDA cores, up to a 20-core CPU, up to 1 petaflop FP4 AI performance, and up to 128GB of unified memory.
Does every RTX Spark laptop have 128GB of memory?
Not necessarily. NVIDIA says up to 128GB unified memory. Exact configurations depend on OEM SKUs.
Can I compare RTX Spark FP4 performance with NPU TOPS?
Only cautiously. FP4 GPU AI performance and NPU TOPS are different metrics tied to different workloads and precision levels.
Sources and limits
Update this guide only when NVIDIA, Microsoft, an OEM, or a first-party benchmark/review publishes stronger evidence.