From the NVIDIA edition of October 16, 2026
Moor Insights & Strategy: RTX Spark leads Snapdragon on GPU and AI but its CPU finished last in every test and Windows on Arm remains immature
Moor Insights & Strategy (Anshel Sag) · Industry analyst firm / hands-on testing · United States
- Publisher
- Moor Insights & Strategy
- Original publication
- 2026-10-16
- Source retrieved
- 2026-10-17
- Event date
- 2026-10-16
What happened
In Oct 16 field notes, the analyst found RTX Spark's GPU more than doubled Snapdragon X2 Elite Extreme in 3DMark tests and beat Apple's M5 Max on MLPerf time to first token, but its CPU finished last in every CPU test (Geekbench 7 single-core about 2,600 versus 3,300 and 3,600). GPU-Z crashed the whole system on every run, some anti-cheat games would not run, the 55 TOPS NPU is unused with Copilot+ support undated, and battery life trails Qualcomm and Apple. He sees 27-28 billion parameter models as the current local sweet spot and says the CPU gap likely waits for a next-generation chip.
Why it matters
Useful counterweight for SLED device standards teams: it frames RTX Spark as a portable local AI machine rather than a general-purpose replacement, which bounds where institutions should pilot it. The crash and compatibility findings matter for managed fleets that rely on monitoring and utility software.
Evidence and measured results
Analyst's own benchmarks on one $5,899 128GB unit, with comparison table versus Snapdragon X2 Elite Extreme and M5 Max. Method details (runs, settings) are limited; the page does not disclose whether NVIDIA or Microsoft are clients of the firm.
Limitations and uncertainty
Single tester and unit; anecdotal issues (screen flicker resolved itself) may be fixed by drivers. Launch memory for the $2,599 base unit is stated as 32GB here but 24GB in the Computerworld review; unresolved.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-17; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Device standards teams need to know where an Arm-based local AI laptop fits before buying beyond a pilot.
- Stakeholders
- Endpoint standards owner
- Research computing
- Information security / endpoint protection team
- Discovery questions
- Which roles are CPU-bound versus GPU or AI-bound?
- Which monitoring and security agents must run on every device?
- Is the refresh timeline flexible enough to wait for 2027 models?
- Value hypothesis
- Interpretation: targeting only AI-heavy roles may capture the GPU and memory advantage while avoiding CPU and compatibility shortfalls.
- Potential engagement
- Role-segmentation workshop plus a compatibility test plan covering security agents, utilities and CPU-bound tasks.
- Unsupported claims
- General-purpose CPU leadership
- Copilot+ NPU features at launch
- Fully mature Windows on Arm software support
Pre-sales engineering
Role takeaway
- Fit
- Strong for local inference; weaker for CPU-bound and compatibility-sensitive workloads.
- Architecture and integration
- Blackwell GPU plus 20-core Arm CPU over NVLink-C2C; unified LPDDR5X memory; NPU present but not yet used by Copilot+.
- Prerequisites
- Full software-stack compatibility list
- CPU-bound benchmark set
- Driver update process for Arm devices
- Constraints
- CPU last in analyst tests
- Utility-induced system crash observed
- Anti-cheat and some apps unsupported on Arm
- Security
- Test endpoint protection, monitoring and management agents on Arm before deployment; driver maturity is a stability risk.
- Proof of value
- Run security and monitoring agents for a week without crashes
- Compare CPU-bound task times to current devices
- Benchmark a 27-28B model task end to end
Delivery
Role takeaway
- Work
- Execute a compatibility and stability test of the institutional software stack, then a role-limited user pilot with crash and ticket tracking.
- Dependencies
- Endpoint protection vendor Arm support
- OEM driver updates
- Pilot user group
- Ownership
- Endpoint engineering team.
- Skills and adoption
- Support staff need Arm troubleshooting skills; users need guidance on which tasks suit the device.
- Governance checkpoints
- Security stack certification on Arm
- Pilot exit review
- Refresh-timing decision versus 2027 platform
- Acceptance criteria
- Proposed: zero system crashes from required software over the pilot
- Proposed: CPU-bound tasks within an agreed tolerance of current devices
- Risks
- Driver or utility crashes
- Users in CPU-heavy roles see regressions
- Buying ahead of a cheaper 2027 platform
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
GPU and memory suit local inference; CPU-bound workloads (compilation, CPU rendering, general responsiveness) lag peers. Large 99 Wh batteries and 140W chargers suggest a heavier device class.
Governance
Who approves, reviews and stays accountable for outcomes?
Undated Copilot+ NPU support and immature software suggest institutions should gate adoption on a compatibility review rather than vendor timelines.
Security and privacy
What data, permissions and controls need testing?
System-level crash from a common utility points to driver maturity risk; security tooling compatibility on Arm should be tested before deployment.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Users of CPU-heavy or niche tools may see worse experience than on current devices; training should set expectations on what runs well.
Procurement
What should contracts, pricing and exit terms secure?
Price comparisons cited: Snapdragon X2 Elite Extreme about $2,199 with 48GB; M5 Max with 128GB $6,899. A lower-priced platform is planned for 2027, which may affect refresh timing.
Operating model
Which teams own the service once it runs?
Suggests a split device strategy: RTX Spark for local AI roles, other platforms for CPU-heavy or compatibility-sensitive users.
Publication history
- 2026-10-16NVIDIA · Issue 416 resources
Stable resource ID: moor-insights-rtx-spark-cpu-software-limits-20261016