Nvidia RTX Spark Superchip beats M5 Max by 25% in local AI benchmarks

Nvidia's RTX Spark Superchip in the Surface Laptop Ultra beats the M5 Max by 25% in local AI benchmarks, but gaming performance is limited.

Nvidia RTX Spark Superchip beats M5 Max by 25% in local AI benchmarks

's RTX Spark Superchip makes its debut in Surface Laptop Ultra. The focus is firmly on local AI workloads, not gaming. This is a significant platform because it paves a way for you to run your LLMs on laptops without needing the cloud. For those interested in having on-device AI inference capabilities for development/productivity use cases, this is very useful right away, albeit pricey.

The $5899 Surface Laptop Ultra excels at local AI but struggles with gaming

The platform is built around Nvidia's GB10 Superchip. It packs 20 CPUs and 6144 CUDA cores. Microsoft has deployed the silicon in the Surface Laptop Ultra, where you get the superchip and 128GB of LPDDR5X-9400 RAM. That's a max bandwidth of 300 GB/s for shared memory that can be used for both system and AI workloads.

Specifications

  • CPU Cores: 20
  • CUDA Cores: 6144
  • RAM: 128GB LPDDR5X-9400
  • Memory Bandwidth: 300 GB/s
  • Price: $5899

We can see that the RTX Spark platform crushes its competition on local AI benchmarks. On the AgentPerf-local test (Qwen 3.8 27B), it's 25% faster than the MacBook Pro M5 Max. For the Qwen 3.6 35B (A3B), the RTX Spark platform completes the test in 8 minutes. This is 25% faster than the 10 minutes it takes on the Apple machine.

But because of Windows on Arm, you're limited in gaming performance. On Gears of War: E-Day, the system averages 77.7 FPS. On Control Resonant, you're looking at 109.2 FPS. That shows that due to the CPU emulation overhead and shared memory, your GPU isn't able to shine in a standard title.

The RTX Spark Surface Laptop Ultra retails for $5899USD (~$5899). Lead times go out as far as late December. Nvidia is trying to reinvent the PC with this.

However, some of the promised secure agentic AI features for Windows 11 are not ready yet. So right now, this platform is for those who are happy to pay a premium for early access to local AI workloads.

We've been keeping tabs on this silicon's progress. Remember the Microsoft Surface Laptop Ultra Launches Oct? report we did? The Nvidia RTX Spark Superchip is great when it comes to local AI.

However, you'll need to live with gaming trade-offs. And it's also expensive. This is something for a very niche market focused on AI use cases. Not something that will generally replace your laptop.

Discussion

0 comments

Log in to join the thread with a thoughtful take, question, or correction.

Add to the discussion