NVIDIA RTX Spark is a new superchip for Windows laptops and compact desktops, and it uses the shared memory design Apple Silicon has used since the M1. A 20-core Grace CPU and a Blackwell RTX GPU with 6,144 CUDA cores draw from one pool of up to 128GB of LPDDR5X memory.
RTX Spark laptops went on sale October 16 from ASUS, Dell, HP, Lenovo, Microsoft and MSI, with more systems to follow. The lineup includes the ASUS ProArt P16, Dell XPS 16, HP OmniBook Ultra, Lenovo Yoga Pro 9n and MSI Prestige N16 Flip AI+. Microsoft’s Surface Laptop Ultra is the one aimed squarely at the MacBook Pro.
For an RTX Spark vs Apple Silicon comparison, the closest match is the MacBook Pro with M5 Pro and M5 Max. Those chips ship in the 14-inch and 16-inch models and use the same unified memory approach.
RTX Spark and the M5 Max Both Reach 128GB of Unified Memory
Memory capacity is where the two line up most closely. The top RTX Spark laptop and the MacBook Pro with M5 Max both reach 128GB, while M5 Pro stops at 64GB and the 18-core RTX Spark laptop variant stops at 32GB.
| Spec | NVIDIA RTX Spark (20-core laptop) | Apple M5 Max | Apple M5 Pro |
|---|---|---|---|
| CPU | 20-core NVIDIA Grace | 18-core | Up to 18-core |
| GPU | 6,144-core Blackwell RTX | Up to 40-core | Up to 20-core |
| Maximum memory | 128GB LPDDR5X | 128GB | 64GB |
| Memory bandwidth | Up to 300GB/s | Up to 614GB/s | Up to 307GB/s |
| AI hardware | Fifth-generation Tensor Cores, NVIDIA NPU, 1 petaflop FP4 | Neural Accelerator in every GPU core, 16-core Neural Engine | Neural Accelerator in every GPU core, 16-core Neural Engine |
| Operating system | Windows 11 | macOS Golden Gate | macOS Golden Gate |
| Battery claim | All-day, no hour count | Up to 22 hours of video streaming (16-inch) | Up to 24 hours of video streaming (16-inch) |
| Ports and displays | HDMI 2.1b and DisplayPort 2.1b support, varies by laptop maker | Three Thunderbolt 5 ports, HDMI, SDXC, up to four external displays | Three Thunderbolt 5 ports, HDMI, SDXC, up to three external displays |
The lower RTX Spark laptop variant pairs an 18-core CPU with a 5,120-core GPU. NVIDIA also lists a 140W desktop version with the same 20-core CPU and 6,144-core GPU, against 45 to 80W for the laptop chips.
The 128GB figure comes with a condition on the Windows side. Microsoft says the GPU in Surface Laptop Ultra can address only part of the 128GB, and the exact share depends on the configuration and workload.
RTX Spark Matches the M5 Pro on Memory Bandwidth and Trails the M5 Max
NVIDIA rates RTX Spark at 1 petaflop of FP4 AI performance on the 20-core chip and lists 827 teraflops of AI performance for the 18-core version. It also says the platform can run models of up to 120B parameters with a 1M context window. The petaflop figure is theoretical FP4 performance that relies on sparsity.
Apple states its AI gains as speedups over older chips, with the MacBook Pro rated at up to 8x faster AI performance than the M1 family. It pitches the M5 Max for running LLMs with hundreds of billions of parameters on the device.
Each company measures against a different baseline, so the headline gains do not line up. NVIDIA compares RTX Spark with Intel’s Core Ultra X9 388H and claims 4.6x on coding agents, 2.7x on fine-tuning and 1.8x on code compilation. Apple rates the M5 Max at up to 4x faster LLM prompt processing than the M4 Max, tested with a 14-billion-parameter model and a 16K-token prompt.
Memory bandwidth is where the two diverge. NVIDIA’s developer documentation puts RTX Spark at 300GB/s over a 256-bit interface. Apple lists 307GB/s for the M5 Pro and 614GB/s for the M5 Max with the 40-core GPU, or 460GB/s with the 32-core GPU.
Apple’s machine learning research describes the first token as compute-bound and every later token as limited by memory bandwidth. For a model that fits in memory, that makes bandwidth the number that sets text generation speed, and 300GB/s is roughly half of what the M5 Max offers.
RTX Spark Brings CUDA and DLSS 5 to Windows on Arm
CUDA runs natively on RTX Spark, which is the reason AI developers will look at it. NVIDIA and Microsoft released the first PyTorch-CUDA build for Windows on Arm, a Python 3.15 release that is native by default, more than 2,000 ready-made Conda packages, and Arm versions of over 97% of the top 15,000 Python packages.
Gaming is the other draw. RTX Spark supports ray tracing, Reflex and the full DLSS 5 suite, and NVIDIA lists hundreds of games ready at launch, including Hogwarts Legacy, NBA 2K27, Marvel Rivals and Fortnite.
The MacBook Pro runs macOS Golden Gate and tens of thousands of apps optimized for Apple silicon, from Final Cut Pro to Xcode.
Surface Laptop Ultra starts at $2,599.99 and goes up to 128GB of memory and 2TB of storage. The MacBook Pro with M5 Max comes with 36GB of memory and 2TB of storage at $4,099 for the 14-inch and $4,399 for the 16-inch after Apple’s June price increase, and the 14-inch MacBook Pro with M5 Pro starts at $2,499. Reaching 128GB requires the 20-core chip on RTX Spark and the M5 Max with the 40-core GPU on the MacBook Pro.
At 128GB, Surface Laptop Ultra costs $5,899.99 with 1TB of storage. The 16-inch MacBook Pro with M5 Max costs $6,999 with 2TB of storage, or $7,149 with the nano-texture display.
