First NVIDIA RTX Spark Laptop Reviews Are In: How Fast Are the New ARM Chips?

NVIDIA is entering the laptop processor market with ambitions to compete with Intel, AMD, Qualcomm, and Apple. Its RTX Spark is not a traditional processor, but a complete chip that combines a CPU and GPU, with both sharing the same memory. The platform is designed for creative work and gaming, but above all for running AI models locally. In today's launch, NVIDIA is also highlighting CUDA, local AI, and DLSS 5 support.

The first laptops are from ASUS, Dell, HP, Lenovo, MSI, and Microsoft. Prices start at $2,599, while the top configurations cost more than $5,000. Reviews so far have focused mainly on the more powerful RTX Spark N1X, with 20 CPU cores and 6,144 CUDA cores. A less powerful version is also available, with 18 CPU cores and 5,120 CUDA cores.

Graphics performance closer to an RTX 5060 than an RTX 5070, but with good efficiency

The new chip's biggest strength is its graphics performance. In synthetic benchmarks, the more powerful RTX Spark can perform at roughly the level of a laptop GeForce RTX 5070. For example, Notebookcheck measured 3,134 points in 3DMark Steel Nomad on the ASUS ProArt P16, while the ProArt P16 with an RTX 5070 scored 2,935 points. The RTX Spark scored 13,030 points in Time Spy.

In actual games, however, the picture is more complicated, and performance is often closer to that of a laptop RTX 5060. Because the RTX Spark uses an ARM processor and Windows on ARM, many games still run through the Prism emulation layer. This adds CPU overhead and can noticeably reduce performance depending on the game. CPU-intensive titles in particular can hold back the graphics side.

For example, PC Gamer compared the RTX Spark in the ASUS ProArt P16 with a laptop RTX 5070 capped at 115 W. At 1080p, it measured:

• Cyberpunk 2077, RT Ultra: 36 fps vs. 45 fps
• Black Myth: Wukong, High: 66 fps vs. 65 fps
• Metro Exodus Enhanced, Ultra: 75 fps vs. 84 fps
• Baldur's Gate 3, Ultra: 42 fps vs. 65 fps

The GPU's efficiency is particularly notable. In Metro Exodus, the RTX Spark delivered 75 fps at an average GPU power draw of approximately 43 W, while the RTX 5070 in the Razer Blade 14 needed approximately 97 W to reach 84 fps. These figures apply to the graphics component alone, not the entire laptop.

The RTX Spark also supports NVIDIA technologies including DLSS 5 and Frame Generation, which can further boost performance in supported games. The platform is certainly capable of gaming, but given the laptops' high prices, it is not yet the best-value alternative to conventional gaming systems.

CPU performance is the weak point

The CPU side is less impressive. The more powerful RTX Spark has 20 ARM cores: ten Cortex-X925 cores and ten Cortex-A725 cores. The chip's CPU, I/O, and memory components were developed in collaboration with MediaTek.

Notebookcheck measured 1,726 points in the Cinebench 2024 multi-core test. That's more than enough performance for everyday work, but the RTX Spark falls behind the fastest solutions in demanding CPU applications, compilation, and some games. For example, LTT Labs recorded better results on Apple and Qualcomm processors when compiling Godot, and also found the CPU side weaker in other tests. Read more at LTT Labs.

The ARM CPU cores used here have room for improvement. The RTX Spark uses Cortex cores from the 2024 generation, while Apple and Qualcomm already have newer or more extensively customized designs. This is an area where NVIDIA has room to improve in future generations.

128 GB of memory and AI are its biggest advantages

The RTX Spark also targets a very different market. Thanks to its unified memory, the GPU can access significantly more capacity than a typical discrete graphics card. The highest configurations come with up to 128 GB of RAM, making it possible to run very large language models locally or work with extensive 3D scenes. NVIDIA says it can handle 3D scenes larger than 90 GB.

This does not automatically mean that the RTX Spark generates images or AI responses faster than conventional high-performance GeForce cards. Discrete graphics cards with higher memory bandwidth can be considerably faster with smaller models. The RTX Spark's main advantage is its ability to run models that will not fit in their graphics memory at all.

Support for CUDA, TensorRT, and other NVIDIA technologies also helps. In its LLM tests, LTT Labs also recorded significantly better results for the RTX Spark when processing prompts, while the differences between platforms were smaller when generating tokens. The combination of NVIDIA's software ecosystem and large memory makes the platform an interesting option for developers and users working with local AI.

Battery life falls behind

Reviews also cover battery life, with results varying significantly depending on the laptop and how it is used. For example, LTT Labs measured 12 hours and 2 minutes of video playback on the ASUS ProArt P16. The laptop does, however, have a large 99.9 Wh battery. Under heavy graphics workloads, it lasted just 1 hour and 9 minutes.

Using ARM therefore does not automatically mean efficiency on par with the best Snapdragon devices or MacBooks. For example, LTT Labs got 12 hours and 40 minutes from the Snapdragon X2 Elite Extreme with a battery of just 70 Wh.

As for Windows on ARM compatibility, Microsoft has made significant progress, and most everyday applications now work without major issues. Some programs and games can still encounter problems, however. Apps without a native ARM version have to use Prism, and some games with anti-cheat systems can also be problematic.

A conventional laptop with a GeForce RTX is still the better choice for gamers, while the RTX Spark may be a much more interesting alternative for developers, creators, and professionals working with large AI models. However, prices are high, as the laptops come with 64 GB or 128 GB of memory.

Comments

Name:

:)