Has NVIDIA developed a chip that takes on all the workloads of both CPUs and GPUs?

That's right, all the previous rumors are true. NVIDIA has also launched its own PC processors. From now on, when you purchase a computer, you can choose not only NVIDIA's graphics cards, but also NVIDIA's CPUs...

Here's the backstory: On October 8, we arrived in San Francisco, USA, and attended a closed-door event named NVIDIA Spark Editor's Day.

At the event, we witnessed the official launch of NVIDIA's brand-new superchip series, the NVIDIA RTX Spark.

Through some special behind-the-scenes arrangements, we even got a Lenovo YOGA Pro 15 RTX Spark equipped with the NVIDIA RTX Spark N1X right on site.

In fact, as early as two years ago, there were leaks saying that NVIDIA was going to develop ARM architecture processors for PCs.

We even received leaks that we should never look down on this chip just because it is an ARM processor, it performs extremely well in games, and it is even better than other ARM architecture chips from Qualcomm and Apple.

After waiting for two years, we finally got the official release.

It has to be said that the specifications of this chip look really impressive...

It has up to 20 CPU cores, and the GPU consists of 6144 CUDA cores.

To make a rough non-rigorous comparison, the number of GPU cores is already comparable to the desktop RTX 5070.

Backed by NVIDIA's NVLink-C2C technology, the interconnection bandwidth between CPU and GPU can reach 600GB/s.

It also adopts the unified memory architecture that most other ARM processors use, supporting up to 128GB of LPDDR5X memory with a theoretical bandwidth of 300GB/s.

Of course, it would be boring to just read out the specifications, so we will directly show you the benchmark results to see what level this chip actually reaches ——

The classic CineBench R23 does not have a Windows ARM version, so running it directly on this YOGA Pro 15 RTX Spark will bring x86 to ARM translation loss. So we mainly ran CineBench 2026, and also released the R23 results for you to roughly estimate its performance level by comparing with your own devices...

Finally, the single-core and multi-core scores of the RTX Spark N1X CPU are 548 and 7116 respectively.

When running on battery power, the scores will drop slightly, with single-core and multi-core scores reaching 511 and 6554 respectively.

We also brought in another "test subject" —— we happen to have a test platform with AI MAX+ 395 in our lab. This x86 architecture chip has a similar design concept to Jensen Huang's N1x, so we can use it for comparison.

As you can see, in the same CineBench 2026 test, the single-core and multi-core scores of the AI MAX+ 395 CPU are 457 and 7793 respectively.

And its single-core score in CineBench R23 is 2031, and the multi-core score is 35312.

Here is a summary of the results for your convenience of comparison ——

Considering that the CPU part of this chip is mainly developed by MediaTek, which has never made PC processors before, it is already quite impressive that its first work can compete head to head with the star product from industry veteran AMD.

But when it comes to the GPU part, the situation is completely different.

We first tested its AI model performance. When deploying Qwen3.8 locally, the RTX Spark N1X can output about 32 Tokens per second when summarizing articles.

When generating scripts, the output speed is about 30 Tokens per second.

For comparison, the data obtained by AI MAX+ 395 on these two tasks are 25.37 and 25.83 respectively.

After testing AMD's chip, we also asked @Dakuang to test the performance of Apple's M5 Max chip: with the same prompt, the output speed of article summarization is 43.86 tok/sec, and the output speed of script generation is 36.59 tok/sec.

As you can see, although RTX Spark is very powerful, Apple's chip has larger scale, so NVIDIA lost this round narrowly. (Fun fact: The 128GB unified memory versions of N1X and M5 Max laptops are both priced above 40,000 RMB, one is over 40,000 and the other is over 50,000)

Then let's look at the 3DMark results: The RTX Spark N1X scores 13517 in Steel Nomad Light, and 11294 in SolarBay Extream.

While the scores of AI MAX+ 395 in these two tests are 10791 and 6881 respectively.

It can only be said that when it comes to GPU performance, Uncle Huang's products are really outstanding, and they achieved excellent results immediately.

However, this does not mean that the RTX Spark N1X is perfect and invincible in all aspects.

We ran the dedicated battery life test script from Geekwan on this Yoga Pro15 RTX Spark to test its battery performance. The result shows that this laptop with a 92 Wh battery only lasted 520 minutes from full charge to shutdown.