Nvidia's Arm CPU dream hits a hardware wall, debut pushed to 2026

Alfonso Maruccia

Posts: 1,864   +564
Staff
TL;DR: Nvidia's first CPU based on the Arm architecture has been the subject of growing buzz, signaling the company's ambitions to challenge dominant players like Intel, AMD, and Apple. But that move is now hitting serious turbulence. A newly uncovered hardware issue has forced another delay, and engineers are reportedly going back to the drawing board to resolve this latest issue.

Nvidia has encountered a new hardware problem with its much-anticipated N1x Arm CPU, and this time it is a major one. According to multiple industry sources cited by SemiAccurate, the issue may require a significant change to the silicon design. As a result, both the chip's debut and shipping dates have been pushed back to 2026.

Nvidia engineers met a first significant issue earlier this year, but they were allegedly able to solve it without messing with the actual silicon design. The company previously claimed that the N1 and N1x chips had entered full production, which now appears to have been an overly optimistic assessment if the latest rumors leaks are accurate.

A few weeks ago, a supposed "Nvidia N1x" sample appeared in Geekbench's online benchmark database, showing some pretty interesting performance scores. The chip appeared capable of competing with some of the best mobile processors currently on the market. The Nvidia N1 series is expected to go head-to-head with Qualcomm's Snapdragon X Elite, Intel's Core Ultra 200HX series, Apple's M3, and AMD's Ryzen AI Max.

Despite the reported setback in the silicon design, the Nvidia N1 project is real and gaining momentum among industry insiders. Multiple reports have confirmed that the AI behemoth is developing its Arm-based chip with several hardware partners reportedly involved in the venture. Needless to say, the company has gained the resources in the past couple of years to develop this and any other project they wanted.

Nvidia showcased its "AI PC" in January and is now expected to release a more affordable version of the system developed in collaboration with MediaTek.

More recently, insiders have hinted at a possible partnership with Alienware for a gaming laptop featuring an Arm-based CPU paired with a built-in GeForce GPU.

Customers are not exactly queueing up to buy Arm-based laptops for gaming, but Nvidia seems committed to push forward. Some observers have noted that promoting the Arm architecture in the traditional PC market is a questionable move, especially since Intel and AMD continue to improve the x86 architecture at a steady pace.

Nvidia's struggles with the N1 project feel like a case of history repeating itself. In the company's early days, its initial graphics chips failed to gain traction, and Nvidia came close to bankruptcy before rebounding with the successful launch of the RIVA 128 in the late 1990s. For those interested in the company's early history, the 86Box blog offers a fascinating deep dive, where a crazy programmer is working on a functional emulation of the RIVA 128 and NV3 architecture.

Permalink to story:

 
I wasn’t aware of either Intel or AMD making any major improvements to x86 recently. IIRC x86S was shelved by Intel…
Starting 2017, AMD has released ar least Zen, Zen+, Zen2, Zen3, Zen4 and Zen5. Zen6 comes 2026. Intel has also released multiple architectures like Sunny cove, Willow cove, Golden cove, Lion cove etc.

AMD has had more "significant" advances but only because starting point was much lower and ARM still mostly competes on low power only.

X86S was probably cancelled because most legacy stuff can be done with microcode anyway.
 
Starting 2017, AMD has released ar least Zen, Zen+, Zen2, Zen3, Zen4 and Zen5. Zen6 comes 2026. Intel has also released multiple architectures like Sunny cove, Willow cove, Golden cove, Lion cove etc.

AMD has had more "significant" advances but only because starting point was much lower and ARM still mostly competes on low power only.

X86S was probably cancelled because most legacy stuff can be done with microcode anyway.
I think the biggest issue an nVidia ARM chip will have is that it's compatibility isn't going to match the price. The Windows ARM laptops were bad enough and they were actually cheaper than it's x86 counter parts. I have a feeling that nVidia thinks their branding alone doubles the MSRP and it will likely have all the same problems that ARM for windows. I originally thought there ARM chips were supposed to be server chips designed specifically to be used in their data centers. I'll go RISC-V over an ARM chip for a desktop or laptop anyday and it has nothing to do with nVidia.
 
Starting 2017, AMD has released ar least Zen, Zen+, Zen2, Zen3, Zen4 and Zen5. Zen6 comes 2026. Intel has also released multiple architectures like Sunny cove, Willow cove, Golden cove, Lion cove etc.

AMD has had more "significant" advances but only because starting point was much lower and ARM still mostly competes on low power only.

X86S was probably cancelled because most legacy stuff can be done with microcode anyway.
Biggest reason X86S was taken behind the woodshed was it was ultimately pointless. Slimming x86 down would have broken a metric ton of legacy applications, and for no real perceivable benefit, as the x86 code block is absolutely tiny and most modern software is written for AMD64 anyway.
 
Biggest reason X86S was taken behind the woodshed was it was ultimately pointless. Slimming x86 down would have broken a metric ton of legacy applications, and for no real perceivable benefit, as the x86 code block is absolutely tiny and most modern software is written for AMD64 anyway.
Idea was pretty good as CPU development do suffer from all sorts of basically useless legacy stuff. However since backwards compatibility is only advantage that x86 has, trimming out useless legacy is very hard. As for tiny size, leaving out 32 bit "full speed" could free some resources from critical positions on CPU. Now most 32 bit software run without much performance penalty and probably at some time x86s could be achieved just doing 64 bit full speed and putting separate unit for 32 bit code.
 
Back