Nvidia Arm chip surfaces with strong Geekbench scores, could rival top Intel and AMD laptop CPUs

[Daniel Sims]

Something to look forward to: A mysterious "Nvidia N1x" Arm processor recently appeared on Geekbench with a single-core score of 3,096 and an 18,837 multi-core score. The result provides some of the earliest concrete evidence of Nvidia's long-rumored upcoming SoC.

Word is that Nvidia has been developing an Arm-based processor for the past few years. Despite a teaser released last year in collaboration with Dell and the January unveiling of Nvidia's Arm workstation, a consumer-level product has yet to appear. However, a recent Geekbench score suggests that whatever Nvidia is working on might rival some of the best mobile processors currently available.

Reports that Nvidia is developing an Arm processor to rival Apple's M series and Qualcomm's Snapdragon X first emerged in 2023. Last year, Nvidia and Dell advised curious observers to stay tuned for updates, and Nvidia unveiled a $3,000 Arm-based mini workstation at this year's CES in January.

Last month, additional reports indicated that Nvidia and MediaTek are co-developing a pair of SoCs, codenamed N1 and N1X. More recently, United Daily News reported that Nvidia and Dell-owned Alienware are collaborating on an Arm-based laptop APU with GPU performance resembling that of a mobile RTX 4070.

This project is likely a downscaled version of Project Digits, the high-performance mini PC Nvidia introduced in January. That machine features MediaTek's 20-core GB10 CPU, 128 GB of RAM, a 4 TB SSD, and a 1-petaflop Nvidia Blackwell graphics chip.

In contrast, the "N1x" that appeared on Geekbench reported one core with 20 threads, though it may actually have 20 cores, depending on how the Geekbench software interprets the in-development hardware. Other reported specs include an HP 8EA3 motherboard, a 2.81 GHz base clock frequency, and 119 GB of RAM – likely rounded down from 128 GB. While this test was run on Ubuntu 24.04.1, the final product will likely ship with Windows on Arm to compete directly with Qualcomm.

Although comparing results from different operating systems and instruction sets is never straightforward, Nvidia's chip seems capable of trading blows with recent high-end mobile SoCs. The N1x clearly trounces Qualcomm's Snapdragon X Elite and broadly resembles Intel's Core Ultra 200HX family, Apple M3, and AMD Ryzen AI Max.

Arm, for its part, has promised that other vendors will soon begin manufacturing chips based on its architecture, which is more efficient than x86. While most software has traditionally been designed for x86, Arm developer support is improving rapidly.

The company recently stated that Windows on Arm users now spend 90 percent of their time in native apps – suggesting that the most critical applications have already been ported and no longer rely on emulation.

Permalink to story:

Nvidia Arm chip surfaces with strong Geekbench scores, could rival top Intel and AMD laptop CPUs

 

[Squid Surprise]

"The company recently stated that Windows on Arm users now spend 90 percent of their time in native apps – suggesting that the most critical applications have already been ported and no longer rely on emulation."

Not really relevant. Those ALREADY on Arm will obviously have most of their apps running natively - otherwise they'd have bought an x86 system...

The real stat we need is the percentage of ALL Windows apps that have a native Arm app. Once that is over 95%, we have something.

And yes - that includes games.

 

[Liranan]

"The company recently stated that Windows on Arm users now spend 90 percent of their time in native apps – suggesting that the most critical applications have already been ported and no longer rely on emulation."

Not really relevant. Those ALREADY on Arm will obviously have most of their apps running natively - otherwise they'd have bought an x86 system...

The real stat we need is the percentage of ALL Windows apps that have a native Arm app. Once that is over 95%, we have something.

And yes - that includes games.

Totally correct. Windows on ARM has a return rate of 90%, you have to love novelty and expensive toys to buy such a laptop over x86.

 

[HardReset]

Arm, for its part, has promised that other vendors will soon begin manufacturing chips based on its architecture, which is more efficient than x86. While most software has traditionally been designed for x86, Arm developer support is improving rapidly.

Jim Keller disagrees. Keller is btw one of few people that actually developed ARM and x86 chip at same time so better not to tell every time this ARM propaganda BS.

 

[SRB]

I don't see this going anywhere. 20 cores, Blackwell, NPU + Nvidia means this will make the AI Max+ 395 look inexpensive by comparison, yet not out perform it. The 90% is covered by the browser, email, office and social media apps, a pretty low bar. That bar also explains why the vast majority of laptops sell between $400 and $1200. Surface Pro 12 core laptops started at $1900 with keyboard, and Qualcomm's 12 core will get pasted by the Nvida chip, as well as the AI Max and Max +.

To be fair, the Qualcomm will get batter battery life, but I doubt the Blackwell Igpu will sip power noticeably better than the 8060s

What your left with is Qualcomm and Nvidia are dying to get a piece of the PC market with products that at best offer better battery life with prices out of the mainstream.

 

[AnilD]

A laptop APU that could soon have RTX 4070-level graphics baked in? Sounds too good to be true (this year). We’ve gone from “can it run Crysis?” to “can it run Crysis natively on ARM?” in just over a decade.

Between this, Snapdragon X Elite, and Apple’s M series, the x86 monopoly on high-performance laptops is looking shakier by the month.

 

[GeoffreyA]

Jim Keller disagrees. Keller is btw one of few people that actually developed ARM and x86 chip at same time so better not to tell every time this ARM propaganda BS.

An important difference is that ARM instructions are fixed length and x86 variable length. Traditionally, this made decoding of x86 difficult because, in predecode, the length has to be first worked out, and decoding always took lots of energy in x86 (at least until Sandy Bridge's micro-op cache).

Keller noted in his Anandtech interview that over the years, they had found ways to mitigate or solve the problem and that decoding was only a smart part of the die.

Whilst Keller is right, the variable-length instructions might be constraining how wide the decoders and backend grow. Nonetheless, ARM's decoders also use power; they too have got a micro-op cache. If the industry must move, an open-source ISA like RISC-V is more to everyone's benefit than ARM. But, until there's a Windows on RISC-V, that's going to be a dream.

 

[godrilla]

I don't see this going anywhere. 20 cores, Blackwell, NPU + Nvidia means this will make the AI Max+ 395 look inexpensive by comparison, yet not out perform it. The 90% is covered by the browser, email, office and social media apps, a pretty low bar. That bar also explains why the vast majority of laptops sell between $400 and $1200. Surface Pro 12 core laptops started at $1900 with keyboard, and Qualcomm's 12 core will get pasted by the Nvida chip, as well as the AI Max and Max +.

To be fair, the Qualcomm will get batter battery life, but I doubt the Blackwell Igpu will sip power noticeably better than the 8060s

What your left with is Qualcomm and Nvidia are dying to get a piece of the PC market with products that at best offer better battery life with prices out of the mainstream.

Between Nvidia and Qualcomm. I believe that Nvidia ( software with hardware centric company) with throw as much money as it can to improve the software to establish another monopoly. The silver lining is that X86 can't stay stagnated and needs to innovate just by the threat of arm chipping away at their market share.

 

[HardReset]

An important difference is that ARM instructions are fixed length and x86 variable length. Traditionally, this made decoding of x86 difficult because, in predecode, the length has to be first worked out, and decoding always took lots of energy in x86 (at least until Sandy Bridge's micro-op cache).

Keller noted in his Anandtech interview that over the years, they had found ways to mitigate or solve the problem and that decoding was only a smart part of the die.

Whilst Keller is right, the variable-length instructions might be constraining how wide the decoders and backend grow. Nonetheless, ARM's decoders also use power; they too have got a micro-op cache. If the industry must move, an open-source ISA like RISC-V is more to everyone's benefit than ARM. But, until there's a Windows on RISC-V, that's going to be a dream.

That is, when x86 and I mean literally x86 is used. x86-64 basically got rid of x86 legacy on floating point (compilers only make SSE2 FPU code, although x87 still works). Then we have AVX and probably will have even more new instruction sets that fall under x86 although not having anything to do with x86 and not even being x86 compatible.

Decoding is already pretty small problem and when new instruction sets gain more popularity, even more instructions will be microcode emulated because they are so rarely used.

That "width" will probably remain problem with x86 code but again most new code is not, or at least it should not be, something that has basically nothing to do with x86. Perhaps problem will shift on maintaining backwards compatibility and that is something RISC-V shines. No legacy = no power consumption for maintaining legacy. Like you said, RISC-V Windows is, well, dream

 

[Dreadcthulhu]

A laptop APU that could soon have RTX 4070-level graphics baked in? Sounds too good to be true (this year). We’ve gone from “can it run Crysis?” to “can it run Crysis natively on ARM?” in just over a decade.

Between this, Snapdragon X Elite, and Apple’s M series, the x86 monopoly on high-performance laptops is looking shakier by the month.

We already have laptop APUs with (mobile) RTX 4070 level graphics - the Ryzen 395 Max chips have iGPUs that perform in that ballpark. Making an SoC with a bigger iGPU portion & a wider memory interface is not really all that difficult at this point for the major chip makers, they just need to actually make such chips.

 

[CapNemo72]

This is aimed at AI workstation, not gaming or average office usage.
And it will be priced accordingly.

Software support for these tasks either already exists or will be ready at the launch.

But this is not some kind of x86 killer. Its just very focused product for its market.

 

[Eldrach]

I've been selling enterprise computers for over 7 years - and every second year they predict "the rise of ARM for Windows" - But I still have 0 demand for them from my global clients. There's just too much legacy software out there that doesn't work well with ARM

 

[SRB]

I've been selling enterprise computers for over 7 years - and every second year they predict "the rise of ARM for Windows" - But I still have 0 demand for them from my global clients. There's just too much legacy software out there that doesn't work well with ARM

It doesn't have to be legacy. There is a lot of corporate software and job/industry specific software such that it might be difficult and/or not enough of a user base to develop for ARM on Windows. This applies to drivers for a good deal of mainstream hardware. Thre will be less than zero specialized drivers written for WART (Windows on ARM RT).