Nvidia N1X leak points to limited 2026 availability

[Redaktion]

Moore's Law is Dead has revealed Nvidia plans to show off its N1X laptop CPU at Computex this year. However, one might have to wait a while before they can buy a laptop powered by the chip.

https://www.notebookcheck.net/Nvidia-N1X-leak-points-to-limited-2026-availability.1282855.0.html

[opckieran]

There's NO WAY this performs close to a desktop 5060 with essentially half the power budget.

[._.]

There's NO WAY this performs close to a desktop 5060 with essentially half the power budget.

You do realize this has 6144 CUDA Cores compared to a desktop 5060's 3840, right?

Not that any of this matters as unfortunately it's looking like this is a flop. :/

[Huang]

Nvidia is clearly not targeting the hardcore standalone RTX5070 GPU desktop gamers, Jensen is going for the laptop users, and probably gamers AI developers who are also in the Mac ecosystem.

[yee]

There's NO WAY this performs close to a desktop 5060 with essentially half the power budget.

The chip will be support up to 128 GB of LPDDR5x-8533 RAM and be manufactured on an unspecified 3 nm node from TSMC.

Gaming performance is based on memory bandwidth, let's calculate it:
N1X: 273 GB/s = 256-bit * 8533 MT/s / 8 / 1000.

Strix Halo: 256 GB/s = 256-bit * 8000 MT/s / 8 / 1000.

Basically the N1X falls into the 256-bit APU (e.g. Strix Halo (=4060 Mobile/Laptop performance)) category.

the same as the DGX Spark

Another confirmation. en.wikipedia.org/wiki/Blackwell_(microarchitecture): same GB10 chip.

3dmark.com/search - Steel Nomad:
Strix Halo' Radeon 8060S iGPU: Average score: 2031
RTX 4060 (notebook): Average score: 2262

Memory bandwidth of a 4060 Laptop is 224 - 256 GB/s (aligns perfectly with Strix Halo' gaming performance).
Memory bandwidth of a 5060 (Ti) desktop is 448 GB/s (64% more than N1X).
Memory bandwidth of a 5050/5060/5070 Laptop is 384 GB/s.
(en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units)

Based on the memory bandwidth alone, I'd agree with you. But maybe due to it being based on TSMC' 3nm node variant, a bit more of performance can be squeezed out.

But between "between an RTX 5070 and RTX 5070 Ti laptop", would be very surprising, if N1X' memory bandwidth is at 273 GB/s.

[noo]

Gaming performance is based on memory bandwidth

Only at higher resolution gaming at 4k and beyond. You don't need much for lower resolutions (e.g. 720p-1080p)

Blackwell is slightly more efficient at data compression than RDNA3.5 too.

Basically the N1X falls into the 256-bit APU (e.g. Strix Halo (=4060 Mobile/Laptop performance)) category.

Yeh, because games aren't compiled for arm and there is fare bit CPU overhead.

If a game was ever released for arm, it's going to be faster.

Wendell from level1tech has tested all these systems (DGX Spark GB10, Halo) and said pretty much the same.

Maybe laptop implementation will be slightly slower because less power. DGX Spark can use over 100+ watts.

[indeed]

There's NO WAY this performs close to a desktop 5060 with essentially half the power budget.

The chip will be support up to 128 GB of LPDDR5x-8533 RAM and be manufactured on an unspecified 3 nm node from TSMC.

Gaming performance is based on memory bandwidth, let's calculate it:
N1X: 273 GB/s = 256-bit * 8533 MT/s / 8 / 1000.

Strix Halo: 256 GB/s = 256-bit * 8000 MT/s / 8 / 1000.

Basically the N1X falls into the 256-bit APU (e.g. Strix Halo (=4060 Mobile/Laptop performance)) category.

the same as the DGX Spark

Another confirmation. en.wikipedia.org/wiki/Blackwell_(microarchitecture): same GB10 chip.

3dmark.com/search - Steel Nomad:
Strix Halo' Radeon 8060S iGPU: Average score: 2031
RTX 4060 (notebook): Average score: 2262

Memory bandwidth of a 4060 Laptop is 224 - 256 GB/s (aligns perfectly with Strix Halo' gaming performance).
Memory bandwidth of a 5060 (Ti) desktop is 448 GB/s (64% more than N1X).
Memory bandwidth of a 5050/5060/5070 Laptop is 384 GB/s.
(en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units)

Based on the memory bandwidth alone, I'd agree with you. But maybe due to it being based on TSMC' 3nm node variant, a bit more of performance can be squeezed out.

But between "between an RTX 5070 and RTX 5070 Ti laptop", would be very surprising, if N1X' memory bandwidth is at 273 GB/s.

Just to make it clear why memory bandwidth is so important:
While the AI / LLM token generation (tg) (the generation of text) depends on the memory bandwidth, the prompt processing (pp) (the processing of input text) depends on the compute of the GPU (basically, the GPU's 3DMark score / FPS performance), but this, again, depends on the memory bandwidth (as you can see in the quote above).

That doesn't mean that a workstation GPU, like a RTX PRO 6000 (96 GB VRAM) (same GB202 chip as a 5090 (32 GB VRAM)), can't have 3 times the VRAM memory, which it has, but it also has 1.8 TB/s of memory bandwidth (thanks to its 512-bit chip and using the, much faster per bit, GDDR7 VRAM chips).
If this N1X or Strix Halo had a bigger variant, using the same 512-bit memory bus width, its memory bandwidth would only be:
546 GB/s = 256-bit * 8533 MT/s / 8 / 1000.
(again, a RTX 5090, using the same 512-bit wide memory bus, has 1.8 TB/s)

This APU, on the other hand, uses LPDDR5X memory, so while it can have up to 128 GB RAM (not VRAM, that uses GDDR6 or GDDR7), its system memory bandwidth is still rather (very) slow at 273 GB/s.

As such, depending on how big your input text size is, 273 GB/s, or a 4060 Laptop of performance that this memory bandwidth allows for, may take a while to process the amount of text.

On reddit.com/r/localllama many users don't seem very happy with their Strix Halo' (=256 GB/s memory bandwidth) prompt processing speed in e.g. agentic workloads.

What you instead may want to build for AI / LLMs is a simple desktop gaming PC with as fast of a GPU and as of VRAM that you want to afford. If AI / LLMs are a priority, look at the workstation cards (RTX PRO 4000, PRO 4500, PRO 5000 and the mentioned PRO 6000, tho they are expensive), because they have double or more VRAM (see given example of a 5090 (32GB VRAM) vs PRO 6000 (96GB VRAM), but it's the same GB202 chip (see en.wikipedia.org/wiki/Blackwell_(microarchitecture))), than the desktop mainstream gaming GPUs using the same GPU chips.

[wrong post]

nevermind, wrong topic

[yee]

RTX Spark is reviewed (notebookcheck.net/Nvidia-RTX-Spark-Analysis-Finally-an-answer-to-Apple-s-M5-Pro-M5-Max.1422820.0.html) and we can compare now:

There's NO WAY this performs close to a desktop 5060 with essentially half the power budget.

[..]
Based on the memory bandwidth alone, I'd agree with you. But maybe due to it being based on TSMC' 3nm node variant, a bit more of performance can be squeezed out.

But between "between an RTX 5070 and RTX 5070 Ti laptop", would be very surprising, if N1X' memory bandwidth is at 273 GB/s.

3dmark.com/search: Steel Nomad:

• Strix Halo Radeon 8060S iGPU (256 GB/s): "Average score: 2032" (= 7.94 points per GB/s).
• RTX Spark N1X 5120-core iGPU (301 GB/s): "Average score: 2764" (= 9.18 points per GB/s).
• RTX Spark N1X 6144-core iGPU (301 GB/s) (45-80 Watt): "Average score: 2941" (= 9.77 (+37.6%) points per GB/s) and "3134 Points" according to .1422820.0.html (= 10.4 (+46.5%) points per GB/s) (= 39 points/Watt).
• desktop 5060 (448 GB/s) (145 Watt TDP): "Average score: 3179" (= 7.1 points per GB/s) (= 21.9 points/Watt).
• 5070 Laptop (384 GB/s) (50-100 Watt TDP): "Average score: 3009" (= 7.84 points per GB/s).
• 5070 Ti Laptop (672 GB/s): "Average score: 3864" (= 5.75 points per GB/s).

(en.wikipedia.org/wiki/GeForce_RTX_50_series, en.wikipedia.org/wiki/Nvidia_RTX_Spark)

In terms of pure performance, not performance per Watt, the 5060 desktop is just 1.4% (3179/3134) to 8% (3179/2941) faster..pretty close, but not

between an RTX 5070 and RTX 5070 Ti laptop

as is stated in this news and more like on par with a 5070 Laptop.

And in terms of performance per Watt:
39 points/Watt/21.9 points/Watt = +78% (the TDP value was used, and assuming 3dmark utilizes a GPU more or less to its fullest, this should be good enough to being correct). An increase of 78% is excellent, but not unsurprising, because from power scaling (= automatic undervolting) or manual undervolting it's known that the power efficiency increases when the TDP Watt value is reduced to about -50% (below -50% the power efficiency (aka FPS per Watt) worsens again).

The RTX Spark N1X has a maximum TDP of 80 Watts for mobile applications, while desktop systems can consume up to 120 Watts.

The TDP is already reduced from 120 to 80 Watt (-33%). If it's further reduced, the score/Watt may increase from 78% to 100%, but then the performance will decrease to above the +1.4% (to maybe -1x to -2x%).

Assuming those 5070 Laptop laptops that use 100W and have a score of ~3100 (average is "3009", "median: 3030" (.934942.0.html)), then it scores 31 points/Watt (3100points/100W). This leads to the Spark achieving 25.8% (39/31) more FPS per Watt, despite the 5070 Laptop being already an undervolted GB206-300 chip (that is also used in a 5060 Ti (180 W) btw.).

All in all and if no mistakes were made, a good, full node/one generational, improvement (= +25-33% perf/Watt) in raw compute. The one full node jump to TSMC's 3 nm (does probably the most lifting) + 273 -> 301 GB/s (+10%) did it indeed.

The only issue is the price and while 25% more FPS per Watt is a good minimum generational improvement, it will never amortize itself: +25% more power efficient means you save approx. 15-30 bucks per year, depending on where you live, if used for 8 hours every day (15 = 0.025kW (25W=100W-25%)*8hours per day*365days*0.2kWh).

[yee]

Nevermind, at least for the non-native Cyberpunk x64 version, the FPS per Watt did not improve (see the tables)..:

The numbers are still solid and the N1X is just slightly worse compared to systems with the GeForce RTX 5060/5070.