Aardvark Daily

Please visit the sponsor!

Has NVIDIA just killed AMD and Intel?

Computex is underway in Taipei and although the rise of AI has meant that there have been very few "exciting" announcements for gamers and the rest, NVIDA has unveiled something that could really shake up the industry.

The next big thing could be NVIDIA's N1X high integration APU.

Typically, APUs consist of a CPU and a relatively wimpy GPU on a single die. They are most often used in laptop or low-end desktop systems where low power and low price are a priority over performance.

I'm using an APU (AMD Ryzen 5600G) in the desktop system on which I'm typing this now. The six processor cores and inbuilt GPU provide more than enough horsepower to handle simple tasks such as web-browsing, watching videos and word processing. To use anything more powerful would be a waste of energy and capital.

NVIDIA's new APU however, could pose a very real threat to AMD and Intel and the concept of the "wimpy" iGPUs that an APU contains.

The N1X chip is being billed as an RTX5070-class GPU integrated with a 20 core GPU and that has the potential to do a lot of processing in a very short space of time.

The N1X offers to address up to 128GB of unified memory that can be shared by both the CPUs and the GPU. This has the potential to obliterate the peformance bottleneck normally created by way of the PCIE bus that normal GPUs connect through, thus significantly improving overall system performance.

However, isn't this exactly the way that Apple's M-series of chips work?

Yes it is.

Apple integrates a powerful GPU with a multi-core ARM-based processor on a single die, with unified memory.

So has NVIDIA really invented something new?

Not really -- but what they've done is bought CUDA to the party, something that's going to be useful to many users who rely on CUDA-based software. Applications such as 3D rendering and AI will certainly benefit from NVIDIA's silicon when compared to Apple's M-series.

However, despite the claims of NVIDIA, it seems that the N1X isn't quite the CPU powerhouse they would lead us to believe. Instead of the latest generation of C1-Ultra ARM cores, the N1X has opted for older Cortex-X925 ones. This means that in terms of raw CPU performance, Apple's latest M-series will eat NVIDIA's lunch. However, it's clear that NVIDIA is prepared to make that trade-off in return for better power-efficiency and because it knows that the RTX5070-level GPU performance will more than make up for that in many of the target applications.

Jensen Huang is touting this chip as "the" platform for the next generation of agentic PCs, where the ability to run huge LLM AI models locally will be crucial.

In a move that's likely to anger rather than placate the gaming community, Jensen has also been touting its performance as the cornerstone of a new generation of gaming laptops that will deliver 1440p graphics at high framerates. How many *real* gamers want to spend their time gaming on a laptop I wonder?

As for software, NVIDIA is claiming that they're working with Microsoft and all current Windows apps will run on the new chip -- albeit many of them only in "emulation mode", something that imposes another 15-20 percent performance penalty.

What's the bottom line?

Well the N1X is not a new concept, Apple have been putting multiple ARM cores, unified memory and a pretty decent GPU on a single bit of silicon for years now.

The N1X will be significantly more energy efficient than a traditional X86 CPU with discrete GPU setup and it will offer comparable levels of performance in many applications.

For AI-based applications, which may soon be something *forced* upon us all, the N1X will likely be unmatched by anything Apple currently offers and perhaps even a high-end x86 system with separate GPU. The performance win goes to the N1X because of the incredibly high memory bandwidth the unified approach offers, something really important when dealing with the massive matricies that LLMs rely on.

NVIDIA claim this silicon will be used in "premium" products -- a hint that it will not be cheap. When you look at the price of an RTX5070 GPU I would expect that a top-spec NX1 chip will come in at over US1K so once it's integrated into a high-spec laptop with lots of RAM we're talking US$4K or so. Entry-level pricing may be less when using versions of the N1X (just called the N1) with fewer cores and systems that have less RAM.

It's worth noting that both AMD and Intel share prices took a dip in the wake of the NVIDIA annoucement. Clearly the market is worried that this new silicon could make quite dent in the revenues and profits of those companies.

Only time will tell.

Carpe Diem folks!

Please visit the sponsor!

Here is a PERMANENT link to this column

Rank This Aardvark Page

Change Font