This news from Seeking Alpha -

NVIDIA on Monday (1 june) unveiled NVIDIA RTX Spark, a new superchip that reinvents Windows PCs for
the era of personal AI agents, featuring 1 petaflop of AI performance, power efficiency, full-stack NVIDIA AI and graphics technology, and up to 128 GB of unified memory.

Starting this fall, Nvidia’s new RTX Spark Superchip will debut in laptop and desktop computers from leading PC brands including Dell Technologies and Lenovo Group.

Microsoft will join Nvidia at two key industry conferences next week (I assume that means the week of 1 june, just started) to launch the first Windows computers that use Nvidia’s chips as the main processor, Axios reported on Saturday (30 may).

One wonders if these new speed demons are backwards compatible, or are we stuck waiting for audio
developers to re-do their software. Or maybe we can do just fine without this kind of speed??

update:
This added information from the italian on-line news site, Il Sole 24 Ore:

(first the english then the original) -

The product is a combintion of microprocessor and graphics chip, realized with the support of MediaTek in Taiwan...it will run Microsoft's Windows for Arm operating system.

Il prodotto è una combinazione di microprocessore e chip grafico, realizzata con il supporto della taiwanese MediaTek, e farà girare il sistema operativo Windows for Arm di Microsoft.

Again, do we have a compatibility issue here??

 

Last edited: Jun 1, 2026 at 3:25 PM

I'll check it out as a curious nerd. Will they have the RAM embedded on the main chip like Apple's?

But Nvidia APU + Microsoft OS = AIAAIAIAIAIAGENTSaiaiaiaiaiAGENTSAGENTSAGENTSaiaiaiaiAGENTS

So my hopes as a consumer... don't exist. Glad for the super wealthy who can afford this I guess (?)

 

Yes, we do. Since these will be ARM chips, they will run ARM apps at full speeds and rely on emulation for backwards x64 compatibility.

How well that will work, and at what speed, compared to price point. I will let others test and benchmark before I go anywhere near of those things.
They will probably be outside of any reasonable price range anyway.

After Nvidia did everything it possibly could to screw over the consumer market, and their last 2 generations of consumer graphics cards were majorly flops, it's hard to imagine them making anything good for the consumer market any more, other than to milk whatever trust remains with shoddy products.

 

Yep, that'll be interesting to watch. Microsoft can no longer code a freaking Notepad so funny times ahead of us

well, flops for the consumers, success in sells because there's no competition

that's a fact. NVIDIA is now a 99% AI-focused company because, well, money.

 

Yes. It's kinda ironic though that half a decade of developing for Apple M will give developers a head start in transitioning their Windows software to the ARM instruction set - if they're using architecture specific instructions at all.

Still, this is Microsoft and not Apple. MS doesn't have Apple's dictatorial Eisenfaust, so the transition is going to take much, much longer. Even more so because this is 2026 and who's going to make risky investments in computing equipment you might have to depend on for 7+ years with this economy?

Like I said a couple years ago, don't worry too much until 2030 rolls around.

Hum, my reading: RAM is unified, on-package & inside the SoC (like Apple M), with Nvidia's interlink connecting it to the CPU and GPU. Bandwidth looks to be 273 GB/s (so Apple M4 Pro, or M5 Pro if you round up).

 

i hope its not too expensive.. anyway

 

F*** AI, I want to buy one more audio friendly CPU and then I'm going offline.

 

I hate both Microsoft and Nvidia, but nonetheless a functional Windows ARM computer with a compatibility layer is a great thing to have. I don't think I care enough to have one, but the effect it has on laptops and SBCs are a step forward. I didn't have faith in the M1 chips but I was proved wrong, they're great computers for people who don't want to think about computers. Maybe if it gets really cheap eventually I might even get one for my living room.

 

I only know one system: VHS!
I only know one CPU: AMD Ryzen!
I only know two sound formats: WAV and MP3!

 

This is great.

New stuff means existing stuff gets sold off cheaper.

That is unless we get the same issue as with the RAM, where warehouses start destroying stock to keep the price up.

 

We do not need a ton of GPU for much more than stem separation and if you want to locally create AU generative or AI assisted music. Voice cloning, noise reduction and mastering assistants etc. These tend to work on chunks of audio rather than individual samples. We're not going to be waiting for anything more than AI-assistants in current audio software, not new versions of entirely new plugins. The biggest barrier to GPU dominated production will continue to be low-latency realtime performance in audio production.

 

A current desktop nvidia video card is how much - anywhere between 2K and 5K ($/€)? So how much the super chip is going to cost? 10-15K? Rather niche. Because if is cheaper than a current card then is not a super chip, isn't it?

 

Nah. For now, I'm sticking with what I have and waiting to see which of the new chip design approaches (the ones ASML, TSMC, and Intel are each working on) ends up prevailing, rather than throwing lots of money on a dead horse.

 

And since a "consumer" can't even grab a 4TB SSD now without taking out a 2nd mortgage on the trailer, I would be guessing this "power" would only be affordable to the corps & DC's with the MEGA-$'s.

 

Not too knowledgeable about gear in this spec range. Am I correct in thinking, that with talk of 128gb unified memory and being able to run 100B parameter models locally with multiple agents working with good power efficiency, these are targeting the £5k/$7k+ macbook pro m5 users?

(edit: ok nm I must have scrolled past xorome answering this question)

Sure that's how it used to work but look at the new and second hand market now...both cooked...and is that even a thing in the price range we're talking about where the people buying these likely write them off as a business expense.

 

Last edited: Jun 1, 2026 at 8:52 PM

I already watched a clip about it. It's gonna be expensive as heck, that's for sure.
I saw the first will be laptops with 128GB (LP on laptops) DDR5, and it should be pretty fast cuz' AI depends a lot on RAM bandwidth. That's insane. I guess there'll be models with less memory.

Well, you're right there, that's the only silver lining. Until now, the ARM based Qualcomm (?) with Windows have failed miserably. At least a push for Windows ARM development considering NVIDIA has a humongous amount of money and influence.

 

Once developers catch up to WoA this will be a godsend. MUCH better for DPC latency. Huge.

 

How so? To my knowledge WoA has the same queues for deferred calls from drivers. And the bit that is faster with this architecture (presumably, I've no details on this) is transferring the data to and from audio hardware is not really the problem on x86 either.

It's not like ARM is magically gonna amp up USB/TB (if TB is even supported, that is) controllers and such.

 

https://nvidianews.nvidia.com/news/nvidia-microsoft-windows-pcs-agents-rtx-spark

 

No PCIe hops—CPU/GPU share pool, zero-copy DMA to audio interfaces. WoA kernel: fresh ARM exception model + NPU-driven real-time scheduler slashes DPC queues vs x86 legacy HAL/drivers (ndis/wdf spikes). Mac ARM proves sub-200µs stable; Nvidia stack amplifies. Bus speed same, system overhead not.