Nvidia RTX Spark CPU is now official: "superchip" will power Windows laptops and desktops

[Megalomaniac]

Thank Goodness I invested in Chips during the pandemic!

Nvidia is going to blow Intel and AMD out of the water.

10 years from now, my next computer may be an Nvidia CPU and Nvidia GPU.

 

[Eldrach]

Guess if this goes on - Lisuan will be our only option at some point...we just have to get people to smuggle it out of China

 

[SRB]

I won't even bother to read the rest of the thread as I know anyway what the direction will be. "Boo AI, boo Nvidia... boo... just boo! Because boo! All is bad!!111"

Always happy to see it, sounds good if Nvidia is involved. Maybe we can finally run an AI locally on the GPU soon, which can then help with certain tasks on the PC. Exactly what I want. Something like a Claw integrated into Windows would be awesome.

I'm running some AI programs now, locally, when I want to, on my GPU. And somehow they work WITHOUT Nvidia's hardware....I don't need Microsoft's and Nvidia's quest for ARM pc's.
It would be different if they worked with all my software and ran faster than my current hardware for less money, but they don't

 

[AnilD]

Nvidia somehow convinced the entire industry that the future of computing is buying a GPU and then attaching however much CPU is required to keep feeding it data. I'm not even judging whether that's right or wrong on the merits, just observing how remarkable the shift has been.

 

[ZedRM]

Cool, something else I won't buy.

I'll second this!

 

[Stoly]

Talk about a long time in coming. Not only is RTX Spark a year late, its been like a decade since nvidia was supposed to enter the pc platform space with one of the first iterations of Windows ARM.

We already know the chip, it's the same one on the DGX Spark, only for windows this time. The cpu is Grace architecture with 20 cores, apparently no performance/efficiency bull, while the GPU is a RTX5070 mobile level, pretty fancy.

If performance falls in the DGX ballpark, then Apple has nothing to worry about, but lets watch the performance numbers before the final judgement.

On a side note, no mention of DLSS5... interesting...

 

[Stoly]

(and it would be nice if sleep mode works properly for a change).

I've been waiting for decades for MS to fix sleep mode, never gonna happen...

 

[Luay]

This isn’t a chip launch. It’s CUDA arriving on consumer Windows, and TensorRT wired into the creative stack at the API level.

The next time AMD or Qualcomm want to compete, it won’t be on chip benchmarks, it’ll be on whether Photoshop still works. That is a different game that gets won quietly.

The hustle has matured. G-Sync and RTX IO were the crude colonizer version claiming technological "superiority", a proprietary connector, branded wrapping, the lock-in visible and labeled, but Nvidia got “Mandela’d”

Agentic Windows is the maturer colonizer version. It does not announce itself as superior or even proprietary; it makes itself the ground the software is built on. This time it is the technologically required, I.e., the promised “biblical” stack. By the time AMD or Qualcomm argue that the stack should not be the stack, Photoshop and Premiere have already shipped on TensorRT and unified memory, and by then it is too late.

 

[Biostud]

With Windows on Arm already being a major success, what could make it even better?
Add an AI agent to run the computer.

Yiiihaaaa....

(Is it locked down as Apple or can you install Linux on it?)

 

[Luay]

With Windows on Arm already being a major success, what could make it even better?
Add an AI agent to run the computer.

Yiiihaaaa....

(Is it locked down as Apple or can you install Linux on it?)

By the time we find out, the answer is usually already decided, and it will be too late.

 

[Julio Franco]

Added Ryan's tweet in the conclusion (Q&A video) asking Jensen a very good question:
"With the PC market being low margin and cut throat, why enter it now?"

https://twitter.com/x/status/2061816736308244548

 

[HardReset]

Transistor count is almost double what AMD Ryzen AI Max+ 395 has. Considering that, price tag will be huge. Of course, AMD already announced Ryzen AI max 400 series.

And what is so "genuinely impressive" here? AMD already made similar solution. This is basically just bigger version of AMD version and AMD even offers 192 GB memory instead of just 128 GB. For large AI models, this is already obsolete.

 

[jellowiggler]

I'm all for having good mid level graphics built on a chip that also has lots of cores for low and high power work loads. Moving forward I see the common high speed memory pool to be the way forward. Today that means at least a 32GB RAM pool. This is way more flexible that a basic PC's 16GB+8GB of segmented memory today. Take 6-8GB away for the OS running and you have 8+8GB. No good. Common ram pools gives you so much more ability to shuffle things around for previously constrained software.

I'm still waiting for any good reason to have AI tops be a factor in my local PC. But integrated flexibility is of interest. Just by an integrated PC with 64GB of high speed RAM and don't worry about having enough VRAM for graphics or enough space for your video editing, etc... I can see the benefit of that. Also no more incompatibility issues with RAM, or matching RAM sets, etc if it's already contained and built to run at X speed on the chip.

I'm not sure about adoption though. If Strix Halo was too expensive and had no real push, how will this? The normal answer would have been they will price it to grab market share. But if nvidia is so non-price conscious as they are in all of the in demand segments, why bother? Why sell a pro-sumer product if all the production is in big dollar DGX stations? Unless they aren't selling, or it's all just a big paper launch to say they have 'shipped' into another segment but actually are providing a few thousand units a quarter.

 

[Northy]

I actually like using my computer, that's what makes using my computer good. Telling some bot to use my computer isn't what I want. Even when I'm doing creative stuff I don't want a bot to do it for me, it's literally the last thing I want.

I'm just not seeing the way forward here, it's like the Metaverse to me. Something no one asked for and no one wanted.

 

[BadThad]

And so it starts, the PC to AI Agent model. It could very well take hold.

 

[LemmingO]

Added Ryan's tweet in the conclusion (Q&A video) asking Jensen a very good question:
"With the PC market being low margin and cut throat, why enter it now?"

https://twitter.com/x/status/2061816736308244548

The question is willfully ignorant. This isn't about "entering the PC market". It's about reinventing the PC Market with slave devices that prop up Nvidia's strategy and are driven by a distinct revenue model.

Jensen does as Jensen wants, as he always has. The answer to the question paints a picture of AI Bubble 1.1. Nvidia doesn't need to think about traditional revenue models and margins because they can afford not to while playing the mid-term game: create an ecosystem of AI-enabled devices that will feed the AI bubble further and push Nvidia's market cap.

To be fair - Jensen could've said anything as an answer, but he did say the truth.

You can dress it up any way you want, but AI hasn't really translated into value for business, it's just translated in market stratification. The "Haves" and "Have nots". If your business isn't neck deep into AI then you don't stand a chance because investment funds will chase the pot of gold at the end of the AI Rainbow. Businesses will buy the Nvidia-powered devices because it will promise (and sometimes deliver) on AI, and that will check the box for an "AI strategy".

I honestly hope that someone breaks the foundation model that drives current AI hardware.

PS: AMD is no better than Nvidia, when it comes to this, so anyone setting their sights on Lisa Su as the second coming better think twice.

 

[Megalomaniac]

I signed up for AI trading with Wells Fargo ($500) and Charles Schwab ($5000). Both are turning a profit. That's what I expect AI to do for me.

 

[umeng2002]

I wonder how many slurs and illegal sentences I can type into one of these AI terminals before I mysteriously get arrested for Jay Walking. Because you can do all that on Libre Office without it automatically reporting you to the government.

 

[umeng2002]

I signed up for AI trading with Wells Fargo ($500) and Charles Schwab ($5000). Both are turning a profit. That's what I expect AI to do for me.

An Index Fund turns a profit.

 

[poohbear]

Unfortunately the future seems to be "AI PC's" instead of personal computers, Jensen has said it himself. Once you look past the marketing jargon, we're seeing it happen already, hardware is moving into unaffordable territory as consumers are being pushed to rent a PC, or with these Nvidia SoC's, they will probably need a subscription in the cloud to have full functionality. These laptops with Nvidia SoC's will probably be priced at a premium too, and will be for the wealthy AI tech bros, not the average gamer.
It is also interesting to see Microsoft partner up with Nvidia to push for AI even harder, despite their promise to lessen the amount of AI being forced, so much for that claim.

It's already happening with their GPUs and cloud gaming. An 8 year GeForce NOW Ultimate subscription is still cheaper than a RTX 5080 (which is the GPU provided on the Ultimate membership). Yes... 8 years... and that's JUST for the RTX 5080 (not the rest of the PC)! With any high-speed Fibre optic connection or Wifi 7 speeds & latency... it's much cheaper & dare I say better than buying a full PC with a RTX 5080.

 

[Burty117]

It's already happening with their GPUs and cloud gaming. An 8 year GeForce NOW Ultimate subscription is still cheaper than a RTX 5080 (which is the GPU provided on the Ultimate membership). Yes... 8 years... and that's JUST for the RTX 5080 (not the rest of the PC)!

GeForce NOW Ultimate is £200 a year x 8 years = £1600
5080 pricing is around £1,100

So the math is wrong, but you don't seem to value the ability to keep the GPU forever, if you stop paying that subscription, you lose access, if you've purchased the GPU outright, you keep it forever, plus the other benefits like better image quality, lower latency, re-sale value etc...

 

[Aolish]

This sounds nice, but the real deal here is how well can they implement this. As much as how this is exciting, the real question is how close can they get things in par with how Apple has done things. Their M-series macs are a hit. Rosetta emulation is so good for non-native apps, you get INSANE performance from M-series even to the point of beating out many windows laptops at twice the price with WAAAAYYYY battery life. Its a tall order even for Nvidia which doesn't even include trying to convince people to develop apps for ARM on windows. We still have a long ways to go when it comes to Windows on ARM.