https://nvidianews.nvidia.com/news/...ter-ai-supercomputer-on-every-enterprise-desk

https://www.nvidia.com/en-us/products/workstations/dgx-station/

 

Well I suppose price will scale pretty quickly as we've seen with apple, and might not even matter, as this seems to be aimed at enterprise customers. I figure lightweight, energy efficient laptops like these, that they claim can run a 100b parameter model locally with decent token fetching speeds (or whatever the terminology) are going to be a high value proposition for enterprise customers, where implementing AI is already happening, but data protection is a major concern. Windows is the default system for many of these massive players like govs, healthcare, finance, education etc, so maybe if these do well with the bleeding edge customers they could be looking at some massive orders with the next iteration.

Everyone I know that works in an office uses a windows machine and all of them have been incentivised to use Copilot for all kinds of work tasks. Local Copilot and a large, industry tuneable models, could offload a lot of server costs for ms and nv, provide enterprise customers a lot of flexibility, but still keep ms/nv the primary eco system. Offloading trivial server requests could allow ms/nv to allocate more time for bleeding edge training thus accelerating the development, which is something we all know is their main aim.

I suppose if this scales quickly in the enterprise market, then the unwashed masses like us, might get products that are more cost effective eventually. I guess that's what zalbardar meant that I missed.

Tbh, as a musician who can play, but who is also deep into music tech as well. I've had enough compute power to achieve pro results from day one, which was decades ago.

 

High end stuff does not get cheaper due to economy of scale. More powerful stuff can get cheaper because of mass production, But your direction is right, if not the doubling of the price. They definitely aren't going to be any cheaper than the tech you are mentioning the current prices of.

 

I guess these are also, besides being expensive, gonna be as closed as a Samsung phone with a 5y lifespan. You won't be able to change anything and be happy with two USB ports (4, but anyway...) and a power button. Me, I am saving for a Framework laptop.

Nvidia also has one other CPU in the making: the NV1 and NV1x... They are entering the CPU market, but it's mostly about getting as much money as they can from big companies and server market. I'm not really thrilled as I don't expect anything good to come from Nvidia, or anybody else for that matter anymore. Times are... "interesting". Very. Especially for the not-so-wealthy kind.

 

so it's just rebranded Project Digits from January 2025 which was later renamed to DGX Spark intended for developers while being subpar for local AI inference, and now 2 years later (estimated launch Q4 2026 at best) it will officially run Windows.... that's all or did I miss something?

 

wow, a PC with unified memory. what a amazing idea.

- a mac user (laughs)

 

A wave of layoffs is sweeping across the American tech sector... April 8, 2026... NVIDIA builds the hardware on which AI runs... Meta, too, is laying off around 8,000 employees and eliminating 6,000 open positions due to high AI costs.

 

WHEW! Now THAT is what I call superHype. I will wait until they develop a model that also fixes good coffee (on command) and has at least two spigots, one for red wine the other for white wine - complete with a menu that allows user to choose which of each....maybe a connection (optional, of course) to the next generation Claude that recommends what you should drink with your meal.
Maybe we should open a thread about "best wines for working with audio tech".

 

Dom PérAIgnon
Fuck, that was too bad of a joke even for my low standards... sry guys, I'll see myself out

 

You can't quit, you're fired....

 

so, after bit of research, it's indeed the DGX Spark in a laptop form factor, minus the Mellanox Connext-X7 infiniband 200Gbe network and lacking other connectivity maybe too,
Ubuntu DGX OS swapped with officially supported Windows 11 for ARM, that's the selling point I guess,
performance-wise, you can expect underpowered RTX5070 (so 1080p gaming perhaps),
price-wise, I doubt to expect prices lower than aforementioned DGX Spark mini PC which costs around 5000€

 

@tzzsmk I was reading that most of the claimed improvement in local inference and general performance has come from all the direct partnerships with all the app/game developers, how they're implementing the ai processes (over my head) and a big jump in optimisation of windows on arm by ms. Also the close work with the OEM partners for the laptop hardware optimisation. I read that they've solved most of the issues they were having between their gpu and the mediatek cpu.

Also did you see any of the stuff they were saying about sandboxing agents locally, user defined permissions, encrypted/spoofed calls to the cloud and so on? That seemed to be relevant to the enterprise stuff I was speculating about. Do you think all of that in combination and stuff we don't know about could have bumped performance significantly over the gtx?

I think when Nvidia are making the whole chip themselves we could see Apple M like innovation for windows systems if my understanding of all this is correct. Excuse my ignorance on the topic just very interested and was waiting on your take, especially on the local models and agents side.