what are your thoughts?
Did they increase the bandwidth to 600GB/s?
If these new laptops can run Linux then they would be very interesting. If not, I personally do not care.
For me its the other way round. Would love to use Windows 10 or MacOS X as OS and not Linux.
The big question is: Is the memory bandwidth higher? On their website, they also showcase a new desktop.
I wondered the same but I believe the RTX Spark’s 600GB/s is the aggregate C2C bandwidth, which is the same as DGX Spark. So basically the same C2C (~600 GB/s bidirectional) and memory bandwidth (~300 GB/s) as DGX Spark I think.
Sure seems to be the exact specs at the DGX spark (with less power draw) . Will be interesting to see if they’ve resolved any of the memory bandwidth issues we have, as they are advertising 1 petaflop of AI compute, when we cannot even get near this on our DGX Spark. Or are they expecting 4/ts to be useful on a laptop…
@eugr_nv I am hoping the DGX Spark II Spark II - Let's discuss what we would like to see in the next Spark version ups its game or this is going to massivy cannibalize the DGX Spark space.
From what is reported the RTX Spark is a slower but more portable version of the DGX Spark.
That power cap will not make it faster for certain.
It’s the same chip, same specs, same max 1PFLOP claimed sparse compute as advertised for DGX Spark - at a probably slightly lower TDP except when on wall power but top end models likely will hit similar numbers.
Minus CX-7, so no scaling. Unsure about DGX Linux support. It’s a very different product.
The just announced RTX Spark is DGX Spark’s little sister and to a competitor. It’s design for personal agents. Windows OS is modified to support the NVIDIA OpenShell. Lot’s of security issues with the OpenClaw. The interesting tidbit from NVIDIA and Microsoft Reinvent Windows PCs for the Age of Personal AI | NVIDIA Newsroom is that the notebooks will be available with up to 128G of RAM. Emphasis on up to!
And just with the Spark NVIDIA lists up to 1 petaFLOP of FP4 AI performance for the RTX Spark. Good luck achieving it!
Nvidia’s render on their page for the desktop looks to be an identical Motherboard design/layout as the GB10 just missing ConnectX
And the similarities for the OEM boxes also seem to show the same thing just slightly re-designed cases without Connectx, same layout otherwise.
MSI RTX Spark vs GB10
Asus GX10 vs RTX Spark
I’m not sure how much I believe it, but they’re reporting 600gb/sec memory bandwidth NVIDIA’s Enters The PC Space With RTX Spark, Offers Up To A 20-Core CPU, 128GB Of Unified Memory, 600GB/s Bandwidth To Supercharge AI Operations
If that’s true, then Spark DGX users got burned big time.
600GB/s is die C2C bandwidth, which seems the same as the GB10. Memory is claimed for 300GB/s which might be marginally faster than the 273 TO 300GB/s claimed on the GB10. But the TDP is lower. The silicon might have minor differences but it seems to be nearly identical.
The 600GB/s is C2C bandwith not memory bandwidth. The DGX spark also has 600gb/s C2C bandwidth. The RTX spark is supposed to have 300gb/s of memory bandwidth so its a bit higher than the DGX spark but not by very much.
You can practically fit the Spark in your pocket. How much do they need to shrink it to put it in a laptop?
Same. I refuse to use Windows. I worry that nvidia had to promise MSFT they wouldn’t release linux for the new desktop/laptop chips for a while.
I don’t see how they could dissipate 200w in a laptop, they are saying these chips will burn 80w, so something has to give.
One benefit of them using the same chip is that it should help ensure longer-term support for GB10 systems. It’s also a positive step for ARM, as wider adoption could help drive better software compatibility and encourage more developers to support ARM natively rather than focusing solely on x86
When GB10 systems were announced, NVIDIA’s focus seemed to be on native ARM software support rather than relying on x86 emulation. Wider adoption of the platform could help drive more native ARM applications over time.
RTX Spark is Nvidia’s plan for consumer market. They will move from graphic card to own the whole PC eco-system. It also serves as Nividia’s insurance plan against the future where Data center build out stalled.
In the event where AI bubble pop, one possible path forward for Nvidia is turn into Apple like entity. They are betting AI models will turn into operating system of the future.
Nividia is leverage tinkerers like us to search for no so obvious AI application. It is the same playbook as personal PC.
That is my hope as well. If it’s going to be a mainstream chip we will get more support.
I think the whole point of RTX Spark is to enter the CPU market and sell All in one solution, which maybe more appealing to many laptops. But it’s not the AI niche product at all, so I would believe that they will work more on Windows compatibility, other software and AI accelerator it’s a selling add-on.