Pardon me if this has been discussed and answered already, but I am still struggling to find info around it, and perhaps we simply just do not know yet.
DGX Spark vs RTX Spark. What is really the differences here? Why are there 2 different products for seemingly the exact same thing?
Am I missing something? I understand that RTX Spark is shipping with windows (ARM), and DGX Spark ships with linux. However, from what I can tell is the exact same thing, same chip, literally the same everything between the 2 besides the software it ships with.
If I have DGX Sparks today, do we expect Windows support to also be there? Are there actually silicon differences between the 2 that make the windows support possible? RTX Spark is being touted as a “new” SOC, but I cannot seem to find anything that details what makes it “new”. From what I can tell, its a year old, the hardware already exists, and I am confused why a Fall release when the platform, chip, and everything else has been on the market for some time and is widely available.
RTX drops the ConnectX-7 interface, so it will probably be cheaper, but it loses the fast connection between nodes. It will have marginally better memory bandwidth (300gb/s vs 273 for DGX). Feels like it’s for people who want a mini pc to play around with smaller models and maybe do some gaming.
The big question for me, since both DGX and RTX are GB10 chips, will they use the same LPDDR5X chips as well? If so, it would seem like 300GB/s is only a matter of a firmware update for DGX owners.
To be honest, they looks like very similar in terms of CPU/GPU architecture, and to me what is really interesting it’s to understand if we can get the same software stack to run on our DGX Spark. DGX Spark with linux is great to host and act like an home server, while RTX Spark with windows looks like more capable to directly “fuse” some windows workflow with AI agents, like interact with desktop apps and so on. To me it will be interesting if we’ll be able to test windows and RTX Spark software on DGX box, even loosing connect-x cabability during windows operation/test.
The more I think about it, the more I think the 300GB/s number is just Nvidia and partners rounding up. It is not unheard of for specs to be discussed or disclosed that way. It is very likely the exact same platform with lower memory options and no CX-7 ports.
Which should in theory bring the prices way down and make them more price competitive. The industry is selling this as a direct competitor to Apples lineup of hardware.
I thought long and hard about it, not in a search of new pc, but potentially I want a great working laptop for work and travel. I have a pretty good one, 2 yo i9 with 32g and rtx 4070ti 12g. Ideally I want a rocketship of course. But rtx spark just seems is nowhere. Underdeveloped windows arm, not natively compatible with amd64/x64. So to run or build native windows software an emulator needed. While they exist and pretty good it’s not the same as bare metal. I have some dependency on windows only x86 stack and have to build software for it. Most of my solutions are cross-platform dot net but a vital piece is anchored to windows dotnet x64 and nothing I can do about it. So no easy solutions here. So if we look at hybrid path then rtx spark is actually inferior by a mile to macbook m5 max 16 128g. 800Gb/s ram, really good GPU. Can run models under 80-90Gb faster than single spark. Same quality emulator for windows x64 as rtx arm. Windows. But much better package overall than any contraptions from Microsoft. Still not fully native solution. I imagine myself traveling with 16 inch 7k macbook, 16inch windows i9 laptop, 7k worth of 2 sparks, portable monitor in my backpack in airport security queue. Reassuring picture.
I was thinking the exact same thing. Windows on ARM is the biggest liability here. However, the fact that Microsoft has partnered with NVIDIA on this leads me to believe that they want to make Windows on ARM drivers more mature for NVIDIA products. I just can’t see the upside otherwise.
This article from The Verge mentions the new RTX Spark Asus models are using LPDDR5X 9400 on the P16 and LPDDR5X 9600 on the P14. This could be the reason for the ‘bump’ in memory performance advertised for the RTX Spark platform.
The question I now have is why the shift for RTX-spark? Perhaps the supply of chips used in DGX have dried up, so they are going for slightly faster stuff for RTX. Minus the CX-7.
Wendel from level1tech speculated the RTX Spark used the same chip as the DGX Spark but pointed out it has a NPU feature that would probably require additional silicon and finally concluded it was a different chip. who knows for sure?