[hank808]

You guys that continue to compare DGX Spark to the Mac Studios, please remember two things:

1. Virtually every model that you'd run was developed on Nvidia gear and will run on Spark. 2. Spark has fast-as-hell interconnects. The sort of interconnects that one would want to use in an actual AI DC, so you can use more than one Spark at the same time, and RDMA, and actually start to figure out how things work the way they do and why. You can do a lot with 200 Gb of interconnect.

[nialse]

Also remember that the Mx Ultras have 2-3x the memory bandwidth. Looking at the benchmarks even Strix Halo seems to beat the Spark. Buying a 200 Gbps switch is $10k-$100k+ so don't imagine anyone actually will use the interconnect. The logical thing for Nvidia would be to sell a kit with three machines and cabling, and make it a ring with the dual ports per machine. Helps for some scenarios but not others with the 10 times slower network than memory bandwidth.

[miladyincontrol]

On another note to remember, you can also ring topology mac studios using TB5 for 120Gbps per link with four such ports, all using cheaply available cable

[wmf]

You could also connect Sparks in a 200 Gbps ring with cheapish ($90) cables.

[moondev]

| Buying a 200 Gbps switch is $10k-$100k+

$1,295.00

https://www.balticnetworks.com/products/mikrotik-crs812-ddq-...

[m00x]

At best this is a cheap setup to test distributed training/inference code.

[pavlov]

It would be very interesting to read a tutorial on case 2.

[hank808]

@pavlov here's the tutorial that you wanted. https://youtu.be/rKOoOmIpK3I?si=WgLTee3Kc1SnUbDZ